How biomanufacturing actually works, and what it is worth

Every biologic medicine on the market is made the same improbable way. A cell is engineered to manufacture a protein it was never meant to make, grown by the trillion in a steel or plastic vessel, and then dismantled away from its own product until what remains is pure enough to put into a person. Nothing about that is like making a pill. A chemical reaction run twice gives the same answer twice; a cell run twice does not. A tablet survives heat, time and rough handling; a protein, a viral vector or a living T-cell does not. Almost every process step described below exists to control one of those two facts -- variability or fragility -- and the regulatory framework on top exists because neither can be inspected out of a finished vial.

This piece is the long version of a subject two shorter pieces on this site cover in summary.77"Biomanufacturing's blurry number." BigBio.ai, August 10, 2026. Covers the market-size disagreement, 2025-2026 trends and the regulatory backdrop in summary. (This site's own prior publication)88"The consolidation nobody priced in." BigBio.ai, August 10, 2026. Covers the filings and registry evidence in summary. (This site's own prior publication) It runs from the 1973 experiment that made any of it possible, through each production stage, through every product class the industry actually ships, out to the technologies at its edge -- and then asks what the filings, the trial registry and the patent office say about the size and shape of the whole thing. Where a figure could not be traced to a primary source this pass, it is marked as not established rather than filled in from general knowledge.

Contents

• The arc, 1973 to now
• What counts as biomanufacturing
• How it actually works
• What biomanufacturing actually makes
• The frontier, graded against a real scale
• A market whose size depends on who is counting
• What the filings actually show
• A convergence window, and a pattern worth watching rather than trusting
• The demand already committed
• Innovation intensity, by patent title
• What changed in 2025 and 2026
• AI on the plant floor, graded by tier
• Who makes it, and who stands beside them
• The regulatory and policy backdrop
• Where the current signals point
• Pressures, and what relieves them
• What this piece does not establish
• Sources, by tier

The arc, 1973 to now

The four-inflection-point version of this history -- biologics become a category in the 1980s, contract manufacturers scale in the 2000s, COVID forces a capacity step-change, federal policy names the field -- is directionally right and skips the science that made each step possible. Rebuilt with named people and dated approvals, the pattern that emerges is different and more useful: every commercial step-change was preceded by a scientific runway five to forty years long that hindsight erases.

1973 -- the founding experiment. In spring 1973 Stanley Cohen at Stanford and Herbert Boyer at UCSF spliced a piece of DNA carrying tetracycline resistance into the plasmid pSC101 and replicated it inside living Escherichia coli. Cohen, Chang, Boyer and Helling published the method in the Proceedings of the National Academy of Sciences that November.99Cohen SN, Chang ACY, Boyer HW, Helling RB. "Construction of Biologically Functional Bacterial Plasmids In Vitro." Proceedings of the National Academy of Sciences 1973;70(11):3240-3244. Accessed August 10, 2026.Primarythe original paper Read plainly, what they had shown was that DNA from one organism could be cut out, joined to another organism's genetic material, and then expressed by a host cell that would go on copying it indefinitely. That is the mechanism every biologic in this piece still runs on: a living cell talked into manufacturing a specific protein it would not otherwise make.

1976 -- the first company built to sell it. Boyer partnered with the 29-year-old venture capitalist Robert Swanson to found Genentech in 1976, on the premise that recombinant DNA could be turned into marketable medicines rather than only published science.1010Genentech. "Our Founders." gene.com. Accessed August 10, 2026: "founded ... in 1976, by ... Robert A. Swanson and ... Dr. Herbert W. Boyer."Primarythe company's own founder history In the autumn of 1977, before Genentech had lab space of its own, Boyer's UCSF group and Keiichi Itakura at City of Hope expressed a human hormone, somatostatin, in bacteria -- proof the technique worked on human genetic material and not only bacterial.

1982 -- the first recombinant drug reaches patients. Genentech licensed its recombinant human insulin technology to Eli Lilly, which brought Humulin to market. FDA's own Drugs@FDA database gives an original approval date of October 28, 1982 for Humulin R and Humulin N under BLA018780 and BLA018781.1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database This is the moment biologics stopped being a laboratory result and became a prescribable category of medicine.

1975-1986 -- monoclonal antibodies, from a Nobel-winning trick to a first approval. Georges Kohler and Cesar Milstein invented hybridoma technology in 1975: fuse an antibody-producing spleen cell to an immortal myeloma cell and the resulting hybrid churns out one single, uniform antibody forever. The work won them, with Niels Jerne, the 1984 Nobel Prize in Physiology or Medicine. The first monoclonal antibody approved for human therapeutic use was muromonab-CD3, sold as Orthoclone OKT3, in 1986, for acute rejection in kidney-transplant patients.1212"Muromonab-CD3 (Orthoclone OKT3): the first monoclonal antibody approved for therapeutic use." PubMed, PMID 3557906. Accessed August 10, 2026.Primaryindexed primary literature1313Monoclonal antibody manufacturing history review, 2024. PMC, PMC11218797. Accessed August 10, 2026: covers hybridoma-era ascites production and the shift to recombinant mammalian-cell manufacturing.Primarypeer-reviewed review OKT3's manufacturing method is worth knowing precisely because everything after it was built to avoid repeating it: recombinant mammalian-cell production was not yet good enough to make antibodies at scale, so manufacturers grew the mouse hybridoma cells as tumours in the peritoneal cavity of live mice and harvested antibody from the resulting ascites fluid. The entire downstream-purification discipline described below exists, in part, to have made that unnecessary.

The 1950s -- the cell line that would come to dominate manufacturing, decades before anyone knew. Theodore Puck, at the University of Colorado, isolated and characterised the Chinese hamster ovary line CHO-K1 while developing single-cell mammalian tissue-culture technique -- work aimed at basic genetics, with no manufacturing application in view.1414National Academy of Sciences. "Theodore T. Puck: Biographical Memoir." nasonline.org. Accessed August 10, 2026 via the Internet Archive, the live page returning 403 to scripted fetchers: confirms Puck's isolation and characterisation of the CHO-K1 line during his single-cell mammalian tissue-culture work. The specific 1955-56 dating carried by secondary accounts is not stated in the memoir and is therefore not asserted here.Primaryacademy biographical memoir CHO did not become a production platform until the 1980s, when three of its traits turned out to matter enormously: it grows well in suspension, it accepts stable high-yield foreign-gene insertions, and, being a non-human line, it carries minimal risk of harbouring human-infectious viruses. The large majority of FDA-approved recombinant biologics are made in CHO cells today. It is the most consequential accidental handoff in this history, from a genetics tool to the substrate of a multi-hundred-billion-dollar manufacturing base.

1997-1998 -- mammalian-cell manufacturing proves it scales. Rituxan (rituximab), a chimeric mouse-human antibody from IDEC Pharmaceuticals and Genentech, was the first licensed full-length recombinant monoclonal antibody therapy, produced in a then-unprecedented 12,000-litre CHO bioreactor at Genentech's South San Francisco plant. FDA's database gives rituximab an original BLA approval date of November 26, 1997 (BLA103705). Trastuzumab (Herceptin), the first antibody engineered against a genetic marker -- HER2 -- to select which breast-cancer patients would respond, followed on September 25, 1998 (BLA103792).1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database Between them they mark the shift from murine antibodies like OKT3 to humanised and chimeric antibodies made at commercial CHO scale, which is still the template essentially every subsequent antibody follows.

The 2000s -- contract manufacturing becomes a distinct, investable business. Lonza, a Swiss company founded in 1897 as a hydroelectric and industrial-chemicals business, pivoted into biotech manufacturing across the 1980s and 1990s, acquired Celltech Biologics and its mammalian sites in Slough and Portsmouth in 1996, and completed three 20,000-litre mammalian bioreactors in Portsmouth by 2004 at a cost exceeding $200 million -- at the time the largest capital investment in the company's history.1515Lonza. "Our History." lonza.com. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: 1897 founding; 1996 acquisition of Celltech Biologics; three 20,000-litre mammalian bioreactors completed in Portsmouth by 2004 at a cost exceeding $200 million.Primarycompany history Samsung Biologics, now one of the largest contract manufacturers by capacity, was founded in 2011.1616Samsung Biologics. "Milestones." samsungbiologics.com. Accessed August 10, 2026: "APR. Foundation of Samsung Biologics" under 2011.Primarycompany history That gap matters for reading the revenue figures below: Samsung is a newer entrant compounding faster than the incumbents, not an incumbent defending share. Boehringer Ingelheim's BioXcellence history could not be independently verified this pass, because the company's own history page returned a bot-detection block rather than content, and is marked not established here.1717Boehringer Ingelheim BioXcellence corporate history could not be independently verified this pass; the company's own history page returned a bot-detection block rather than content, and no substitute primary source was located.Not foundno source

2005-2020 -- the most underappreciated thread in the story. Katalin Kariko and Drew Weissman, working together at the University of Pennsylvania from 1997, spent years fighting a problem that made synthetic mRNA look like a dead end: injected into cells, it triggered a violent innate immune response. In 2005 they published the fix -- replacing standard uridine with a chemically modified nucleoside, eventually 1-methylpseudouridine, let synthetic mRNA slip past the innate immune system almost entirely; a 2008 follow-up showed the same modification also raised protein yield.1818Lasker Award retrospective on nucleoside-modified mRNA, 2021. PMC, PMC8462133. Accessed August 10, 2026: Kariko and Weissman's 2005 finding that replacing uridine with a modified nucleoside suppressed innate immune activation, and the 2008 follow-up showing increased protein yield.Primarypeer-reviewed retrospective1919The Nobel Assembly at Karolinska Institutet. "The Nobel Prize in Physiology or Medicine 2023." NobelPrize.org, October 2, 2023. Accessed August 10, 2026: awarded jointly to Katalin Kariko and Drew Weissman "for their discoveries concerning nucleoside base modifications that enabled the development of effective mRNA vaccines against COVID-19."Primarythe awarding body That work is the foundation both COVID-19 vaccine companies were built on. BioNTech was founded in Germany in 2008 by Ugur Sahin and Ozlem Tureci, originally for personalised cancer vaccines rather than infectious disease; Moderna was founded in the US in 2010 by Derrick Rossi and collaborators. Both spent the whole of the 2010s developing mRNA-lipid-nanoparticle candidates against Zika, cytomegalovirus and other targets, none of which had reached approval by January 2020. The real story is not that mRNA vaccines were invented in 2020, but that a fifteen-year, two-company platform effort was sitting nearly ready when a pandemic handed it its first market. Kariko and Weissman received the 2023 Nobel Prize in Physiology or Medicine for exactly this discovery.1919The Nobel Assembly at Karolinska Institutet. "The Nobel Prize in Physiology or Medicine 2023." NobelPrize.org, October 2, 2023. Accessed August 10, 2026: awarded jointly to Katalin Kariko and Drew Weissman "for their discoveries concerning nucleoside base modifications that enabled the development of effective mRNA vaccines against COVID-19."Primarythe awarding body

2010 -- Congress creates a biosimilar pathway, and it takes five years to produce one. The Biologics Price Competition and Innovation Act, signed in March 2010 as part of the Affordable Care Act, created the first US route -- analogous to the generic-drug Hatch-Waxman pathway, but for biologics -- by which a product shown highly similar to an approved reference biologic could reach market without repeating the original clinical programme. The first product approved under it was Zarxio (filgrastim-sndz), Sandoz's biosimilar to Amgen's Neupogen, on March 6, 2015 (BLA125553).1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database Five years from statute to first product is a measure of how much analytical and manufacturing-comparability work a biosimilar sponsor still has to do even when it is excused from repeating Phase 3.

2017 -- cell and gene therapy manufacturing becomes commercial. In a single five-month stretch the FDA approved three products that created two entirely new manufacturing categories. On August 30, 2017 it approved Kymriah (tisagenlecleucel), Novartis's CD19-directed CAR-T for paediatric and young-adult acute lymphoblastic leukaemia -- the first approved gene therapy in the US and the first approved CAR-T, manufactured individually for each patient from that patient's own T-cells.2020US Food and Drug Administration. Press announcement approving Kymriah (tisagenlecleucel), August 30, 2017. web.archive.org, archived September 3, 2017. Accessed August 10, 2026: the first gene therapy approved in the US, and the first CAR-T product.Primaryagency press announcement, via archive Gilead and Kite's Yescarta (axicabtagene ciloleucel) followed weeks later. On December 19, 2017 the agency approved Luxturna (voretigene neparvovec), the first gene therapy in which a corrective gene is delivered directly into patients by an adeno-associated virus vector, for a rare inherited blindness.2121Associated Press. Report on FDA approval of Luxturna (voretigene neparvovec), December 19, 2017. AP News. Accessed August 10, 2026: first gene therapy in the US delivering a corrective gene directly via an AAV vector, for a rare inherited retinal disease.Secondarycontemporaneous wire report of the agency action Those three approvals split the newest frontier of biomanufacturing along a line that still governs its economics: Kymriah and Yescarta are autologous -- one patient's cells in, one patient's product out, a manufacturing run of one, every time -- while Luxturna's AAV vector is allogeneic, one large batch of vector serving many patients.

2016-2017 -- US federal policy names biomanufacturing as its own category. On December 16, 2016 the US Commerce Secretary announced a $70 million award establishing the National Institute for Innovation in Manufacturing Biopharmaceuticals as the eleventh Manufacturing USA institute -- the first with a focus area proposed by industry rather than government, and the first funded directly by the Department of Commerce.2222US Department of Commerce and National Institute of Standards and Technology. "U.S. Secretary of Commerce Penny Pritzker Announces Biopharmaceutical Manufacturing Institute Joining Manufacturing USA Network." NIST, December 16, 2016. Accessed August 10, 2026: $70 million award establishing NIIMBL as the eleventh Manufacturing USA institute, the first with an industry-proposed focus area and the first funded directly by the Department of Commerce.Primaryfederal agency announcement The institute's own materials and NIST's institute page both list its establishment as 2017; the announcement and the formal establishment are a year apart, which is worth citing precisely rather than picking whichever is more convenient.

2020-2022 -- COVID forces a capacity step-change on top of all of it. Making mRNA vaccines at pandemic scale required lipid-nanoparticle and mRNA production capacity that barely existed in 2019, built directly on the platform work of 2005-2019 rather than invented in 2020. Both the Pfizer-BioNTech and Moderna vaccines were authorised within roughly a year of the viral sequence being published, a timeline made possible specifically because the underlying manufacturing science was already fifteen years mature.2323mRNA vaccine platform review. PMC, PMC8502079. Accessed August 10, 2026: in vitro transcription, lipid-nanoparticle formulation, and first US authorisation of the Pfizer-BioNTech vaccine on December 11, 2020 with Moderna's following roughly a week later.Primarypeer-reviewed review Samsung Biologics' own reported revenue traces the resulting build-out directly: KRW 1,165 billion in 2020, KRW 1,568 billion in 2021, KRW 2,437 billion in 2022 -- a 109% two-year increase.33Samsung Biologics. "Annual Business Report." Samsung Biologics Investor Relations. Accessed August 10, 2026: consolidated K-IFRS revenue (operating profit) in KRW billion -- 2020: 1,165 (293); 2021: 1,568 (537); 2022: 2,437 (969); 2023: 2,939 (1,206); 2024: 3,497 (1,321); 2025: 4,557 (2,069). The 2022 figures consolidate a newly acquired US subsidiary and were unaudited per the company's own footnote.Primarycompany-reported consolidated financials

What the long version changes. The compressed four-point history makes it look as though biologics, contract manufacturing, mRNA and federal policy each arrived on schedule. They did not. Every commercial step-change here -- insulin in 1982, antibody scale-up in 1997 and 1998, biosimilars in 2015, cell and gene therapy in 2017, mRNA at scale in 2020 -- sat on top of a long, often unfunded scientific runway that becomes invisible the moment the product ships. That pattern is the one thing in this history worth carrying forward, and it is examined again below.

What counts as biomanufacturing

Two definitions compete for the word, and which one a report uses quietly decides its headline number.

The narrow, pharmaceutical definition is the one the National Institute for Innovation in Manufacturing Biopharmaceuticals and the International Society for Pharmaceutical Engineering use: the production of biologic medicines -- biologics and vaccines -- through specialised, cGMP-regulated processes, spanning lab-scale development through commercial deployment.2424Schiel J, et al. "Biomanufacturing readiness levels [BRL] -- A shared vocabulary for biopharmaceutical technology development and commercialization." Biotechnology and Bioengineering 2022;119:3526-3536, doi 10.1002/bit.28227, PMC9826509, PMID 36071569. Accessed August 10, 2026: the nine-level scale across three phases, defined for technology feeding "the production of a licensed biologic medicine." Note on the citation itself: PMC9463082, carried by earlier coverage for this paper, resolves to an unrelated study of prescription-drug misuse and is not the BRL paper.Primarypeer-reviewed paper2525International Society for Pharmaceutical Engineering. "Guide: Biopharmaceutical Process Development and Manufacturing." ISPE. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers.Secondaryprofessional-society guidance, not independently peer-reviewed The Northeast Biomanufacturing Center's technician-training definition breaks the same activity into the stages an operator would recognise: upstream growth of biomass in CHO, E. coli, yeast or another host; downstream purification and recovery; then formulation, filling and packaging.2626Northeast Biomanufacturing Center and Collaborative. "Skill Standards." biomanufacturing.org. Accessed August 10, 2026.Secondaryeducational-consortium source

The broad, bioeconomy definition is EuropaBio's: any use of biological mechanisms -- human or animal cells, plants, micro-organisms, viruses -- to synthesise products at commercial scale, with applications running from medical therapies to animal feed, fuels, textiles, plastics and cosmetics.2727EuropaBio. "Biomanufacturing Definition 101." europabio.org, May 2023. Accessed August 10, 2026: the broad bioeconomy definition covering medical therapies, animal feed, fuels, textiles, plastics and cosmetics.Secondaryindustry-association source A monoclonal antibody plant and a fermentation vat making vitamin B12 both count under EuropaBio's definition; only the first counts under the narrow one. Every market-size figure below inherits this fork, whether or not the report producing it says so.

Modalities inside the narrow scope are monoclonal antibodies and other recombinant biologics, vaccines including cell-culture and mRNA platforms, cell and gene therapy, and biosimilars. Cell and gene therapy is a distinct manufacturing discipline within that set rather than a variation on it -- vector design, cell isolation and expansion, and platform technology for purifying viral vectors at scale have little in common with antibody production.2828"Manufacturing Cell and Gene Therapies: Challenges in Clinical Translation." PMC, PMC10961620. Accessed August 10, 2026: vector design, cell isolation and expansion, formulation, fill-finish and GMP facility requirements as a distinct manufacturing discipline.Primarypeer-reviewed review

The boundary with adjacent categories matters twice over. Biomanufacturing is a subset of pharmaceutical manufacturing at large: small-molecule, chemically synthesised drugs sit outside it. And it overlaps with, but is not identical to, the contract development and manufacturing industry, which is the outsourced delivery model for biomanufacturing rather than the activity itself.

How it actually works

The shape of the whole thing

Strip away the modality-specific detail and every biomanufactured product -- an antibody, a vaccine, a CAR-T therapy, a gene therapy -- moves through the same four phases in the same order, even though what happens inside each differs enormously.

Build the factory, run the factory, separate the product from the factory, put it in a vial. Each phase in the source of the constraint it exists to manage: variability upstream, purity downstream, sterility at the end.

The one-sentence version: biomanufacturing is the discipline of getting a living cell, or a process derived from one, to make a specific therapeutic molecule reliably and at scale, and then separating that molecule from everything else the cell also made, without letting anything unsafe reach a patient.

Two facts drive nearly every step that follows, and neither applies to conventional chemical manufacturing. Biological systems are variable: a cell today is not identical to a cell tomorrow, in a way a chemical reaction is not. And biological products are fragile: proteins, viral vectors and living cells can be destroyed by heat, shear, a pH swing, or simply time, in ways a small-molecule tablet cannot. Read every process control below as an answer to one of those two.

Upstream and downstream is the field's own shorthand and the most load-bearing vocabulary in it. Upstream is everything up to and including growing the cells that make the product: cell-line development and the bioreactor culture step. Downstream is everything after: purifying, polishing and formulating what the cells made. Service offerings, capacity metrics and even hiring categories are organised around that split.

Cell line development, or building the factory before you can run it

Before anything can be manufactured, someone has to build the biological factory: a single, genetically defined cell -- or, for viral-vector products, a producer cell paired with a viral genome -- engineered to express the therapeutic molecule and proven stable and safe enough to carry from a lab flask to a multi-thousand-litre vessel.

For a recombinant protein or antibody the dominant workflow begins with CHO cell transfection: the DNA encoding the therapeutic protein is pushed into a CHO host line by electroporation, a lipid-based carrier, or viral transduction. Transfection is inefficient and random. Most cells either fail to take up the gene at all, or take it up in a place or a copy number that produces little product. What comes out is therefore not a cell line but a large, genetically mixed pool, of which a small fraction happen to be good producers.

Clone selection is the search through that pool for the one cell worth keeping, and then growing an entire population from that single cell so every daughter is genetically identical to it. The winner has to satisfy four things at once: high and stable expression of the target; correct folding and post-translational modification (many therapeutic proteins, antibodies above all, need particular sugar chains attached in particular places to work, and to avoid provoking an immune reaction); reliable growth in suspension; and genetic stability across the dozens of population doublings a commercial run demands. Modern selection uses single-cell sorting -- fluorescence-activated cell sorting, or dedicated single-cell imaging -- with automated screening across hundreds to thousands of candidates, because the gap between a mediocre clone and an excellent one can be several fold in yield. That gap is not academic: it decides directly how large, and therefore how expensive, a bioreactor has to be to make a given quantity of drug.

The Master and Working Cell Bank system is what lets a manufacturer make the same biologic, to the same quality, for decades after that one clone was picked. Once the production clone is chosen, a large batch of it is expanded and frozen into thousands of cryogenic vials. That is the Master Cell Bank: characterised exhaustively for identity, genetic stability, absence of adventitious viruses and growth behaviour, then essentially never touched again except to make Working Cell Banks. A Working Cell Bank is a second-generation stock expanded from a handful of Master vials and frozen in bulk; Working vials are what actually get thawed to start a campaign. The two tiers exist so that nobody ever has to repeat clone development and full characterisation for a licensed product. Every commercial batch of Herceptin made since 1998 traces back to vials drawn from the same characterised banks -- and that traceability is precisely the evidence a regulator needs in order to believe batch four thousand will behave like batch one.

Cell and gene therapy products have an analogous but different front end, described below: an autologous cell-therapy manufacturer banks or directly processes each individual patient's cells, and a gene-therapy manufacturer banks the producer cell line and the plasmids used to generate vector, not a protein-producing clone.

Upstream processing, or growing the cells

Once a Working Cell Bank vial is thawed, upstream processing expands that small starting population through progressively larger vessels -- shake flasks, then seed bioreactors, then the full production bioreactor. Throughout, operators feed and control the culture environment (temperature, pH, dissolved oxygen, nutrient and waste concentrations) closely enough that the cells grow densely and express the target at high yield without triggering the stress responses that cut productivity or degrade quality.

Three bioreactor operating modes, in rising order of process complexity:

• Batch. Cells, nutrients and everything else go in at the start. The culture runs until nutrients deplete or waste products such as lactate and ammonia reach inhibitory levels, then the whole vessel is harvested at once. Simplest to run; lowest yield per litre and per day.
• Fed-batch, the industry-standard workhorse for proteins and antibodies. Concentrated nutrient feeds go in incrementally through the run rather than all at once, which lets the culture reach higher cell densities and run longer before depletion or waste forces a harvest. Markedly higher titres per batch, at the cost of tighter process control.
• Perfusion, or continuous culture. Fresh medium goes in continuously while spent medium carrying the product comes out continuously, the cells being retained in the vessel by a filter or centrifuge. The culture sits in a near-steady metabolic state indefinitely instead of running as a single time-limited batch. Perfusion reaches very high cell densities and output per litre and enables smaller-footprint continuous manufacturing -- the mode the FDA's pending cGMP amendments are written to accommodate. It is also the hardest to run and demands the finest real-time monitoring.

Single-use against stainless steel is the other big upstream design choice. A traditional stainless-steel bioreactor is a permanently installed vessel that must be cleaned and sterilised in place between every batch: capital-intensive, facility-defining, but efficient at very large volumes and durable over decades. A single-use system replaces the product-contact vessel with a pre-sterilised disposable bag, used once and discarded. That removes the cleaning and sterilisation window between batches -- faster changeover, lower cross-contamination risk when different products share a facility -- and lowers upfront capital, at the cost of higher per-batch consumables and a lower practical ceiling on scale. The rise of single-use since the 2010s is a direct enabler of the modern contract-manufacturing model: a plant built around disposable bioreactors can pivot between clients far faster than one built around dedicated stainless trains, and pivoting between clients is the entire business.

Scale-up -- moving a cell line and a process recipe from a two-litre bench vessel through pilot scale to a commercial vessel of thousands or tens of thousands of litres -- is one of the genuinely unsolved engineering problems in the field, and the reason is physical rather than biological. The environment a cell experiences does not scale with volume. Mixing patterns, oxygen transfer rates and the shear forces from impellers and gas sparging all change as the tank grows, so a process that runs perfectly at two litres can give lower yield or different product quality at two thousand. This is why scale-up work, called tech transfer when a process moves from an originator to a contract manufacturer, is a specialised discipline rather than a matter of building a bigger tank.

Downstream processing, or purification

At harvest the bioreactor holds a mess: the therapeutic protein, but also host-cell proteins and DNA from the CHO cells themselves, cell debris, spent medium, and variant forms of the product -- aggregates and fragments. Downstream processing is the purification train that separates the target from all of it, to a purity regulators will accept as safe and consistent enough to dose.

Protein A affinity chromatography is the near-universal first capture step for antibodies, and it works on a piece of borrowed biology. Protein A is a bacterial cell-wall protein, originally from Staphylococcus aureus, that happens to bind tightly and specifically to the Fc region shared by essentially all IgG antibodies. Pack it onto a resin, pour the crude harvest over it, and the antibody sticks while most host-cell protein, DNA and media components run straight through -- a jump to roughly 95% purity in one step. It is also why Protein A resin is among the single most expensive consumables in antibody manufacturing, and why Protein A supply is itself occasionally an industry capacity constraint.

Viral clearance and inactivation is a mandated set of steps rather than optional polishing: its purpose is to demonstrate that no infectious virus -- from the cell source, from a contaminated raw material, or from an adventitious event during manufacturing -- survives into the final product. Two mechanisms are typically used together precisely because they work on different principles and so are unlikely to fail together: low-pH inactivation, holding the post-Protein-A eluate at low pH for a defined time, which denatures many enveloped viruses without harming the antibody, and viral filtration, passing the product through a filter whose pores physically exclude virus particles by size while letting the much smaller antibody through. Regulators require dedicated spiked studies for every biologic: deliberately add a known virus to a scaled-down model of the process and show the process removes or kills it by a very large, specified margin, expressed as a log reduction value.

Polishing chromatography follows capture -- typically ion-exchange, hydrophobic-interaction, or both. Each exploits a different physical property of the molecule than Protein A did, which is the point: what capture leaves behind is precisely the material that resembles the product in the one dimension capture was reading. Polishing removes aggregates, fragments, residual host-cell protein and DNA, and any Protein A that leached off the column. Capture is a large fast jump; polishing closes the last gap to specification.

Ultrafiltration and diafiltration is the final concentration and buffer-exchange step. The purified product arrives in whatever buffer the last column left it in, and usually too dilute for a practical dose volume. Pressing it against a semi-permeable membrane both concentrates it -- water and small molecules pass, the large therapeutic molecule does not -- and swaps it into the formulation buffer it will actually be dosed in. That is the handoff into fill-finish.

Formulation, fill and finish

Formulation is the science of keeping a fragile molecule intact outside the plant. It means choosing the buffer, the pH, the stabilising excipients -- sugars, surfactants, amino acids -- and the container and closure that will hold a protein, a viral particle or a living cell potent through shipping, storage and administration, often across a shelf life of one to several years.

Fill-finish is the physical act of filling formulated product aseptically into its final container -- vials, pre-filled syringes, or a delivery device -- and freeze-drying it where required. Everything about it is governed by one constraint: a biologic cannot be terminally sterilised by heat the way many small-molecule injectables can, because the heat that kills the contaminant also destroys the drug. So the entire filling environment, and every step inside it, has to be sterile already. Lyophilisation removes water by freezing and then subliming it under vacuum, leaving a stable dry cake reconstituted with diluent at the point of use. It is reserved for products whose liquid form will not survive the required shelf life, and it costs a slower and more complex manufacturing step plus an extra handling step for whoever administers the dose.

Fill-finish is very often run as a distinct step, at a separate site or by an entirely different company, from production of the drug substance -- and the reason is specialisation. Aseptic filling needs its own equipment, cleanroom infrastructure and container-specific expertise; a line optimised for filling small-molecule vials at enormous scale is not automatically good at filling a low-volume, high-value biologic into pre-filled syringes. This is why the industry distinguishes drug substance, the purified active biologic, from drug product, the formulated, filled, patient-ready item. The two are frequently made in different buildings, by different companies, on different timelines -- and that separation is what made a fill-finish specialist a tier-one acquisition target, as the filing trail below shows.

Quality control, and what a batch has to prove

No batch of a biologic can be released without passing a defined panel of tests confirming it is what it is supposed to be, safe, and effective at the labelled dose:

• Identity -- confirming the molecule is the right one, by peptide mapping, mass spectrometry or specific binding assays.
• Purity -- quantifying purity and characterising process- and product-related impurities (aggregates, fragments, host-cell protein and DNA residuals) against specification, typically by chromatography and electrophoresis.
• Potency -- a functional assay, often cell-based, showing the product actually does its biological job at the expected strength. Structure is not function, and only this test asks about function.
• Sterility -- confirming no viable microbial contamination, traditionally by a fourteen-day compendial test. That fourteen-day hold is a direct throughput constraint on the whole plant, which is exactly why validated rapid microbiological methods are an active regulatory-science priority.
• Endotoxin -- testing for bacterial endotoxin, a fever-inducing contaminant that survives after every viable bacterium has been killed, typically by Limulus amebocyte lysate assay.

Batch release is the formal quality-assurance decision -- made by an authorised person, not by the scientists who ran the process -- that a specific batch meets every specification and may ship. It is the checkpoint everything above exists to support, and the one that real-time-release and automated-testing initiatives are trying to compress without weakening.

The regulatory frame: cGMP, and BLA against NDA

Current Good Manufacturing Practice is the FDA framework -- 21 CFR Parts 210 and 211 for drugs generally, Parts 600-680 for biologics -- governing the systems, facilities, equipment, training, documentation and quality practices a manufacturer must have to produce a safe, effective product consistently. It is not a certification anyone passes once. It is an operating standard the agency inspects against for the whole commercial life of a product, and every process step above is designed and documented to satisfy it.

BLA against NDA is the distinction that defines "biologic" for regulatory purposes. Most biologics are licensed through a Biologics License Application under the Public Health Service Act, reviewed by FDA's Center for Biologics Evaluation and Research or the relevant divisions of its drug centre, rather than through the New Drug Application pathway under the Food, Drug and Cosmetic Act that most small molecules use. The difference is not administrative. A BLA licenses both the product and the specific facility and process that makes it -- which follows directly from the variability and fragility described at the top of this section, and which is also the underlying reason a biosimilar sponsor cannot simply prove chemical equivalence the way a generic sponsor can.

Where each modality diverges from the spine

Monoclonal antibodies and recombinant proteins follow it most directly: CHO upstream, Protein-A-anchored downstream, standard fill-finish. This is the most mature, most standardised and per-unit generally cheapest modality here, which is also why contract capacity was historically built around it.

Vaccines split in two. Traditional vaccines are made by recombinant-protein expression, inactivation of a grown pathogen, or attenuation of a live one -- different upstream biology, broadly similar downstream logic. mRNA vaccines are almost entirely different: the drug substance has no living-cell step at all. mRNA is produced by a cell-free enzymatic in vitro transcription reaction, in which an RNA polymerase copies a DNA template in a bioreactor-like but cell-free system. The product is then encapsulated inside a lipid nanoparticle by microfluidic mixing -- combining the mRNA with an ionisable lipid plus cholesterol, a PEGylated lipid and a phospholipid at controlled pH, forming a particle that shields the fragile mRNA from degradation and gets it into cells after injection.2323mRNA vaccine platform review. PMC, PMC8502079. Accessed August 10, 2026: in vitro transcription, lipid-nanoparticle formulation, and first US authorisation of the Pfizer-BioNTech vaccine on December 11, 2020 with Moderna's following roughly a week later.Primarypeer-reviewed review That encapsulation step, not a bioreactor, is the real bottleneck in mRNA manufacturing, and it is why mRNA capacity required essentially new infrastructure rather than a repurposing of existing antibody plants.

Cell therapy is governed by the autologous and allogeneic distinction. Autologous manufacturing, which is what every current CAR-T product uses, begins with apheresis -- collecting one specific patient's white blood cells -- ships them to a facility, engineers them (usually by viral transduction to insert the chimeric antigen receptor gene), expands them, formulates the final cell product and ships it back to be infused into that same patient.

The loop closes on the person it started from. Because each run serves exactly one patient, none of the economies of scale a large fed-batch run gives an antibody are available; the industry measures the elapsed time from collection to reinfusion as vein-to-vein time, typically weeks, and for a patient with rapidly progressing cancer that is a clinical constraint, not an operational metric. Allogeneic, or off-the-shelf, cell therapies are made from a single donor or an engineered line at batch scale, many doses from one run -- closer to conventional biologic economics, though allogeneic cell therapy is less clinically mature than autologous CAR-T as of this writing.

Gene therapy centres on producing the viral vector itself as the product. AAV vectors, used in Luxturna and Zolgensma, are typically made by transfecting a producer line -- commonly HEK293 -- with plasmids encoding the therapeutic gene and the viral genes needed to package it. The resulting particles are then purified by specialised chromatography, analogous in principle to the antibody train but adapted for a much larger and more fragile particle. That train also carries a problem with no antibody analogue: separating full capsids, which contain the intended gene, from empty ones, which carry no payload, are otherwise nearly indistinguishable, and cannot be dosed as though they were active product. Published process work reports empty-to-full ratios around 50:50 before purification, which gives a sense of the size of the problem.2929AAV upstream-process methods paper. PMC, PMC8418526. Accessed August 10, 2026: reports an empty-to-full capsid ratio around 50:50 in the harvested material, before purification.Primarypeer-reviewed methods paper Lentiviral vectors, used to make the vector that in turn transduces cells in most current CAR-T processes, follow similar transient-transfection logic with a more complex genome and packaging design.

Biosimilars do not differ from their reference product in process architecture at all. A biosimilar sponsor builds its own CHO line, its own bioreactor process and its own purification train largely from scratch, because it cannot obtain the innovator's actual cell line. What differs is the analytical burden: because the 2010 statute does not require repeating the reference product's full clinical programme, the sponsor must instead generate an extensive head-to-head structural, physicochemical and functional comparison showing the product is highly similar notwithstanding minor differences in clinically inactive components. The characterisation substitutes for, rather than supplements, large parts of a clinical programme.

What biomanufacturing actually makes

This is a catalogue of product classes, not of brands: for each, what it is, how it is made, and two or three real, dated examples. Every approval date below was checked against FDA's Drugs@FDA database or an equivalent primary regulatory source; none is asserted from memory.

Monoclonal antibodies. An engineered antibody protein, made by the CHO-upstream, Protein-A-downstream route above, used either to block a disease-driving target directly or to flag diseased cells for destruction by the immune system. Verified examples: adalimumab (Humira), approved December 31, 2002 (BLA125057), an anti-TNF antibody for autoimmune disease and for years the best-selling drug in the world; trastuzumab (Herceptin), September 25, 1998, HER2-targeted for breast cancer; rituximab (Rituxan), November 26, 1997, anti-CD20 for B-cell lymphomas and autoimmune disease.1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database

Antibody-drug conjugates. An antibody chemically linked to a cytotoxic payload by a purpose-designed linker, so that the antibody delivers a poison specifically to cells carrying its target antigen. Manufacturing combines the antibody route above with small-molecule synthesis and a dedicated bioconjugation step that attaches the payload without destroying binding. Verified examples: ado-trastuzumab emtansine (Kadcyla), approved February 22, 2013 (BLA125427), conjugating trastuzumab's HER2 targeting to a microtubule-inhibitor payload;1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database enfortumab vedotin (Padcev), a Nectin-4-directed conjugate for bladder cancer whose label carries an initial US approval of 2019.3030Enfortumab vedotin (Padcev) prescribing information. DailyMed. Accessed August 10, 2026: the label carries "Initial U.S. Approval: 2019" and describes a Nectin-4-directed antibody-drug conjugate. A direct Drugs@FDA record was not returned under the generic name this pass; the label statement is used instead of a day-precise database date.PrimaryFDA-approved label

Bispecific antibodies. Engineered constructs that bind two different targets at once -- most often used to drag a patient's own T-cell into physical contact with a cancer cell and trigger it to kill, a mechanism a single-target antibody cannot achieve at all. Manufacturing follows the antibody spine with extra cell-line-engineering and purification complexity, because the process must ensure the correctly paired molecule assembles rather than the mispaired and single-target byproducts that form alongside it. Verified example: blinatumomab (Blincyto), a CD19-by-CD3 bispecific for B-cell acute lymphoblastic leukaemia, named in Kymriah's own FDA review documents as "the only available product in a related product class."3131US Food and Drug Administration. BLA review documents for tisagenlecleucel. fda.gov. Accessed August 10, 2026: reviewers describe blinatumomab as "the only available product in a related product class."PrimaryFDA review memorandum

Recombinant proteins and enzymes. Any therapeutic protein made by the recombinant route that is not an antibody. Insulin: Humulin, approved October 28, 1982, the founding product of the whole industry. Growth hormone: Genentech's Protropin (somatrem) was approved October 18, 1985 per the company's own primary account -- FDA's separate Orphan Drug Product database lists October 17, a one-day discrepancy flagged here rather than silently resolved.3232Genentech. "25th Anniversary of First Product Approval." gene.com. Accessed August 10, 2026: Protropin (somatrem) approved October 18, 1985, the first product Genentech itself manufactured and marketed. FDA's Orphan Drug Product database lists October 17; the one-day discrepancy is recorded rather than resolved.Primarycompany account, cross-read against the FDA database It was the second genetically engineered pharmaceutical approved in the US and the first Genentech itself manufactured and marketed; production ceased in 2004, superseded by Nutropin. Eli Lilly's Humatrope (somatropin) followed on October 16, 1986 (BLA019640) and remains active.1111US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16.PrimaryFDA's own approvals database Clotting factors: Recombinate, from Baxter Healthcare and Genetics Institute, approved December 10, 1992, was the first recombinant Factor VIII, ending the era of plasma-pooled Factor VIII and its bloodborne-infection risk; Bayer's Kogenate reached market as the second, its current label carrying an initial US approval of 1993.3333Recombinate approval, December 10, 1992, per contemporaneous pharmaceutical trade reporting. Citeline / Pink Sheet archive. Kogenate: current prescribing information. FDA, which states "Initial U.S. Approval: 1993." Accessed August 10, 2026. Pre-1996 BLA documents are not served by FDA's own approvals database and openFDA carries no record, so Kogenate is dated to the year rather than the day; a February 25, 1993 date carried by secondary accounts is not asserted. (Secondary for Recombinate's date; Primary for Kogenate's label year) Interferons: Intron A (interferon alfa-2b) and Roferon-A (interferon alfa-2a) were approved simultaneously in June 1986 for hairy cell leukaemia, FDA treating them as first approvals of two distinct recombinant interferon brands on the same day -- making interferon the third major recombinant-protein class to reach market within four years of Humulin. The day-precise date reported contemporaneously as June 4 sits behind a paywall in the only primary-adjacent record located this pass and is not asserted here at that precision.3434Contemporaneous pharmaceutical trade reporting of the simultaneous US approvals of Intron A and Roferon-A for hairy cell leukaemia, June 1986. Citeline / Pink Sheet archive, dated June 9, 1986. Accessed August 10, 2026: headline and metadata confirm the simultaneity and the indication; the body text, which carries the day-precise date, is paywalled, so the date is given here to the month.Secondarycontemporaneous trade reporting Enzyme replacement: the Gaucher-disease lineage is the clearest documented case of a first-generation extracted-protein therapy being replaced by a biomanufactured one -- alglucerase (Ceredase), approved in 1991 from placenta-derived enzyme, then its recombinant successor imiglucerase (Cerezyme), which the enzyme-therapy history dates to 1995, made in genetically engineered CHO cells.3535Gaucher-disease enzyme-replacement therapy history. PMC, PMC3340106. Accessed August 10, 2026: alglucerase (Ceredase) approved 1991 from placenta-derived enzyme; the recombinant successor imiglucerase (Cerezyme) dated to 1995 in this review. A 1994 date appears in label metadata elsewhere; the review's own date is used here and the discrepancy is recorded rather than resolved.Primarypeer-reviewed history

Vaccines. Two manufacturing families. Verified mRNA examples: the Pfizer-BioNTech COVID-19 vaccine, first authorised in the US on December 11, 2020, and the Moderna vaccine roughly a week later -- both built on the nucleoside-modified-mRNA platform above, both formulated by lipid nanoparticle.2323mRNA vaccine platform review. PMC, PMC8502079. Accessed August 10, 2026: in vitro transcription, lipid-nanoparticle formulation, and first US authorisation of the Pfizer-BioNTech vaccine on December 11, 2020 with Moderna's following roughly a week later.Primarypeer-reviewed review Verified traditional example: Recombivax HB, Merck's recombinant-DNA hepatitis B vaccine, produced in genetically engineered yeast expressing hepatitis B surface antigen, and one of the first recombinant-protein vaccines to reach market -- predating the mRNA platform by roughly three decades. Its day-precise original approval date was not independently re-derived beyond label metadata this pass and is marked not established at that precision.3636Current marketed status and label content for Recombivax HB, Amtagvi (lifileucel) and BeneFIX. DailyMed. Accessed August 10, 2026. Day-precise original approval dates were not independently re-derived beyond label metadata this pass and are marked not established at that precision.PrimaryFDA label database The manufacturing routes are genuinely different: Recombivax is a subunit protein grown and purified through the cell-culture spine, while an mRNA vaccine never passes through a living cell at all.

Cell therapies. Living cells, collected or engineered, as the product itself. Verified autologous CAR-T examples: Kymriah (tisagenlecleucel), approved August 30, 2017, the first approved gene or cell therapy in the US, and Yescarta (axicabtagene ciloleucel) shortly after, both CD19-directed for B-cell malignancies.2020US Food and Drug Administration. Press announcement approving Kymriah (tisagenlecleucel), August 30, 2017. web.archive.org, archived September 3, 2017. Accessed August 10, 2026: the first gene therapy approved in the US, and the first CAR-T product.Primaryagency press announcement, via archive A verified non-CAR-T example: Amtagvi (lifileucel), from Iovance Biotherapeutics, which its own label calls a "tumor-derived autologous T cell immunotherapy" -- made by harvesting a patient's tumour-reactive T-cells from a resected tumour, expanding them outside the body and reinfusing them. The shorthand "tumour-infiltrating lymphocyte" or TIL is the company's and the press's, not the label's, and is used here with that attribution.3636Current marketed status and label content for Recombivax HB, Amtagvi (lifileucel) and BeneFIX. DailyMed. Accessed August 10, 2026. Day-precise original approval dates were not independently re-derived beyond label metadata this pass and are marked not established at that precision.PrimaryFDA label database It is manufacturing-wise distinct from CAR-T: it uses cells that were already tumour-reactive, without the viral-vector transduction step that defines CAR-T. Broader stem-cell therapies exist and are in active development, but no further named and dated example was independently verified this pass and none is asserted here.

Gene therapies. A corrective gene delivered directly into a patient's cells by a viral vector. Verified examples: Luxturna (voretigene neparvovec), Spark Therapeutics, approved December 19, 2017, an AAV gene therapy for inherited retinal disease and the first gene therapy for a genetic disease fully approved in both the US and EU;2121Associated Press. Report on FDA approval of Luxturna (voretigene neparvovec), December 19, 2017. AP News. Accessed August 10, 2026: first gene therapy in the US delivering a corrective gene directly via an AAV vector, for a rare inherited retinal disease.Secondarycontemporaneous wire report of the agency action and Zolgensma (onasemnogene abeparvovec), Novartis and AveXis, an AAV9 therapy for spinal muscular atrophy, dated through the FDA-sourced regulatory literature rather than a direct database hit -- flagged as such rather than presented as database-confirmed.3737AAV clinical-trial-design review citing FDA's May 24, 2019 summary basis for regulatory action for Zolgensma. Cold Spring Harbor Perspectives in Medicine. Accessed August 10, 2026. A direct Drugs@FDA database record was not returned for either Zolgensma or Kymriah under their brand names this pass, so both dates rest on FDA-sourced literature rather than a direct database hit.Secondarypeer-reviewed review of the primary regulatory record

Blood and plasma-derived products are adjacent to this scope rather than inside it. Immunoglobulins and clotting factors extracted from pooled donated human plasma are made by fractionation, not by an engineered cell line, and so diverge from everything described above. They are flagged as adjacent rather than catalogued.

GLP-1 and peptide therapeutics deserve a note precisely because they are so often discussed in the same breath as biologics capacity while being manufactured by an entirely different route. Semaglutide -- Ozempic, Wegovy, and the oral formulation Rybelsus -- is a synthetic peptide, not a cell-expressed protein. Novo Nordisk's own account describes chemists, not cell-line engineers, synthesising and iterating candidates; semaglutide was literally compound number 217 in that campaign.3838Novo Nordisk. "Semaglutide." Annual Report 2025. Accessed August 10, 2026: semaglutide described as compound number 217 in the synthesis campaign.Primarycompany annual report The manufacturing-relevant innovation was chemical attachment of a fatty-acid side chain, which lets the peptide bind blood albumin and resist rapid clearance, enabling once-weekly dosing.3939GLP-1 drug-discovery review. PMC, PMC11441540. Accessed August 10, 2026: the fatty-acid side chain enabling albumin binding and resistance to rapid clearance, and therefore once-weekly dosing.Primarypeer-reviewed review Practically, GLP-1 production scales like specialised peptide chemistry, not like a bioreactor build-out -- worth stating plainly, because capacity pressure from GLP-1s and capacity pressure from biologics are frequently reported as one thing and are not.

The frontier, graded against a real scale

Everything above is approved and on the market. This section is different: it is the edge of the field as of 2026, checked against trial registries, drug labels and peer-reviewed literature rather than characterised from news coverage -- and graded honestly, because the edge is exactly where hype and reality are hardest to separate.

A shared yardstick. The institute at the centre of US biomanufacturing research published its own technology-maturity framework in 2022, modelled on the Department of Defense's Manufacturing Readiness Levels: the Biomanufacturing Readiness Level, nine levels across three phases.2424Schiel J, et al. "Biomanufacturing readiness levels [BRL] -- A shared vocabulary for biopharmaceutical technology development and commercialization." Biotechnology and Bioengineering 2022;119:3526-3536, doi 10.1002/bit.28227, PMC9826509, PMID 36071569. Accessed August 10, 2026: the nine-level scale across three phases, defined for technology feeding "the production of a licensed biologic medicine." Note on the citation itself: PMC9463082, carried by earlier coverage for this paper, resolves to an unrelated study of prescription-drug misuse and is not the BRL paper.Primarypeer-reviewed paper Phase 1, laboratory operations, runs from concept ideation (BRL 1) through proof-of-concept (BRL 3). Phase 2, pilot operations, runs from a built prototype (BRL 4) through a GMP-ready system producing pivotal batches for an approved investigational new drug application (BRL 7). Phase 3, commercial operations, is BRL 8 -- regulator-approved and used in at least one licensed product -- and BRL 9, broad routine deployment across multiple products. The scale is defined specifically for technology feeding the production of a licensed biologic medicine, which matters below: one of the examples here is a regulated food, and does not map onto the scale at all.

Ranges are ranges because the best available evidence supports a span, not a point. The two rows below the axis are not low on the scale; they are outside its defined scope, which is a different statement and the more interesting one.

Xenotransplantation. United Therapeutics, through its Revivicor subsidiary, and eGenesis are both running registered, FDA-regulated human trials of gene-edited pig organs -- not compassionate-use one-offs but structured Phase 1/2 studies with defined enrolment. United Therapeutics has two recruiting kidney studies of fifty participants each; eGenesis has a not-yet-recruiting study of a porcine liver paired with an external perfusion device, enrolling twenty.4040ClinicalTrials.gov records NCT06878560 and NCT07224763 (United Therapeutics, gene-edited pig kidney, both recruiting, enrolment 50 each) and NCT07429838 (eGenesis, EGEN-5784 porcine liver with external perfusion, not yet recruiting, enrolment 20). Queried August 10, 2026.Primarytrial registry The organs are edited to remove the pig antigens that trigger hyperacute rejection, in some cases adding human transgenes, using the same CRISPR toolkit described above for cell and gene therapy but applied to a whole living donor animal rather than a cell line. No xenotransplant product has FDA approval. BRL 6-7: human trials underway, no licensed product.

In vivo against ex vivo gene editing, precisely, because this is where a careless claim breaks. Casgevy (exagamglogene autotemcel), the first CRISPR-based therapy approved by FDA, in December 2023,4141Casgevy (exagamglogene autotemcel) prescribing information, setid 7c3e12ad-e2fe-4d3f-a630-ea7364d9e846. DailyMed. Accessed August 10, 2026: "Patients are required to undergo hematopoietic stem cell (HSC) mobilization followed by apheresis to obtain CD34+ cells for CASGEVY manufacturing," and after infusion "the edited CD34+ cells engraft in the bone marrow" -- an ex vivo, not in vivo, edited autologous cell therapy.PrimaryFDA-approved label is frequently described as in vivo gene editing. It is not. Its own FDA label states plainly that patients undergo haematopoietic stem cell mobilisation followed by apheresis to obtain the CD34+ cells from which Casgevy is manufactured, and that after infusion the edited CD34+ cells engraft in the bone marrow.4141Casgevy (exagamglogene autotemcel) prescribing information, setid 7c3e12ad-e2fe-4d3f-a630-ea7364d9e846. DailyMed. Accessed August 10, 2026: "Patients are required to undergo hematopoietic stem cell (HSC) mobilization followed by apheresis to obtain CD34+ cells for CASGEVY manufacturing," and after infusion "the edited CD34+ cells engraft in the bone marrow" -- an ex vivo, not in vivo, edited autologous cell therapy.PrimaryFDA-approved label The patient's own blood stem cells are removed, edited outside the body during manufacturing, and returned after myeloablative conditioning: an ex vivo gene-edited autologous cell therapy, manufacturing-wise a direct descendant of the CAR-T process. True in vivo editing -- a construct delivered into the body, editing cells in place -- exists in earlier-stage development, but no in-vivo-edited product was verified as FDA-approved this pass and none is claimed. Casgevy sits at BRL 8: one approved, commercially available product.

Tooth regeneration, a genuinely strange and genuinely Phase-I-complete example. Toregem BioPharma, in Kyoto, is developing TRG035, a humanised monoclonal antibody that neutralises USAG-1 -- a protein that normally suppresses a dormant third generation of tooth buds humans carry beyond their two ordinary sets.4242USAG-1 and tooth regeneration. Journal of Oral Biosciences 2024, PMID 39389160. Accessed August 10, 2026: confirms the USAG-1 mechanism in mouse models.Primarypeer-reviewed paper A Phase I double-blind, placebo-controlled dose-escalation safety trial in thirty healthy adult men is listed as complete in Japan's national registry, and Japan's health ministry designated TRG035 an orphan drug for severe congenital hypodontia in September 2025.4343Japan Registry of Clinical Trials, record jRCT2051240154. jrct.mhlw.go.jp. Accessed August 10, 2026: TRG035 Phase I, double-blind, placebo-controlled dose-escalation safety trial in 30 healthy adult men, listed complete.Primarynational trial registry Manufacturing-wise it is entirely unremarkable: Toregem contracted WuXi Biologics for CHO-and-Protein-A GMP production under a 2022 agreement, which is the point worth stating plainly. "Frontier" here means a novel target and mechanism riding on completely conventional manufacturing infrastructure. Phase I proved safety, not efficacy; whether the antibody actually regrows a properly positioned, functional tooth is unproven pending Phase II, and the current indication is congenital tooth agenesis, not adult tooth loss. BRL 6-7.

Bioprinting and in vivo triggered tissue growth: two attacks on the same problem, both research-stage. A direct registry query for bioprinted tissue transplant trials returned zero registered studies4444ClinicalTrials.gov searches for bioprinted tissue transplant studies and for patient-derived organoid studies. ClinicalTrials.gov. Queried August 10, 2026: zero registered trials for bioprinted tissue transplant; 181 for patient-derived organoids, every one sampled using organoids as a diagnostic or drug-screening tool rather than as the therapeutic product.Primarytrial registry -- a real, checkable negative result rather than an assumption. Whatever progress exists in smaller tissue constructs, organ-replacement bioprinting has not reached human testing. In April 2026 Science Advances published a different strategy aimed at the same underlying obstacle, which is that an engineered tissue implant cannot be built at full therapeutic size outside the body. Sangeeta Bhatia at MIT and Christopher Chen at Boston University and the Wyss Institute demonstrated BOOST: they implanted a small engineered human liver-tissue construct into mice and used a doxycycline-inducible synthetic-biology switch to trigger it to grow roughly six-fold in place, after implantation, instead of trying to build a full-size liver first.4545Stoddard A, et al. "Bioengineered on-demand outgrowth via synthetic biology triggering." Science Advances 2026;12(16):eadz8362, PMID 41996502; also at PMC13089339. Accessed August 10, 2026: an implanted engineered human liver-tissue construct triggered to grow roughly six-fold in place in mice; in vitro and mouse-model only, no IND filed.Primarypeer-reviewed paper This is preclinical -- in vitro and mouse only, no investigational-new-drug filing, no human trial -- but it is a genuinely different scaling strategy from printing, and a useful reminder that "grow a replacement organ" is being attacked from more than one direction. Both sit at BRL 1-3.

Organoids, a real clinical tool today but not the therapeutic product. A registry search for patient-derived organoid trials returned 181 studies;4444ClinicalTrials.gov searches for bioprinted tissue transplant studies and for patient-derived organoid studies. ClinicalTrials.gov. Queried August 10, 2026: zero registered trials for bioprinted tissue transplant; 181 for patient-derived organoids, every one sampled using organoids as a diagnostic or drug-screening tool rather than as the therapeutic product.Primarytrial registry every one sampled this pass -- ovarian, breast, pancreatic, lung, colorectal -- uses organoids grown from a patient's own tumour biopsy to predict which chemotherapy will work for that patient, not as a manufactured product infused as therapy. That is a scoping distinction rather than a maturity gap. Organoids are genuinely deployed in clinical practice today, just not as a biomanufactured therapeutic.

Cultivated meat, approved and on sale, and outside the scale entirely. Upside Foods and Good Meat received final USDA grants of inspection on June 21, 2023, following separate FDA safety determinations -- the first cultivated-meat sales approvals in the US, after Singapore's 2020 approval was the first worldwide. The product is grown from animal cells in steel bioreactors, mechanically closer to the upstream cell culture described above than to any conventional agriculture, but it is regulated as food under joint FDA and USDA authority rather than licensed as a biologic. It therefore sits outside the readiness scale's defined scope rather than at a low or high level on it.4646"Two companies can now sell lab-grown chicken in the US." MIT Technology Review, June 21, 2023, and Douglas L. "'A new era': US regulator allows first sales of lab-grown meat." Reuters, June 21, 2023. Accessed August 10, 2026: USDA grants of inspection to Upside Foods and Eat Just's Good Meat, following FDA no-questions letters in November 2022 and March 2023; Singapore approved cultivated chicken first, in 2020; product grown in steel tanks and served initially in restaurants.Secondarycontemporaneous press, cross-confirmed across two outlets It is currently sold in a small number of restaurants; broader retail availability was not established as achieved this pass.

Point-of-care and battlefield manufacturing. DARPA has funded distributed, on-demand pharmaceutical manufacturing since at least 2010, for a reason that maps directly onto the FDA's own pending rulemaking below: the current defence supply chain can take weeks to months to get pharmaceuticals and protein therapeutics to a battlefield. Battlefield Medicine funds two thrusts -- Pharmacy on Demand for small-molecule active ingredients and Bio-MOD for protein therapeutics -- building miniaturised continuous-flow platforms.4747Defense Advanced Research Projects Agency. "Battlefield Medicine." darpa.mil. Accessed August 10, 2026: two thrusts, Pharmacy on Demand for small-molecule active ingredients and Bio-MOD for protein therapeutics, building miniaturised continuous-flow platforms.Primaryagency programme page Reimagining Protein Manufacturing targeted initiating protein production within roughly twenty-four hours of receiving a DNA or RNA template in a fully distributed setting; DARPA's own page marks it complete. The newest programme, EQUIP-A-Pharma, is explicitly building the regulatory-qualification data needed to let multiple finished drug products be made on a single reprogrammable point-of-need platform; its programme manager cites $550 million in annual US hospital costs from rolling drug shortages as the civilian motivation, and the programme is coordinating with HHS to generate exactly the kind of real-world evidence the FDA's continuous-manufacturing framework will need.4848Defense Advanced Research Projects Agency. "Point-of-need pharmaceuticals." darpa.mil, January 24, 2024. Accessed August 10, 2026: EQUIP-A-Pharma building the regulatory-qualification data for multiple finished drug products on a single reprogrammable platform; "$550 million annually" in US hospital costs from drug shortages cited as the civilian motivation.Primaryagency announcement Working prototypes exist -- BRL 3-5, varying by sub-programme -- but no fielded, regulator-qualified system was independently verified as deployed this pass.

A market whose size depends on who is counting

The sources disagree, and the disagreement is the finding. Across eight analyst reports pulled this pass, current-year market-size estimates range from roughly $19.1 billion to $40.1 billion, and forecast growth from 6.6-28% compound annual, depending almost entirely on how narrowly each report scopes the word.

Rows are ordered by base-year size, so the two sub-segment reports sit at the top and the whole-market reports below them. The filled mark is the base year, the open mark the forecast. The axis is logarithmic because the smallest estimate here is a hundredth of the largest, and a linear axis would render two of the seven as marks indistinguishable from the origin. The spread, not any single point, is the finding.

Ordered by base-year size, which is the only ordering that makes the spread legible. The eighth source in this set is the capacity survey cited further below, which sizes production rather than revenue and so does not belong in this column.5454BioPlan Associates. "2026 23rd Annual Report and Survey of Biopharmaceutical Manufacturing Capacity and Production." GII Research, July 14, 2026. Accessed August 10, 2026. Sizes production capacity rather than market revenue, which is why it does not appear as a row in the revenue comparison.Secondarycommercial survey report

What drives the spread, as far as these sources allow: different base-year scoping (the whole biomanufacturing market against "biologics manufacturing" against narrower platform sub-segments treated as markets of their own); different vendor rigour, since Grand View Research is an established firm with a documented methodology while several others are smaller syndication sites whose methods are not independently auditable; and the narrow-against-broad definitional fork above. What is consistent across every source regardless of base number is the direction: high-single-digit to double-digit growth through 2030-2033, driven by biologics and biosimilar demand and cell and gene therapy scale-up.

A stress test the low end has to survive. Samsung Biologics' fiscal-2025 revenue of KRW 4,557 billion converts, at the prevailing rate,5555USD/KRW exchange rate. Trading Economics. Accessed August 10, 2026: 1,418.11 on August 10, 2026. The conversion above uses approximately 1,420 as a round current-rate approximation -- a spot translation, not the company's own reporting currency.Secondaryfinancial-data aggregator, for illustrative conversion only to roughly $3.2 billion. The lowest whole-market estimate above is $19.08 billion. One company -- a company that does nothing but contract biomanufacturing, unlike the diversified conglomerates in its tier -- would then already account for about one-sixth of the entire global market, before counting Lonza, WuXi Biologics, the now-Novo-owned Catalent, or Thermo Fisher's own biologics business. That does not rule out a $19 billion market. It does require the other four named tier-one contract manufacturers to hold collectively a noticeably smaller share than the single largest player, which is not the distribution the public rankings describe. This is a back-of-envelope share check, not a market-sizing exercise of its own.

One figure carried by earlier coverage is dropped here rather than repeated: a $47.7 billion current-year estimate could not be re-verified on the source's own page this pass. The confirmed figures from that source are $40.1 billion for 2025 and $140.6 billion for 2033.22Grand View Research. "Biologics Manufacturing Market Size, Share and Trends Analysis Report, 2033," report ID GVR-4-68040-743-9. Grand View Research. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: "valued at USD 40.1 billion in 2025 and is anticipated to reach USD 140.6 billion by 2033, growing at a CAGR of 16.7% between 2026 and 2033." A $47.7 billion 2026 figure carried by earlier coverage was not present on the page this pass and is not repeated.Secondarycommercial market research, proprietary methodology not independently auditable

What the filings actually show

A market-size estimate is an opinion about the future. A 10-K is a sworn statement about the past. This section stays with the second kind of number.

Three series, one axis of time. The sparklines carry shape, not magnitude -- each is scaled to its own range, which is what a word-sized graphic is for. The Thermo Fisher sparkline runs from 2016, four years earlier than its columns, because the full filed series is what makes its shape legible; the columns stay aligned with Samsung's for comparison. Samsung's 2022 figures consolidate a newly acquired US subsidiary and were unaudited per the company's own footnote.

Samsung Biologics. Revenue almost quadrupled in five years, and the operating margin moved with it rather than against it -- 25.2% in 2020 to 45.4% in 2025.33Samsung Biologics. "Annual Business Report." Samsung Biologics Investor Relations. Accessed August 10, 2026: consolidated K-IFRS revenue (operating profit) in KRW billion -- 2020: 1,165 (293); 2021: 1,568 (537); 2022: 2,437 (969); 2023: 2,939 (1,206); 2024: 3,497 (1,321); 2025: 4,557 (2,069). The 2022 figures consolidate a newly acquired US subsidiary and were unaudited per the company's own footnote.Primarycompany-reported consolidated financials That combination is the tell. A company scaling revenue on flat or thinning margins is selling more of the same thing at the same price. A company scaling revenue while margin nearly doubles is running fuller plants against a fixed cost base: utilisation rising faster than capacity itself. Fiscal 2025 revenue grew 30.3% year over year,5757Samsung Biologics. "Samsung Biologics Reports Fourth Quarter and Fiscal Year 2025 Financial Results." Samsung Biologics, January 21, 2026. Accessed August 10, 2026: FY2025 revenue KRW 4,557.0 billion, up 30.3% year over year; operating profit KRW 2,069.2 billion, up 56.6%.Primarycompany-reported consolidated results a rate no whole-market growth estimate above comes close to. Read plainly, the single biggest contract manufacturer by capacity is compounding faster than the analysts who size the category it competes in -- and doing it as a company founded in 2011, markedly younger than the incumbents it has overtaken.1616Samsung Biologics. "Milestones." samsungbiologics.com. Accessed August 10, 2026: "APR. Foundation of Samsung Biologics" under 2011.Primarycompany history

Thermo Fisher. Total company revenue, from XBRL company-facts pulled directly from SEC EDGAR, peaked at $44.9 billion in 2022, fell to $42.9 billion in both 2023 and 2024, and only in 2025 climbed back to $44.6 billion, still short of the 2022 high.5656Thermo Fisher Scientific Inc. XBRL company-facts, tag RevenueFromContractWithCustomerExcludingAssessedTax. SEC EDGAR, CIK 0000097745. Retrieved August 10, 2026: FY2020 $32.2 billion, FY2021 $39.2 billion, FY2022 $44.915 billion, FY2023 $42.857 billion, FY2024 $42.879 billion, FY2025 $44.556 billion; the FY2025 figure confirmed against the FY2025 10-K, filed February 26, 2026. Company-wide revenue; XBRL company-facts does not carry segment-level tags.Primaryfiled XBRL data and the underlying annual report Pulled back further, the full filed series runs from $18.3 billion in 2016, and the peak-and-plateau of 2022-2024 is a three-year interruption in a decade-long climb rather than the whole shape. That framing matters for how the growth rate is quoted: the bookend compound rate from 2016 to 2025 is 10.4%, the mean year-over-year change across the ten filed observations is 10.8% with a bootstrap 95% confidence interval of 4.5-17.2%, and the same compound rate recomputed across every five-year-or-longer window in the series ranges from 6.7% to 16.5%.5858Statistics computed independently for this piece from the raw series above, in Python, rather than taken from any source's own summary figure. Method: growth stated as the mean year-over-year change with a 20,000-resample percentile bootstrap 95% confidence interval, alongside the bookend compound rate and the full range of that compound rate recomputed across every window of five years or longer. The bootstrap is used in preference to a closed-form interval because ten to sixteen annual observations of noisy counts do not satisfy the assumptions a t interval would need. Where an interval crosses zero, that is stated rather than resolved. The intent is Motulsky's: an interval and a sample size, never a bare point estimate carrying more precision than the data holds. Raw pulls (SEC EDGAR XBRL company-facts, USPTO file-wrapper records, ClinicalTrials.gov API v2) and the analysis script are held with this piece's working files.Primarycomputed from the filed and registry data cited above A single growth number for this company is therefore a choice of endpoints as much as a measurement, which is worth saying out loud in a document that spends a section on other people's single growth numbers. The 2022 peak lines up with COVID-era diagnostics and vaccine-support revenue rolling off through 2023; 2025 is the first year that decline reversed. This is company-wide revenue, not the bioproduction segment specifically: XBRL company-facts data does not carry segment tags cleanly, because segment breakdowns live in the 10-K's narrative tables rather than as machine-readable dimensional facts. That limitation is flagged rather than approximated around. Whether the recovery reflects genuine biomanufacturing demand or simple post-pandemic normalisation is not established in the filing data alone -- but the direction, after two flat years, is up.

Catalent, and what a filing trail records. Catalent's last 10-K as an SEC-reporting public company was filed September 6, 2024, for the fiscal year ended June 30, 2024.5959Catalent, Inc. Form 10-K for the fiscal year ended June 30, 2024. SEC EDGAR, filed September 6, 2024, accession 0001596783-24-000087.Primaryfiled annual report On December 18, 2024, Novo Holdings completed its acquisition; Catalent's own 8-K says so directly.6060Catalent, Inc. Form 8-K, introductory note: "On December 18, 2024, Novo Holdings A/S ... completed the acquisition of Catalent, Inc." SEC EDGAR, filed December 18, 2024, accession 0001193125-24-280922.Primaryfiled merger-completion document The same day it filed a Form 25-NSE de-listing its shares from the New York Stock Exchange, and by December 30 had filed a Form 15-12G terminating its SEC reporting obligations.6161Catalent, Inc. Form 15-12G terminating registration. SEC EDGAR, filed December 30, 2024.Primaryfiled deregistration document By its own paperwork, one of five named tier-one contract manufacturers went from routine annual filing to fully deregistered in sixteen weeks. Whether Novo Nordisk redirected any of Catalent's fill-finish capacity toward its own GLP-1 production is not established here; the timing is checkable, and what it coincides with is Thermo Fisher's revenue decline stopping and Samsung Biologics' margin reaching a five-year high. Whether fewer independent competitors and rising margins at the survivors are connected or merely concurrent, this piece does not resolve. It is the kind of juxtaposition a market-size estimate built from aggregated analyst inputs cannot surface at all.

A convergence window, and a pattern worth watching rather than trusting

Three separate threads point at the same eighteen-month stretch. The FDA's Advanced Manufacturing Technologies designation programme, guidance finalised December 2025, creates a real regulatory fast lane for qualifying technology.6262US Food and Drug Administration. "Advanced Manufacturing Technologies Designation Program." FDA, final guidance December 2025, under authority established by the Food and Drug Omnibus Reform Act of 2022. Accessed August 10, 2026: designation triggers expedited review for drug applications manufactured using the designated technology.Primaryagency guidance page The national biomanufacturing institute's Project Call 10.1 -- $8 million, concepts due September 9, 2026, with "Intelligent Biomanufacturing Through Artificial Intelligence, Digitalization, and Advanced Process Control" as the named lead topic -- funds the standards work underneath it.6363National Institute for Innovation in Manufacturing Biopharmaceuticals. "NIIMBL Announces Project Call 10.1 to Advance Biopharmaceutical Manufacturing Technology and Workforce Capabilities." NIIMBL, August 3, 2026. Accessed August 10, 2026: a combined $8 million of project activities under Project Call 10.1; concept submissions due September 9, 2026 at 5pm ET; lead technical topic "Intelligent Biomanufacturing Through Artificial Intelligence, Digitalization, and Advanced Process Control."Primarythe institute's own announcement And the patent data below shows the intellectual property being staked out in real time rather than after the fact. None of the three is news alone. Together they describe a regulator, a funding body and inventors all moving on the same technology category in the same window.

One historical parallel is worth naming and then discounting to its proper weight. The last time US federal policy formally named biomanufacturing as its own category, in December 2016, patent applications using that exact word in the invention title went from essentially zero to a sustained five to nine a year within about two years. Whether AI-enabled biomanufacturing follows the same naming-to-patenting lag is a pattern match from a single prior instance. One data point is not a rule, and it is offered here as something to watch rather than something to rely on.

Where the next unit of margin has come from. The workforce data below -- 36% of facilities struggling to hire process-development staff -- reads differently beside Samsung Biologics' margin trajectory. A company that raises operating margin from 25.2% to 45.4% while its industry cannot hire is not solving its capacity problem by adding people. The institute's own funding priority names the alternative directly: automation, digitalisation and advanced process control, not headcount. Read together, the two describe where the marginal unit of output has been coming from -- sensors, software and closed-loop control raising throughput in plants already built, rather than steel for new ones. This piece has no forward financial data and offers no view on what happens next; it has enough to say the direction is consistent across every filing and funding signal it examined.

The demand already committed

Almost every report of this data quotes a cumulative total. A cumulative total cannot say whether registration is accelerating, flat or slowing -- which is the only thing about it that bears on future manufacturing capacity. So the underlying series was pulled directly from ClinicalTrials.gov year by year and analysed here rather than taken from a summary.44ClinicalTrials.gov API v2, /api/v2/studies with query.intr set to the quoted exact phrase, plus filter.advanced=AREA[StudyFirstPostDate]RANGE[...] for the annual series. ClinicalTrials.gov. Queried August 10, 2026: 1,189 gene-therapy studies (259 recruiting), 2,506 cell-therapy (712 recruiting), 2,321 CAR-T (885 recruiting), plus first-posted counts for every year 2005-2026. The query form matters and is stated for that reason: dropping the quotation marks returns 3,454, 28,162 and 2,581 for the same three terms. The counts overlap and are not additive. Raw series and the analysis script are held with this piece's working files.Primarytrial registry5858Statistics computed independently for this piece from the raw series above, in Python, rather than taken from any source's own summary figure. Method: growth stated as the mean year-over-year change with a 20,000-resample percentile bootstrap 95% confidence interval, alongside the bookend compound rate and the full range of that compound rate recomputed across every window of five years or longer. The bootstrap is used in preference to a closed-form interval because ten to sixteen annual observations of noisy counts do not satisfy the assumptions a t interval would need. Where an interval crosses zero, that is stated rather than resolved. The intent is Motulsky's: an interval and a sample size, never a bare point estimate carrying more precision than the data holds. Raw pulls (SEC EDGAR XBRL company-facts, USPTO file-wrapper records, ClinicalTrials.gov API v2) and the analysis script are held with this piece's working files.Primarycomputed from the filed and registry data cited above

Counts are of studies first posted in each calendar year. 2026 is excluded because it is a partial year, and a partial year plotted beside full ones reads as a collapse the calendar produced. The y-axis starts at zero, as a count axis must.

All three modalities set their registry record in 2025, and their second-highest in 2024. Computed across the sixteen annual observations rather than from two endpoints: cell-therapy registrations grew at a mean 17.3% a year (bootstrap 95% confidence interval 5.3-28.7%), gene therapy at 13.2% (95% CI -0.3 to 27.2%), and CAR-T at 41.5% (95% CI 21.1-63.8%). Those intervals are wide because sixteen noisy annual counts do not support a tighter one, and the gene-therapy interval crosses zero -- which is the honest way of saying that gene-therapy registration growth is real over the full period but not statistically distinguishable from flat on a year-to-year basis. Bookend compound growth rates, the figure most reports quote, are 14.8%, 9.9% and 35.7% respectively; the same rate recomputed across every five-year-or-longer window in the series ranges 8.0-21.6%, 0.0-23.4% and 14.7-90.4%. A single compound rate for any of these is an artefact of which two years someone picked.

A note on the counts themselves, because it changes a claim. These totals are sensitive to the exact query. Searching the registry for the unquoted term rather than the exact phrase returns 2,581 CAR-T studies and 28,162 cell-therapy studies rather than 2,321 and 2,506. Earlier coverage of this data, including this site's own, reported 2,580 CAR-T against 2,505 cell-therapy trials and noted the oddity that a nominal subset appeared larger than its superset. Run with one consistent query form, the oddity disappears: cell therapy is larger than CAR-T, as it should be. The anomaly was a mixed query, not a registry tagging failure -- which is worth correcting in public rather than leaving as a curiosity.44ClinicalTrials.gov API v2, /api/v2/studies with query.intr set to the quoted exact phrase, plus filter.advanced=AREA[StudyFirstPostDate]RANGE[...] for the annual series. ClinicalTrials.gov. Queried August 10, 2026: 1,189 gene-therapy studies (259 recruiting), 2,506 cell-therapy (712 recruiting), 2,321 CAR-T (885 recruiting), plus first-posted counts for every year 2005-2026. The query form matters and is stated for that reason: dropping the quotation marks returns 3,454, 28,162 and 2,581 for the same three terms. The counts overlap and are not additive. Raw series and the analysis script are held with this piece's working files.Primarytrial registry

Read narrowly, every registered study still represents a real demand claim on GMP viral-vector or cell-processing capacity that either exists today or has to exist inside that trial's own timeline. Set that against the supply side. One fewer independent tier-one contract manufacturer, and the largest one running at a five-year-high margin. If margin expansion at Samsung Biologics reflects capacity running close to full, and Catalent's absorption removed one alternative at the same time registrations were setting records, then the setup is a demand curve that is not slowing meeting a supply side that just got one name shorter. This piece does not have the data to say whether that gap is already binding. It has enough to say the gap is worth attention before it is.

Innovation intensity, by patent title

US patent applications with "biomanufacturing" literally in the invention title, by filing year, queried directly against USPTO patent file-wrapper data: 76 total matches across all years.6464US Patent and Trademark Office, patent file-wrapper data, applications with "biomanufacturing" in the invention title; the by-year series recomputed here from the 76 returned application records' own filingDate fields rather than from a summary table. USPTO Patent Public Search; a reader without an API key can reproduce the shape of the search at Google Patents. Queried August 10, 2026. Title-text only, not classified by CPC code -- a narrower proxy than a field-wide patent count.Primarypatent-office data, narrowly scoped

Near-zero before 2015, then a step to five to nine filings a year from 2018 that has held for eight straight years. Years with no filings are drawn as zero rather than omitted, because the source table skips them and their omission would make the step read as a plotting artefact. 2026 is a partial year and is drawn hollow.

Recomputed from the raw application records rather than from a summary: filings averaged 0.6 a year across 2003-2017 and 7.9 a year across 2018-2025 (bootstrap 95% confidence interval 6.9-8.8), a roughly thirteen-fold shift in LEVEL.5858Statistics computed independently for this piece from the raw series above, in Python, rather than taken from any source's own summary figure. Method: growth stated as the mean year-over-year change with a 20,000-resample percentile bootstrap 95% confidence interval, alongside the bookend compound rate and the full range of that compound rate recomputed across every window of five years or longer. The bootstrap is used in preference to a closed-form interval because ten to sixteen annual observations of noisy counts do not satisfy the assumptions a t interval would need. Where an interval crosses zero, that is stated rather than resolved. The intent is Motulsky's: an interval and a sample size, never a bare point estimate carrying more precision than the data holds. Raw pulls (SEC EDGAR XBRL company-facts, USPTO file-wrapper records, ClinicalTrials.gov API v2) and the analysis script are held with this piece's working files.Primarycomputed from the filed and registry data cited above What the data does not support is a shift in TREND after the step: the mean year-over-year change within 2018-2025 is -0.14 filings a year, with a 95% interval running from -1.86 to +1.43. That interval spans zero, so on eight annual counts no within-plateau trend is established in either direction. The honest summary is a step followed by a plateau, not a step followed by a climb -- and anyone reading this series as evidence of accelerating innovation is reading something that is not in it.

Read this narrowly in one further respect. It counts only applications using the exact word in the title, not the field's whole patent activity -- that would require a classification search across C12M and related classes, which was not run this pass and is flagged as a gap rather than estimated. What it does show cleanly is the step-change, which is consistent with "biomanufacturing" becoming an established enough term of art to patent against by name around the same time the contract-manufacturing and single-use-systems build-out was accelerating.

What changed in 2025 and 2026

The binding constraint moved from capital to people. BioPlan Associates survey data, reported through the trade press, puts 36% of facilities as unable to hire process-development staff, 28% struggling to hire downstream production personnel, and 27% short of process engineers.55Langer E (BioPlan Associates), interviewed in "Industry Outlook 2026: Biopharma Industry Confronts Skills Shortage with Strategic Automation (Part 2)." BioPharm International, January 23, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: "36% ... unable to hire process development staff. 28% struggle to hire downstream production personnel. 27% ... lack sufficient process engineers."Secondarytrade-press report of survey data; the underlying survey not independently accessed The strongest near-term innovation demand has shifted toward continuous upstream automation specifically to reduce reliance on scarce talent, rather than toward further expansion of low-cost single-use consumables.

Onshoring is real and is moving faster than the workforce can follow. Over $370 billion has been committed to domestic US pharmaceutical manufacturing plants since 2025, while core technical hiring -- scientists, quality specialists, process engineers -- is expected to ramp over a decade rather than a year.6565Moran L. "Onshoring Is Here. The Hiring Wave Is Still a Decade Out." Genetic Engineering and Biotechnology News, August 4, 2026. Accessed August 10, 2026: over $370 billion committed to domestic US pharmaceutical manufacturing plants since 2025; decade-scale technical hiring ramp.Secondarytrade press Contract manufacturers are explicitly filling the gap while in-house plants come online, and future plants skew toward automation and lights-out operation, which shifts demand toward automation engineers and regulatory specialists over traditional production roles.

AI and digitalisation moved from pilot to adoption. Multiple sources report uptake across bioprocess control, real-time monitoring, analytical testing and batch-release documentation -- explicitly including regulators promoting AI-enabled manufacturing and automated documentation rather than merely tolerating it.6666Mirasol F, Smith AK. "How AI-Driven Batch Review May Reduce Biomanufacturing Labor Burden." BioPharm International, April 23, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: AI and machine learning handling routine batch-record review under a review-by-exception model.Secondarytrade press5454BioPlan Associates. "2026 23rd Annual Report and Survey of Biopharmaceutical Manufacturing Capacity and Production." GII Research, July 14, 2026. Accessed August 10, 2026. Sizes production capacity rather than market revenue, which is why it does not appear as a row in the revenue comparison.Secondarycommercial survey report The next section takes that claim apart by maturity tier.

Contract-manufacturing business models are bifurcating. Trade coverage describes a split between regionalising, specialising manufacturers and broad integrated one-stop providers, with capacity increasingly aligned to specific modalities -- cell and gene therapy, antibody-drug conjugates -- rather than held as general-purpose capacity. The same coverage describes geography itself becoming a quality attribute, with proximity to regulators and regional compliance infrastructure treated as competitive differentiators, and names the BIOSECURE Act as a driver of geographically constrained sourcing.6767"The End of Business as Usual: How CDMOs will Evolve Operations in 2026." Contract Pharma, April 1, 2026. Accessed August 10, 2026: "some CDMOs are regionalizing and specializing, others are doing the opposite -- aggressively expanding their service breadth to become integrated, 'one-stop-shop' partners"; a move toward late-stage complex commercial work in cell and gene therapies and antibody-drug conjugates; "Local strategy will become a quality attribute," with the BIOSECURE Act named as signalling "a shift toward geographically constrained sourcing and manufacturing models." A Section 232 tariff framing carried by earlier coverage was not present in the article text this pass and is not repeated.Secondarytrade press

Policy is now a direct input to site selection, not background noise. Both the contract-manufacturing and onshoring coverage above treat legislation and tariffs as active factors in sourcing and site decisions rather than risks to be monitored.6767"The End of Business as Usual: How CDMOs will Evolve Operations in 2026." Contract Pharma, April 1, 2026. Accessed August 10, 2026: "some CDMOs are regionalizing and specializing, others are doing the opposite -- aggressively expanding their service breadth to become integrated, 'one-stop-shop' partners"; a move toward late-stage complex commercial work in cell and gene therapies and antibody-drug conjugates; "Local strategy will become a quality attribute," with the BIOSECURE Act named as signalling "a shift toward geographically constrained sourcing and manufacturing models." A Section 232 tariff framing carried by earlier coverage was not present in the article text this pass and is not repeated.Secondarytrade press6565Moran L. "Onshoring Is Here. The Hiring Wave Is Still a Decade Out." Genetic Engineering and Biotechnology News, August 4, 2026. Accessed August 10, 2026: over $370 billion committed to domestic US pharmaceutical manufacturing plants since 2025; decade-scale technical hiring ramp.Secondarytrade press

AI on the plant floor, graded by tier

"AI in biomanufacturing" is not one thing, and the difference between its uses is the difference between drafting a document and controlling a process that ends in a patient's vein. Three tiers, in descending order of production maturity. Every claim below was re-fetched from its source this pass, and where a source is a vendor's own account of its own product, this piece says so in the prose rather than leaving the reader to infer it from a URL.

Tier 1, documentation. This is the most mature tier by a wide margin, for a structural reason: drafting and formatting a structured technical document from data that already exists is closer to conventional software engineering than to novel science, and the results are reported by named operating companies rather than projected.

Novo Nordisk built NovoScribe, a documentation platform combining Claude Code, Claude models on Amazon Bedrock, and MongoDB Atlas with retrieval-augmented generation, according to Anthropic's own published customer case study -- which is to say, a vendor describing its own customer's use of its own product, not independent verification.6868Anthropic. "Novo Nordisk" customer case study. claude.com. Accessed August 10, 2026: NovoScribe combining Claude Code, Claude models on Amazon Bedrock and MongoDB Atlas with retrieval-augmented generation; clinical study report production "10+ weeks to 10 minutes"; writing time cut by 90%; a stated baseline of "2.3 CSRs per year" by hand. A vendor describing its own customer's use of its own product, not independent verification.Secondaryvendor case study The claims recorded there: clinical study report production dropping from "10+ weeks to 10 minutes," writing time on those reports cut by 90%, against a stated baseline of staff writers previously averaging 2.3 reports a year by hand. The case study names a digitalisation strategy director with a molecular-biology doctorate and, by his own description, little programming experience, who now prototypes features in natural language rather than filing tickets. The stated next step is automating full Common Technical Documents -- complete regulatory submission packages -- not individual study reports.

Bluenote, per a second Anthropic case study and so also a vendor account, builds agents for life-sciences clients including Guardant Health and Eledon Pharmaceuticals across three types: document-processing agents, technical-documentation agents that produce regulatory documents with citations, and multi-step workflow agents.6969Anthropic. "Bluenote" customer case study. claude.com. Accessed August 10, 2026: three agent types for life-sciences clients including Guardant Health and Eledon Pharmaceuticals; Guardant Health cited as having boosted a specialised workflow's efficiency by 40-50%; broader "10x faster analysis" and "50-75% acceleration" figures self-reported and unaudited.Secondaryvendor case study Guardant Health is cited as having improved a specialised workflow's efficiency by 40-50% through a quality-control agent. Bluenote's broader unattributed metrics -- "10x faster analysis" for protocol parsing, "50-75% acceleration in regulatory document production" -- are self-reported and unaudited, and are presented here as the vendor's claim rather than as a finding.

i3 Pharmaceuticals is the closest of the three to manufacturing operations proper. Per a case study published by the engineering firm that did the build -- again vendor-authored, though the operational detail is unusually specific -- i3 is a specialty generic pharmaceutical company in Warminster, Pennsylvania, running a 150,000-square-foot facility under full cGMP oversight with FDA prior-approval and GMP inspections, manufacturing posaconazole delayed-release, maraviroc and guanfacine.7070AgenticBricks. "i3 Pharmaceuticals custom SaaS." agenticbricks.com. Accessed August 10, 2026: a specialty generic pharmaceutical company in Warminster, Pennsylvania, 150,000 square feet under full cGMP oversight; a production system delivered in three months from an empty repository; "Claude Code proposes, engineers verify, tests validate, reviewers approve. No generated code reached production without human review and automated verification." Published by the engineering firm that did the build, about its own client.Secondaryvendor-authored case study The build used Claude Code to produce a custom operational platform, deployed in three months from an empty repository against typical regulated-pharma software timelines the case study puts at twelve to eighteen months. The stated discipline is quoted directly: "Claude Code proposes, engineers verify, tests validate, reviewers approve. No generated code reached production without human review and automated verification." The tool generated audit-trail middleware, role-based access control, input validation and logging, wrote test suites, performed cross-file refactoring and drafted documentation from the codebase -- and was explicitly not used as an autonomous agent in the regulated environment: "We used it as an accelerator under engineering supervision."

Tier 2, discovery and design. Here the evidence gets both more interesting and more mixed, because one source is an independent third-party benchmark rather than a vendor's own account.

Scale AI's research arm published a benchmark in June 2026 running three agentic coding tools -- Claude Code on Opus 4.7, Codex on GPT-5.5, and Gemini CLI on Gemini 3.1 Pro -- head to head across 66 verifiable, domain-expert-curated tasks spanning target identification and validation, hit discovery, hit-to-lead and lead optimisation, plus a smaller set of protein-engineering and cancer-genomics tasks.7171Scale AI Labs. "Coding agents for drug discovery." labs.scale.com, June 6, 2026. Accessed August 10, 2026: 66 domain-expert-curated tasks run through the Biomni framework against Claude Code (Opus 4.7), Codex (GPT-5.5) and Gemini CLI (Gemini 3.1 Pro); Claude Code declined 5 tasks on content-policy grounds; "on 43 of the 66 tasks at least one of the three agents produced a correct answer"; the bottleneck identified as high-level planning rather than domain knowledge. Scale's own research blog, benchmarking competing products rather than its own.Secondaryindependent industry benchmark All three ran inside Biomni, an open-source biomedical agent framework from Stanford's SNAP lab that supplies a few hundred biomedical functions. Codex and Gemini answered all 66; Claude Code answered 61, declining five on content-policy grounds -- tasks involving HIV-1 reverse transcriptase, SARS-CoV-2 PLpro and two competitor-patent questions. That is an honest tradeoff worth reporting plainly rather than glossing: a conservative content policy cost one tool five otherwise answerable tasks. On the 61 all three attempted, Scale's own summary is that Codex is most accurate on well-defined single answers, Gemini explores more and reaches answers on open-ended tasks the others miss, and Claude Code holds up better on long tasks that pile up several constraints at once. By category, Codex led on structure tasks at 73% and database screening at 62%; Claude Code led on molecular biology at 58% and patent mining at 61%. Across the full set, on 43 of the 66 tasks at least one of the three produced a correct answer -- and the benchmark's central finding is that the bottleneck is high-level planning rather than domain knowledge, since supplying an expert's step-by-step method, without the answer, made most previously unsolved tasks solvable. Long-horizon tasks stacking many constraints remained hardest for all three.

Boltz, an open molecular-structure-prediction suite now callable through Anthropic's science-focused workbench, published its own account of two internal experiments -- a vendor blog post, though the methodology and numbers are specific rather than promotional.7272Boltz. "Boltz and Claude." boltz.bio. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: a CD3-by-CD19 bispecific designed across a 20,000-call budget reaching minimum binding confidence 0.62 across three interfaces; a smaller model skipping the antibody-numbering step and invalidating all its designs; a hit-to-lead predictor reaching log-scale Pearson correlation 0.38 against 0.23 for the raw model and 0.14 for a fixed non-agentic regressor. A vendor blog post, with specific and reproducible-style methodology.Secondaryvendor blog In the first, an agent designed a CD3-by-CD19 bispecific antibody by iteratively calling the Boltz API across a budget of 20,000 design calls, with no human in the loop per turn; the best result reached a minimum binding confidence of 0.62 across the three interfaces. A parallel run using a much smaller model skipped the antibody-numbering step, redesigned a framework region containing a structurally load-bearing disulphide cysteine, and rendered all of its designs invalid -- a concrete illustration of a capability gap on a task with a hard biological constraint the API itself does not enforce. In the second experiment, an agent fine-tuned a hit-to-lead binding affinity predictor against live medicinal-chemistry assay data, deliberately drawn from the subset of the test set where the base model is known to perform poorly. The agent's blended prediction reached a log-scale Pearson correlation of 0.38 on held-out candidates, against 0.23 for the raw model prediction and 0.14 for a fixed non-agentic regressor on the same inputs. The improvement came specifically from per-target judgement calls -- which features to trust, when to override the model -- that a fixed pipeline never makes.

Tier 3, live process control, and the tier most honest about its own limits.

Eli Lilly uses digital twins and agentic AI in live manufacturing operations, per independent trade journalism rather than a company release.7373"Eli Lilly taps AI, digital twin technology to boost manufacturing capacity." Pharma Manufacturing, June 11, 2026. Accessed August 10, 2026: Scot Lindsey, senior vice-president and information officer for manufacturing and quality -- "We don't want any black box. We want to fully understand why our AI solution is recommending what it's recommending"; GLP-1 products kept off the FDA shortage list since late 2024; the October 2025 NVIDIA partnership.Secondaryindependent trade journalism Lilly credits AI-assisted production scaling with keeping its GLP-1 products off the FDA drug-shortage list since late 2024, after earlier shortages. Its senior vice-president and information officer for manufacturing and quality states the approach rests on three principles -- human-in-the-loop, explainability, transparency -- and puts the second plainly: "We don't want any black box. We want to fully understand why our AI solution is recommending what it's recommending." A digital twin identified a process bottleneck, was tested in a pilot and then implemented, yielding what he describes as a tremendous reduction in the time that bottleneck step took. That is a qualitative claim, and it is reported as one here rather than converted into a percentage the source does not give. Lilly's chief information and digital officer is separately quoted, within the same piece, saying the company literally made more product in the prior year than it possibly could have without AI. In October 2025 Lilly partnered with NVIDIA to build what the two describe as the industry's most powerful pharma-owned AI supercomputer, for both discovery and manufacturing digital twins.

Adaptive Agent-Oriented System Control is a research framework from Seyed Soheil Mansouri at the Technical University of Denmark, described in independent trade coverage and developed with collaborators including Novo Nordisk, Henkel, Boston Consulting Group, Zapata Computing and ORCA Computing.7474"Adaptive agent-oriented control for biomanufacturing systems." Genetic Engineering and Biotechnology News, April 29, 2026. Accessed August 10, 2026: Seyed Soheil Mansouri of the Technical University of Denmark on decentralised "agent hives" of "rule-based, mathematically informed agents"; agentic AI "is not yet fully ready for complete, independent control in biopharmaceutical manufacturing"; "any AI that directly affects medicine quality still needs strong human oversight and full approval"; the shadow-mode adoption path.Secondaryindependent trade press It uses decentralised agent hives -- in Mansouri's framing, rule-based, mathematically informed agents grounded in physics, chemistry and biology -- explicitly not one monolithic model coordinating everything, but distributed specialised agents coordinating digital twins and real-time control. Four case studies reportedly reduced deviation durations and prevented shutdowns during severe fault scenarios. The load-bearing quote for the whole tier comes from Mansouri directly: agentic AI "is not yet fully ready for complete, independent control in biopharmaceutical manufacturing," and "any AI that directly affects medicine quality still needs strong human oversight and full approval." His recommended adoption path is explicit -- run new agentic systems in shadow mode first, where the system watches everything and gives recommendations but makes no changes without human oversight, and expand its role only once operators have built confidence in it.

WuXi Biologics' PatroLab launched as a digital-twin platform combining Raman-based process analytical technology with predictive modelling, per the company's own press release.7575WuXi Biologics. PatroLab digital-twin platform press release. wuxibiologics.com, January 12, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: Raman-based process-analytical-technology monitoring of "more than 40 key process performance and product quality attributes" in real time, with data density per batch increased by "nearly 1,000 times" against traditional methods, framed around enabling real-time release testing.Secondaryvendor press release It monitors more than 40 key process performance and product quality attributes in real time and non-invasively, and claims to increase data density per batch by nearly 1,000 times against traditional methods. Beyond monitoring, it layers in predictive modelling and what-if scenario analysis for process risk and technology-transfer troubleshooting, framed explicitly around enabling real-time release testing in line with FDA and EMA guidance -- which is the same regulatory push toward compressing the fourteen-day sterility hold described above.

Reading the three tiers together. The maturity gradient is not an accident of which stories got published. It tracks the real difficulty gap between drafting a structured document from data that already exists, reasoning over a well-defined design problem with a checkable answer, and exercising real-time judgement inside a live GMP process where an error reaches a patient. Not one of the Tier 3 sources -- independent academic research, independent trade press and a manufacturer's own press release alike -- claims agentic AI is trusted with unsupervised control of anything touching product quality today. Every one of them states human oversight as a design requirement rather than a temporary caveat. Worth naming plainly: this piece was itself produced by an agentic research-and-drafting pipeline, cross-checked against independent sources rather than accepted from a single pass -- the same discipline the i3 build describes, and the same scepticism about unverified vendor claims applied here to every source above.

Who makes it, and who stands beside them

Contract manufacturers, the outsourced production layer. Convergent naming across independent 2026 rankings puts the consistent top tier as Lonza, Samsung Biologics, WuXi Biologics, Catalent and Thermo Fisher (Patheon), with the top nineteen collectively linked to about half of FDA drug approvals since 2015.7676"Top CDMO Companies 2026: Largest Contract Development and Manufacturing Organizations Ranked." IntuitionLabs, August 8, 2026; cross-checked against Slabodkin G, "4 Contract Development and Manufacturing Organizations to Watch in 2026," Pharma Manufacturing, January 8, 2026. Accessed August 10, 2026: consistent top tier of Lonza, Samsung Biologics, WuXi Biologics, Catalent and Thermo Fisher (Patheon).Secondarycommercial ranking sources, cross-checked across two independently verified pages; individual methodologies not audited The next tier includes Fujifilm Diosynth Biotechnologies, Boehringer Ingelheim BioXcellence, AGC Biologics, Rentschler Biopharma, Siegfried, Recipharm and Sandoz. The most consequential recent move in the tier is the one the filings record above: Novo Holdings' acquisition of Catalent, completed December 18, 2024. Reported site acquisitions and capacity figures for other members of the tier were not independently re-verified this pass and are not repeated here. The history above shows this tier is not uniformly old money: Lonza traces to 1897 and entered biotech in the 1980s and 1990s, while Samsung Biologics, now among the largest by capacity, was founded only in 2011.

Equipment and consumables vendors were not individually verified in the sources pulled this pass. Cytiva, Sartorius, Thermo Fisher's instruments arm and Merck's MilliporeSigma are the widely known names in this tier, but no source in this research verified their specific 2025-2026 market positioning, and it is marked not found rather than asserted.7777No source located this pass verified the specific 2025-2026 market positioning of Cytiva, Sartorius, Thermo Fisher's instruments business, or Merck's MilliporeSigma. Flagged as not established rather than asserted from general knowledge.Not foundno source What the manufacturer-focused sources do confirm is the enabling layer: single-use bioreactor systems and continuous-manufacturing tooling are the specific technologies behind capacity expansion and asset-light outsourcing economics, and equipment partnerships are increasingly public-facing -- Lonza with Oxford Nanopore on accelerated GMP quality-control testing for mRNA therapeutics, Catalent with GelMEDIX on iPSC-derived cell therapy manufacturing.

Precompetitive research consortia sit upstream of both. The national biomanufacturing institute is not a manufacturer, a contract manufacturer or an equipment vendor: it is a precompetitive, Manufacturing USA-affiliated consortium doing standards development, workforce training and the kind of research individual companies will not fund alone because the benefit is shared across the industry rather than captured by one firm.2222US Department of Commerce and National Institute of Standards and Technology. "U.S. Secretary of Commerce Penny Pritzker Announces Biopharmaceutical Manufacturing Institute Joining Manufacturing USA Network." NIST, December 16, 2016. Accessed August 10, 2026: $70 million award establishing NIIMBL as the eleventh Manufacturing USA institute, the first with an industry-proposed focus area and the first funded directly by the Department of Commerce.Primaryfederal agency announcement The manufacturer-focused sources gesture at precompetitive consortia in generic terms -- shared standards, regulatory alignment, joint research cutting development time -- without naming any specifically, which is consistent with that layer sitting adjacent to the commercial one rather than competing inside it.

The regulatory and policy backdrop

FDA is building a designation pathway for advanced manufacturing technology. The Advanced Manufacturing Technologies designation programme finalised its guidance in December 2025, under authority established by the 2022 Food and Drug Omnibus Reform Act; designation triggers expedited review for drug applications manufactured with the designated technology.6262US Food and Drug Administration. "Advanced Manufacturing Technologies Designation Program." FDA, final guidance December 2025, under authority established by the Food and Drug Omnibus Reform Act of 2022. Accessed August 10, 2026: designation triggers expedited review for drug applications manufactured using the designated technology.Primaryagency guidance page Separately, a proposed rule is on the Unified Agenda under RIN 0910-AJ31, titled "Amendments to 21 CFR Parts 210 and 211; Current Good Manufacturing Practices; Advanced Manufacturing, Distributed, and Point of Care Manufacturing." Its abstract states the intent directly: to clarify how cGMP applies to continuous, distributed and point-of-care manufacturing, and to establish flexible batch definitions, real-time monitoring control strategies and lifecycle-based validation in place of fixed end-of-batch testing.7878US Office of Information and Regulatory Affairs, Unified Agenda entry RIN 0910-AJ31, "Amendments to 21 CFR Parts 210 and 211; Current Good Manufacturing Practices; Advanced Manufacturing, Distributed, and Point of Care Manufacturing." Reginfo.gov. Accessed August 10, 2026: proposed rule stage; the abstract describes science- and risk-based compliance approaches, flexible batch definitions, real-time monitoring control strategies and lifecycle-based validation, building on Section 3213 of the 2022 Food and Drug Omnibus Reform Act. Not finalised.Primarythe regulatory agenda itself It remains at the proposed-rule stage and is not in force.

AI-based release testing specifically. FDA's advanced-manufacturing page confirms an active Advanced Manufacturing Innovation Hub developing regulatory-science benchmarks for platform technologies including closed-loop controls and smart manufacturing, building on ICH Q13 on continuous manufacturing, finalised in March 2023.7979US Food and Drug Administration. "Advanced Manufacturing," Office of Clinical Policy and Programs. FDA. Accessed August 10, 2026: the Advanced Manufacturing Innovation Hub and regulatory-science benchmarks for platform technologies including closed-loop controls; ICH Q13 on continuous manufacturing finalised March 2023.Primaryagency page Industry reporting independently describes regulators actively promoting AI-enabled manufacturing and automated batch-release documentation rather than merely tolerating it.6666Mirasol F, Smith AK. "How AI-Driven Batch Review May Reduce Biomanufacturing Labor Burden." BioPharm International, April 23, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: AI and machine learning handling routine batch-record review under a review-by-exception model.Secondarytrade press

Domestic capacity is an explicit national-security priority. Executive Order 14293, "Regulatory Relief to Promote Domestic Production of Critical Medicines," signed May 5, 2025, directs FDA, EPA and other agencies to streamline reviews and centralise environmental permitting specifically to cut the time it takes to build new capacity.8080Executive Order 14293, "Regulatory Relief to Promote Domestic Production of Critical Medicines." whitehouse.gov, May 5, 2025. Accessed August 10, 2026.Primarythe order itself The BIOSECURE Act, enacted through the fiscal-2026 National Defense Authorization Act signed December 18, 2025, restricts federal procurement, contracts and grant funding involving equipment or services from designated biotechnology companies of concern; implementation phases in over the following year through amendment of the Federal Acquisition Regulation, and organisations with federal funding are advised to begin planning now.8181BIOSECURE Act, enacted via the National Defense Authorization Act for Fiscal Year 2026, signed December 18, 2025; analysis via K&L Gates, "BIOSECURE Act: What You Need to Know." K&L Gates, January 20, 2026. Accessed August 10, 2026. (Primary for the statute; Secondary for the law-firm analysis of implementation timing)

Drug shortages as a capacity-strain proxy. openFDA's drug-shortage database, queried directly, showed 1,179 currently active US drug shortages.8282openFDA drug-shortage database, queried directly at api.fda.gov/drug/shortages.json with search=status:"Current". open.fda.gov. Queried August 10, 2026: 1,179 currently active US drug shortages against 1,651 records unfiltered, with the API's own meta.last_updated reporting August 8, 2026. Spans the whole pharmaceutical supply chain, not biologics specifically; not broken out by product type this pass.PrimaryFDA data That spans the whole pharmaceutical supply chain rather than biologics specifically -- most active shortages are small-molecule generics -- so it should be read as a general manufacturing-strain indicator, not a biologics number. It was not broken out by biologic against small molecule this pass, and that limit is flagged rather than glossed.

The EU already has a route that bypasses market authorisation entirely. Long before any battlefield-manufacturing programme, European pharmaceutical law carved out a path for a hospital to manufacture a bespoke advanced therapy for one named patient and use it without ever seeking a marketing authorisation.

The left path is the one every product in the catalogue above took. The right path skips it entirely -- and skips the trial programme with it.

Article 3(1) of Directive 2001/83/EC codifies the older, general magistral formula exemption: a pharmacist compounds a drug against an individual prescription for one named patient, and because the product never enters the market in the directive's sense, no marketing authorisation is required. Article 28 of the Advanced Therapy Medicinal Products Regulation extends the same logic specifically to cell, gene and tissue-engineered therapies, through what practitioners call the hospital exemption.8383Regulation (EC) No 1394/2007 of the European Parliament and of the Council on advanced therapy medicinal products, Article 28, read with Article 3(1) of Directive 2001/83/EC. EUR-Lex. Accessed August 10, 2026: the hospital exemption for custom-made advanced therapy medicinal products prepared on a non-routine basis, under individual prescription, for an individual patient, within a single member state.PrimaryEU legislative text A hospital may manufacture a custom ATMP on a non-routine basis, under a physician's individual prescription, for one specific patient, and administer it within the same member state, entirely outside the European Medicines Agency's centralised pathway -- no Phase 1-3 programme, no EMA filing. The exemption is not unconditional: manufacturing still requires national authorisation, and the regulation mandates traceability and pharmacovigilance standards equivalent to those imposed on centrally authorised products.

Industry legal analysts are explicit that innovators see this as a competitive threat rather than a curiosity. A review of the pending EU pharmaceutical-legislation overhaul states plainly that expanded compounding rules mean "we are likely to see pharmacies compounding product in a way that increasingly threatens the market for approved innovator products," since a hospital can produce a functional CAR-T-like therapy under the exemption that competes directly with an approved product whose maker spent a decade and billions on trials.8484"EU Pharma Legislation Review Series: Pharmacy and Hospital Exemptions." Inside EU Life Sciences, April 28, 2023. Accessed August 10, 2026: "we are likely to see pharmacies compounding product in a way that increasingly threatens the market for approved innovator products."Secondaryspecialist legal analysis of primary legislative text The European Commission has itself acknowledged that application is fragmented and inconsistent across member states -- Spain in particular is flagged as treating exemption products almost as de facto approved drugs -- and the reform proposal tries to tighten and harmonise it, adding EMA reporting and data-collection obligations while preserving the core exemption.8585International Bar Association. "The hospital exemption pathway for advanced therapies." ibanet.org, June 2, 2025. Accessed August 10, 2026: fragmented and inconsistent application across member states, with Spain flagged as treating exemption products almost as de facto approved drugs.Secondaryindependent legal analysis The US has no direct equivalent. Right-to-try and compassionate-use pathways are narrower: they route around proof of efficacy for a single desperate case, not around the manufacturing-and-market framework itself, and FDA-regulated 503B outsourcing facilities operate under a materially stricter compounding regime than the EU's magistral formula. That asymmetry is itself part of why US point-of-care manufacturing interest has taken a slower, more centralised shape than Europe's hospital-based path.

Where the current signals point

Inside the next year or two, read forward from what is already scheduled rather than predicted: the September 9, 2026 concept deadline for Project Call 10.1 will show where roughly $8 million in new project funding concentrates, with AI, digitalisation and advanced process control as the named lead topic -- which tracks the workforce constraint above.6363National Institute for Innovation in Manufacturing Biopharmaceuticals. "NIIMBL Announces Project Call 10.1 to Advance Biopharmaceutical Manufacturing Technology and Workforce Capabilities." NIIMBL, August 3, 2026. Accessed August 10, 2026: a combined $8 million of project activities under Project Call 10.1; concept submissions due September 9, 2026 at 5pm ET; lead technical topic "Intelligent Biomanufacturing Through Artificial Intelligence, Digitalization, and Advanced Process Control."Primarythe institute's own announcement Catalent's integration into Novo Holdings closed December 18, 2024; whether that produced any capacity reallocation toward Novo Nordisk's own GLP-1 production is not independently confirmed and is left open here rather than characterised.

Over three to five years, two rulemakings define the environment. The Advanced Manufacturing Technologies designation programme is final; the cGMP amendments for continuous and distributed manufacturing are proposed and not in force. Together they describe a regulatory framework that rewards manufacturers able to demonstrate real-time release and flexible batch definitions -- which is a statement about what the guidance and the proposed rule say, not a claim about what will happen to anyone's revenue. Whether the workforce pipeline catches up to the $370 billion-plus in committed US onshoring capital is genuinely open; the decade-scale hiring ramp described by the trade press is itself an admission that supply and demand are not expected to converge quickly.6565Moran L. "Onshoring Is Here. The Hiring Wave Is Still a Decade Out." Genetic Engineering and Biotechnology News, August 4, 2026. Accessed August 10, 2026: over $370 billion committed to domestic US pharmaceutical manufacturing plants since 2025; decade-scale technical hiring ramp.Secondarytrade press

Pressures, and what relieves them

Pressures. First, the workforce shortage could deepen faster than automation substitutes for it. The survey data above is a 2026 snapshot, not a trend line; if it worsens, capacity utilisation -- apparently already high at Samsung Biologics, per the margin expansion -- becomes a hard ceiling. Second, BIOSECURE and related designation exposure, unresolved and in litigation, as below. Third, the measurement disagreement is itself a structural hazard: participants working from a $40 billion base and participants working from a $19 billion base are planning against different industries. Fourth, supply-chain concentration among a handful of consumables vendors, also below.

Relievers. First, FDA is actively promoting AI-enabled manufacturing and real-time release rather than merely tolerating it, which shortens the qualification path for exactly the technology the workforce constraint makes valuable. Second, the thousands of registered cell and gene trials represent demand already committed through clinical programmes rather than forecast -- a floor under capacity investment independent of whether the market-size forecasts hold. Third, Samsung Biologics' margin trajectory demonstrates that real operating leverage exists at scale, which is a statement about a filed result, not about anyone's future.

A real failure case: Emergent BioSolutions, 2021. Everything above is a growth story; this is the counterweight, and it is the single best argument that capital does not substitute for operational discipline. Emergent's Bayview facility in Baltimore manufactured COVID-19 vaccine drug substance for both AstraZeneca and Janssen in the same building. On March 26, 2021, Janssen notified FDA that it had detected AstraZeneca virus in a Janssen drug-substance batch. FDA's own determination memorandum states the root cause directly: bioreactor media prepared for the cell expansion process "was contaminated in the common weigh and dispense area through contact with the waste path for materials from Area 3," the AstraZeneca manufacturing area.66US Food and Drug Administration. Determination memorandum on Emergent BioSolutions' Bayview facility, June 11, 2021. FDA. Accessed August 10, 2026: bioreactor media "was contaminated in the common weigh and dispense area through contact with the waste path for materials from Area 3 (AZ manufacturing area)"; batches GMP 5 through 9 dispositioned as unsuitable for use.Primaryagency determination memorandum A for-cause inspection that April found the facility not maintained in a clean and sanitary condition, with inadequate waste segregation and training. FDA ordered batches GMP 5 through 9 discarded as unsuitable; two earlier batches were cleared because they predated the facility's waste-overload period. Contemporaneous reporting put the discarded total at 60 million doses' worth of drug substance from this single event, and a later congressional oversight investigation found that "nearly 400 million doses" were destroyed across Emergent's full 2020-2021 quality failures, against a $628 million federal contract.8686Contemporaneous reporting of 60 million doses' worth of drug substance discarded following the March 2021 event. The Washington Post, June 11, 2021. Congressional oversight finding that "nearly 400 million doses" were destroyed across Emergent's 2020-2021 quality failures, against a $628 million federal contract. House Committee on Oversight and Accountability, Democrats. Accessed August 10, 2026. A 75 million upper bound carried by earlier coverage is not supported by any source located this pass and is not repeated.Secondarycontemporaneous press and congressional oversight The lesson generalises: facility design, waste segregation and training failures took down output from a federally subsidised, well-capitalised manufacturing relationship as fast as any funding gap could have.

BIOSECURE and WuXi AppTec, live and unresolved. On June 8, 2026, the Department of War added WuXi AppTec -- a related entity to WuXi Biologics, both part of the WuXi group -- to its statutory 1260H list of Chinese military companies, citing alleged indirect ownership or affiliation with Chinese state and defence-linked entities. WuXi AppTec disputes the designation and is litigating it, having filed a complaint on June 11, 2026 in the US District Court for the District of Columbia. On August 7, 2026, the district's chief judge granted a preliminary injunction blocking enforcement while the underlying suit proceeds, finding that the department's case rested on a misreading of the evidence -- a stake in a fund reported as a stake in a company, and a third party's laboratory work reported as the company's own.8787WuXi AppTec's addition to the Department of War's statutory 1260H list, June 8, 2026, per Holland and Knight client analysis; the complaint and the August 7, 2026 preliminary injunction on the docket of WuXi AppTec v. Department of War, No. 1:26-cv-02069, US District Court for the District of Columbia, via CourtListener. Accessed August 10, 2026: the opinion finds the department's case rested on a misreading of the evidence, specifically a stake in a fund reported as a stake in a company and a third party's laboratory work reported as the company's own. Reported here as an unresolved, actively litigated matter rather than as settled in either direction. (Primary for the docket and opinion; Secondary for the law-firm analysis) A widely quoted phrase about officials having "repeatedly misread the evidence" is press paraphrase rather than the court's own words, and is not used here as a quotation. This is genuinely unresolved in both directions: the merits litigation continues, the formal biotechnology-company-of-concern list is due by December 18, 2026, and a five-year grandfather clause protects pre-existing contracts through 2031 regardless of outcome. The practical stakes are real either way -- one industry estimate cited in law-firm client alerts puts WuXi AppTec's involvement at roughly a quarter of drugs manufactured for the US market, which is why sponsors are reportedly diversifying relationships ahead of any final legal outcome rather than waiting for one.

Supply-chain concentration in consumables. Single-use bioreactor bags, chromatography resins and cell-culture media -- the consumables the entire process above depends on -- are dominated by a small number of vendors. During COVID-driven demand, single-use equipment lead times stretched to 10-12 months, and Cytiva and Pall together committed $1.5 billion over two years specifically to expand capacity -- over $600 million of it to chromatography resins, $400 million to cell-culture media and $300 million to single-use manufacturing, across thirteen sites and 2,000 hires.8888Single-use equipment lead times of 10-12 months during COVID-driven demand. BioProcess International, May 2021, citing ABEC. And the combined Cytiva and Pall commitment: Pall Corporation press release, July 2021. Accessed August 10, 2026: $1.5 billion over two years -- over $600 million to chromatography resins, $400 million to cell-culture media, $300 million to single-use manufacturing and $200 million to other capacity, across thirteen sites with 2,000 hires.Secondarytrade press and a vendor press release Cytiva's own account of a prior single-source disruption -- the 2011 Japan tsunami interrupting supply of a critical raw material for one of its chromatography resins -- illustrates the structural shape of the risk: a handful of vendors, sometimes single-sourcing critical raw materials, sit upstream of essentially the entire industry's purification capacity.

mRNA overcapacity, the direct counter-example. The convergence pattern above -- naming a category precedes an inflection -- has a counter-example worth naming, because a pattern with no counter-example is usually a story. mRNA manufacturing capacity built out for COVID-19 has been markedly underutilised since demand normalised. Moderna's own filings show a strategic resizing initiative launched in the third quarter of 2023 that incurred $1.6 billion in charges that year; full-year 2024 cost of sales still included $495 million in inventory write-downs and $105 million of unutilised manufacturing capacity; and fourth-quarter 2024 results disclose a $238 million non-cash charge to terminate a contract-manufacturing agreement with a Lonza facility as part of that resizing.8989Moderna, Inc. Annual report and quarterly results releases. SEC EDGAR. Accessed August 10, 2026: a strategic resizing initiative launched in the third quarter of 2023 incurring $1.6 billion in charges that year; full-year 2024 cost of sales including $495 million in inventory write-downs and $105 million of unutilised manufacturing capacity; a $238 million non-cash charge disclosed with fourth-quarter 2024 results to terminate a contract-manufacturing agreement.Primaryfiled financial statements This is the same capacity-build enthusiasm documented favourably above, playing out in reverse for one company and one platform.

What this piece does not establish

Several things were looked for and not found, and are recorded here rather than left as silent gaps: segment-level Thermo Fisher bioproduction revenue, which XBRL company-facts data does not carry; a biologics-specific breakdown of the active drug-shortage count; a classification-based patent trend across C12M and related classes, as opposed to the title-text search used above; equipment-vendor 2025-2026 market positioning; Boehringer Ingelheim BioXcellence corporate history, whose own page returned a bot-detection block; day-precise original approval dates for BeneFIX, Recombivax HB and Amtagvi; and a direct Drugs@FDA database record for Zolgensma or Kymriah under their brand names, both of which appear to be indexed under fields the search used does not cover. A $47.7 billion current-year market figure carried by earlier coverage could not be re-verified on its source's own page and is not repeated here. A claim that roughly 23 antibody-drug conjugates are approved globally appeared in a trade source that could not be retrieved by any method this pass, and is dropped rather than asserted. Day-precise approval dates for Kogenate and for the two 1986 interferons could not be pinned to an open primary record and are given to the year and the month respectively. Reported site acquisitions and litre-capacity figures for individual contract manufacturers were not re-verified and are omitted.

Three of this piece's own claims changed under independent re-analysis, and are recorded here because a correction that is not stated is not a correction. The registry counts for CAR-T and cell therapy invert depending on the query form used, which resolves an anomaly earlier coverage reported as a registry tagging failure. The patent series supports a step in level but not a trend after it. And a single compound growth rate for Thermo Fisher's revenue, or for any of the three trial series, is sensitive enough to the choice of endpoints that quoting one without its range is misleading.

No figure above was filled from recall where a live query was available.

Sources, by tier

Ordered by tier, then by subject: filed and regulatory sources first, then registries and peer-reviewed literature, then trade press and commercial research.

Footnotes

1. "Bio-Manufacturing Market -- Global Industry Size," report ID 5022679, a separate report from the preceding one. Research and Markets. Accessed August 10, 2026: $19.08 billion (2024) to $29.57 billion (2030), 7.72% CAGR. (Secondary -- commercial market research) ↩ ↩²

2. Grand View Research. "Biologics Manufacturing Market Size, Share and Trends Analysis Report, 2033," report ID GVR-4-68040-743-9. Grand View Research. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: "valued at USD 40.1 billion in 2025 and is anticipated to reach USD 140.6 billion by 2033, growing at a CAGR of 16.7% between 2026 and 2033." A $47.7 billion 2026 figure carried by earlier coverage was not present on the page this pass and is not repeated. (Secondary -- commercial market research, proprietary methodology not independently auditable) ↩ ↩² ↩³

3. Samsung Biologics. "Annual Business Report." Samsung Biologics Investor Relations. Accessed August 10, 2026: consolidated K-IFRS revenue (operating profit) in KRW billion -- 2020: 1,165 (293); 2021: 1,568 (537); 2022: 2,437 (969); 2023: 2,939 (1,206); 2024: 3,497 (1,321); 2025: 4,557 (2,069). The 2022 figures consolidate a newly acquired US subsidiary and were unaudited per the company's own footnote. (Primary -- company-reported consolidated financials) ↩ ↩² ↩³ ↩⁴ ↩⁵

4. ClinicalTrials.gov API v2, /api/v2/studies with query.intr set to the quoted exact phrase, plus filter.advanced=AREA[StudyFirstPostDate]RANGE[...] for the annual series. ClinicalTrials.gov. Queried August 10, 2026: 1,189 gene-therapy studies (259 recruiting), 2,506 cell-therapy (712 recruiting), 2,321 CAR-T (885 recruiting), plus first-posted counts for every year 2005-2026. The query form matters and is stated for that reason: dropping the quotation marks returns 3,454, 28,162 and 2,581 for the same three terms. The counts overlap and are not additive. Raw series and the analysis script are held with this piece's working files. (Primary -- trial registry) ↩ ↩² ↩³

5. Langer E (BioPlan Associates), interviewed in "Industry Outlook 2026: Biopharma Industry Confronts Skills Shortage with Strategic Automation (Part 2)." BioPharm International, January 23, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: "36% ... unable to hire process development staff. 28% struggle to hire downstream production personnel. 27% ... lack sufficient process engineers." (Secondary -- trade-press report of survey data; the underlying survey not independently accessed) ↩ ↩²

6. US Food and Drug Administration. Determination memorandum on Emergent BioSolutions' Bayview facility, June 11, 2021. FDA. Accessed August 10, 2026: bioreactor media "was contaminated in the common weigh and dispense area through contact with the waste path for materials from Area 3 (AZ manufacturing area)"; batches GMP 5 through 9 dispositioned as unsuitable for use. (Primary -- agency determination memorandum) ↩ ↩²

7. "Biomanufacturing's blurry number." BigBio.ai, August 10, 2026. Covers the market-size disagreement, 2025-2026 trends and the regulatory backdrop in summary. (This site's own prior publication) ↩

8. "The consolidation nobody priced in." BigBio.ai, August 10, 2026. Covers the filings and registry evidence in summary. (This site's own prior publication) ↩

9. Cohen SN, Chang ACY, Boyer HW, Helling RB. "Construction of Biologically Functional Bacterial Plasmids In Vitro." Proceedings of the National Academy of Sciences 1973;70(11):3240-3244. Accessed August 10, 2026. (Primary -- the original paper) ↩

10. Genentech. "Our Founders." gene.com. Accessed August 10, 2026: "founded ... in 1976, by ... Robert A. Swanson and ... Dr. Herbert W. Boyer." (Primary -- the company's own founder history) ↩

11. US Food and Drug Administration, Drugs@FDA approvals database, queried by application number. accessdata.fda.gov. Queried August 10, 2026: Humulin R and Humulin N, BLA018780 and BLA018781, original approval 1982-10-28; rituximab, BLA103705, 1997-11-26; trastuzumab, BLA103792, 1998-09-25; adalimumab, BLA125057, 2002-12-31; ado-trastuzumab emtansine, BLA125427, 2013-02-22; filgrastim-sndz, BLA125553, 2015-03-06; somatropin (Humatrope), BLA019640, 1986-10-16. (Primary -- FDA's own approvals database) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

12. "Muromonab-CD3 (Orthoclone OKT3): the first monoclonal antibody approved for therapeutic use." PubMed, PMID 3557906. Accessed August 10, 2026. (Primary -- indexed primary literature) ↩

13. Monoclonal antibody manufacturing history review, 2024. PMC, PMC11218797. Accessed August 10, 2026: covers hybridoma-era ascites production and the shift to recombinant mammalian-cell manufacturing. (Primary -- peer-reviewed review) ↩

14. National Academy of Sciences. "Theodore T. Puck: Biographical Memoir." nasonline.org. Accessed August 10, 2026 via the Internet Archive, the live page returning 403 to scripted fetchers: confirms Puck's isolation and characterisation of the CHO-K1 line during his single-cell mammalian tissue-culture work. The specific 1955-56 dating carried by secondary accounts is not stated in the memoir and is therefore not asserted here. (Primary -- academy biographical memoir) ↩

15. Lonza. "Our History." lonza.com. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: 1897 founding; 1996 acquisition of Celltech Biologics; three 20,000-litre mammalian bioreactors completed in Portsmouth by 2004 at a cost exceeding $200 million. (Primary -- company history) ↩

16. Samsung Biologics. "Milestones." samsungbiologics.com. Accessed August 10, 2026: "APR. Foundation of Samsung Biologics" under 2011. (Primary -- company history) ↩ ↩²

17. Boehringer Ingelheim BioXcellence corporate history could not be independently verified this pass; the company's own history page returned a bot-detection block rather than content, and no substitute primary source was located. (Not found -- no source) ↩

18. Lasker Award retrospective on nucleoside-modified mRNA, 2021. PMC, PMC8462133. Accessed August 10, 2026: Kariko and Weissman's 2005 finding that replacing uridine with a modified nucleoside suppressed innate immune activation, and the 2008 follow-up showing increased protein yield. (Primary -- peer-reviewed retrospective) ↩

19. The Nobel Assembly at Karolinska Institutet. "The Nobel Prize in Physiology or Medicine 2023." NobelPrize.org, October 2, 2023. Accessed August 10, 2026: awarded jointly to Katalin Kariko and Drew Weissman "for their discoveries concerning nucleoside base modifications that enabled the development of effective mRNA vaccines against COVID-19." (Primary -- the awarding body) ↩ ↩²

20. US Food and Drug Administration. Press announcement approving Kymriah (tisagenlecleucel), August 30, 2017. web.archive.org, archived September 3, 2017. Accessed August 10, 2026: the first gene therapy approved in the US, and the first CAR-T product. (Primary -- agency press announcement, via archive) ↩ ↩²

21. Associated Press. Report on FDA approval of Luxturna (voretigene neparvovec), December 19, 2017. AP News. Accessed August 10, 2026: first gene therapy in the US delivering a corrective gene directly via an AAV vector, for a rare inherited retinal disease. (Secondary -- contemporaneous wire report of the agency action) ↩ ↩²

22. US Department of Commerce and National Institute of Standards and Technology. "U.S. Secretary of Commerce Penny Pritzker Announces Biopharmaceutical Manufacturing Institute Joining Manufacturing USA Network." NIST, December 16, 2016. Accessed August 10, 2026: $70 million award establishing NIIMBL as the eleventh Manufacturing USA institute, the first with an industry-proposed focus area and the first funded directly by the Department of Commerce. (Primary -- federal agency announcement) ↩ ↩²

23. mRNA vaccine platform review. PMC, PMC8502079. Accessed August 10, 2026: in vitro transcription, lipid-nanoparticle formulation, and first US authorisation of the Pfizer-BioNTech vaccine on December 11, 2020 with Moderna's following roughly a week later. (Primary -- peer-reviewed review) ↩ ↩² ↩³

24. Schiel J, et al. "Biomanufacturing readiness levels [BRL] -- A shared vocabulary for biopharmaceutical technology development and commercialization." Biotechnology and Bioengineering 2022;119:3526-3536, doi 10.1002/bit.28227, PMC9826509, PMID 36071569. Accessed August 10, 2026: the nine-level scale across three phases, defined for technology feeding "the production of a licensed biologic medicine." Note on the citation itself: PMC9463082, carried by earlier coverage for this paper, resolves to an unrelated study of prescription-drug misuse and is not the BRL paper. (Primary -- peer-reviewed paper) ↩ ↩²

25. International Society for Pharmaceutical Engineering. "Guide: Biopharmaceutical Process Development and Manufacturing." ISPE. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers. (Secondary -- professional-society guidance, not independently peer-reviewed) ↩

26. Northeast Biomanufacturing Center and Collaborative. "Skill Standards." biomanufacturing.org. Accessed August 10, 2026. (Secondary -- educational-consortium source) ↩

27. EuropaBio. "Biomanufacturing Definition 101." europabio.org, May 2023. Accessed August 10, 2026: the broad bioeconomy definition covering medical therapies, animal feed, fuels, textiles, plastics and cosmetics. (Secondary -- industry-association source) ↩

28. "Manufacturing Cell and Gene Therapies: Challenges in Clinical Translation." PMC, PMC10961620. Accessed August 10, 2026: vector design, cell isolation and expansion, formulation, fill-finish and GMP facility requirements as a distinct manufacturing discipline. (Primary -- peer-reviewed review) ↩

29. AAV upstream-process methods paper. PMC, PMC8418526. Accessed August 10, 2026: reports an empty-to-full capsid ratio around 50:50 in the harvested material, before purification. (Primary -- peer-reviewed methods paper) ↩

30. Enfortumab vedotin (Padcev) prescribing information. DailyMed. Accessed August 10, 2026: the label carries "Initial U.S. Approval: 2019" and describes a Nectin-4-directed antibody-drug conjugate. A direct Drugs@FDA record was not returned under the generic name this pass; the label statement is used instead of a day-precise database date. (Primary -- FDA-approved label) ↩

31. US Food and Drug Administration. BLA review documents for tisagenlecleucel. fda.gov. Accessed August 10, 2026: reviewers describe blinatumomab as "the only available product in a related product class." (Primary -- FDA review memorandum) ↩

32. Genentech. "25th Anniversary of First Product Approval." gene.com. Accessed August 10, 2026: Protropin (somatrem) approved October 18, 1985, the first product Genentech itself manufactured and marketed. FDA's Orphan Drug Product database lists October 17; the one-day discrepancy is recorded rather than resolved. (Primary -- company account, cross-read against the FDA database) ↩

33. Recombinate approval, December 10, 1992, per contemporaneous pharmaceutical trade reporting. Citeline / Pink Sheet archive. Kogenate: current prescribing information. FDA, which states "Initial U.S. Approval: 1993." Accessed August 10, 2026. Pre-1996 BLA documents are not served by FDA's own approvals database and openFDA carries no record, so Kogenate is dated to the year rather than the day; a February 25, 1993 date carried by secondary accounts is not asserted. (Secondary for Recombinate's date; Primary for Kogenate's label year) ↩

34. Contemporaneous pharmaceutical trade reporting of the simultaneous US approvals of Intron A and Roferon-A for hairy cell leukaemia, June 1986. Citeline / Pink Sheet archive, dated June 9, 1986. Accessed August 10, 2026: headline and metadata confirm the simultaneity and the indication; the body text, which carries the day-precise date, is paywalled, so the date is given here to the month. (Secondary -- contemporaneous trade reporting) ↩

35. Gaucher-disease enzyme-replacement therapy history. PMC, PMC3340106. Accessed August 10, 2026: alglucerase (Ceredase) approved 1991 from placenta-derived enzyme; the recombinant successor imiglucerase (Cerezyme) dated to 1995 in this review. A 1994 date appears in label metadata elsewhere; the review's own date is used here and the discrepancy is recorded rather than resolved. (Primary -- peer-reviewed history) ↩

36. Current marketed status and label content for Recombivax HB, Amtagvi (lifileucel) and BeneFIX. DailyMed. Accessed August 10, 2026. Day-precise original approval dates were not independently re-derived beyond label metadata this pass and are marked not established at that precision. (Primary -- FDA label database) ↩ ↩²

37. AAV clinical-trial-design review citing FDA's May 24, 2019 summary basis for regulatory action for Zolgensma. Cold Spring Harbor Perspectives in Medicine. Accessed August 10, 2026. A direct Drugs@FDA database record was not returned for either Zolgensma or Kymriah under their brand names this pass, so both dates rest on FDA-sourced literature rather than a direct database hit. (Secondary -- peer-reviewed review of the primary regulatory record) ↩

38. Novo Nordisk. "Semaglutide." Annual Report 2025. Accessed August 10, 2026: semaglutide described as compound number 217 in the synthesis campaign. (Primary -- company annual report) ↩

39. GLP-1 drug-discovery review. PMC, PMC11441540. Accessed August 10, 2026: the fatty-acid side chain enabling albumin binding and resistance to rapid clearance, and therefore once-weekly dosing. (Primary -- peer-reviewed review) ↩

40. ClinicalTrials.gov records NCT06878560 and NCT07224763 (United Therapeutics, gene-edited pig kidney, both recruiting, enrolment 50 each) and NCT07429838 (eGenesis, EGEN-5784 porcine liver with external perfusion, not yet recruiting, enrolment 20). Queried August 10, 2026. (Primary -- trial registry) ↩

41. Casgevy (exagamglogene autotemcel) prescribing information, setid 7c3e12ad-e2fe-4d3f-a630-ea7364d9e846. DailyMed. Accessed August 10, 2026: "Patients are required to undergo hematopoietic stem cell (HSC) mobilization followed by apheresis to obtain CD34+ cells for CASGEVY manufacturing," and after infusion "the edited CD34+ cells engraft in the bone marrow" -- an ex vivo, not in vivo, edited autologous cell therapy. (Primary -- FDA-approved label) ↩ ↩²

42. USAG-1 and tooth regeneration. Journal of Oral Biosciences 2024, PMID 39389160. Accessed August 10, 2026: confirms the USAG-1 mechanism in mouse models. (Primary -- peer-reviewed paper) ↩

43. Japan Registry of Clinical Trials, record jRCT2051240154. jrct.mhlw.go.jp. Accessed August 10, 2026: TRG035 Phase I, double-blind, placebo-controlled dose-escalation safety trial in 30 healthy adult men, listed complete. (Primary -- national trial registry) ↩

44. ClinicalTrials.gov searches for bioprinted tissue transplant studies and for patient-derived organoid studies. ClinicalTrials.gov. Queried August 10, 2026: zero registered trials for bioprinted tissue transplant; 181 for patient-derived organoids, every one sampled using organoids as a diagnostic or drug-screening tool rather than as the therapeutic product. (Primary -- trial registry) ↩ ↩²

45. Stoddard A, et al. "Bioengineered on-demand outgrowth via synthetic biology triggering." Science Advances 2026;12(16):eadz8362, PMID 41996502; also at PMC13089339. Accessed August 10, 2026: an implanted engineered human liver-tissue construct triggered to grow roughly six-fold in place in mice; in vitro and mouse-model only, no IND filed. (Primary -- peer-reviewed paper) ↩

46. "Two companies can now sell lab-grown chicken in the US." MIT Technology Review, June 21, 2023, and Douglas L. "'A new era': US regulator allows first sales of lab-grown meat." Reuters, June 21, 2023. Accessed August 10, 2026: USDA grants of inspection to Upside Foods and Eat Just's Good Meat, following FDA no-questions letters in November 2022 and March 2023; Singapore approved cultivated chicken first, in 2020; product grown in steel tanks and served initially in restaurants. (Secondary -- contemporaneous press, cross-confirmed across two outlets) ↩

47. Defense Advanced Research Projects Agency. "Battlefield Medicine." darpa.mil. Accessed August 10, 2026: two thrusts, Pharmacy on Demand for small-molecule active ingredients and Bio-MOD for protein therapeutics, building miniaturised continuous-flow platforms. (Primary -- agency programme page) ↩

48. Defense Advanced Research Projects Agency. "Point-of-need pharmaceuticals." darpa.mil, January 24, 2024. Accessed August 10, 2026: EQUIP-A-Pharma building the regulatory-qualification data for multiple finished drug products on a single reprogrammable platform; "$550 million annually" in US hospital costs from drug shortages cited as the civilian motivation. (Primary -- agency announcement) ↩

49. "Biomanufacturing Market -- Global Strategic Business Report," report ID 6094468. Research and Markets. Accessed August 10, 2026: $21.5 billion (2024) to $31.5 billion (2030), 6.6% CAGR. (Secondary -- commercial market research) ↩

50. "Global Biomanufacturing Market Strategic Research Report." Market Research Reports. Accessed August 10, 2026: $28.4 billion to $62.7 billion, 10.40% CAGR. (Secondary -- commercial market research, methodology not independently verified) ↩

51. "BioManufacturing Market Share and Industry Trends." Real Time Data Stats. Accessed August 10, 2026: "$35.2 billion in 2025" to "$70.8 billion by 2033," 9.1% CAGR. (Secondary -- commercial market research, methodology not independently verified) ↩

52. "Next-Generation Biomanufacturing Market by Product Type, Technology, Process Stage -- Global Forecast 2026-2032." GII Research, March 13, 2026. Accessed August 10, 2026: $466.22 million (2025) to $2,624.87 million (2032), 28.00% CAGR. (Secondary -- commercial sub-segment report) ↩

53. "Global Biomanufacturing Platform Market Outlook, In-Depth Analysis and Forecast to 2032." QY Research. Accessed August 10, 2026: $400 million (2025) to $1,727 million (2032), 23.6% CAGR. (Secondary -- commercial sub-segment report) ↩

54. BioPlan Associates. "2026 23rd Annual Report and Survey of Biopharmaceutical Manufacturing Capacity and Production." GII Research, July 14, 2026. Accessed August 10, 2026. Sizes production capacity rather than market revenue, which is why it does not appear as a row in the revenue comparison. (Secondary -- commercial survey report) ↩ ↩²

55. USD/KRW exchange rate. Trading Economics. Accessed August 10, 2026: 1,418.11 on August 10, 2026. The conversion above uses approximately 1,420 as a round current-rate approximation -- a spot translation, not the company's own reporting currency. (Secondary -- financial-data aggregator, for illustrative conversion only) ↩

56. Thermo Fisher Scientific Inc. XBRL company-facts, tag RevenueFromContractWithCustomerExcludingAssessedTax. SEC EDGAR, CIK 0000097745. Retrieved August 10, 2026: FY2020 $32.2 billion, FY2021 $39.2 billion, FY2022 $44.915 billion, FY2023 $42.857 billion, FY2024 $42.879 billion, FY2025 $44.556 billion; the FY2025 figure confirmed against the FY2025 10-K, filed February 26, 2026. Company-wide revenue; XBRL company-facts does not carry segment-level tags. (Primary -- filed XBRL data and the underlying annual report) ↩ ↩²

57. Samsung Biologics. "Samsung Biologics Reports Fourth Quarter and Fiscal Year 2025 Financial Results." Samsung Biologics, January 21, 2026. Accessed August 10, 2026: FY2025 revenue KRW 4,557.0 billion, up 30.3% year over year; operating profit KRW 2,069.2 billion, up 56.6%. (Primary -- company-reported consolidated results) ↩

58. Statistics computed independently for this piece from the raw series above, in Python, rather than taken from any source's own summary figure. Method: growth stated as the mean year-over-year change with a 20,000-resample percentile bootstrap 95% confidence interval, alongside the bookend compound rate and the full range of that compound rate recomputed across every window of five years or longer. The bootstrap is used in preference to a closed-form interval because ten to sixteen annual observations of noisy counts do not satisfy the assumptions a t interval would need. Where an interval crosses zero, that is stated rather than resolved. The intent is Motulsky's: an interval and a sample size, never a bare point estimate carrying more precision than the data holds. Raw pulls (SEC EDGAR XBRL company-facts, USPTO file-wrapper records, ClinicalTrials.gov API v2) and the analysis script are held with this piece's working files. (Primary -- computed from the filed and registry data cited above) ↩ ↩² ↩³

59. Catalent, Inc. Form 10-K for the fiscal year ended June 30, 2024. SEC EDGAR, filed September 6, 2024, accession 0001596783-24-000087. (Primary -- filed annual report) ↩

60. Catalent, Inc. Form 8-K, introductory note: "On December 18, 2024, Novo Holdings A/S ... completed the acquisition of Catalent, Inc." SEC EDGAR, filed December 18, 2024, accession 0001193125-24-280922. (Primary -- filed merger-completion document) ↩

61. Catalent, Inc. Form 15-12G terminating registration. SEC EDGAR, filed December 30, 2024. (Primary -- filed deregistration document) ↩

62. US Food and Drug Administration. "Advanced Manufacturing Technologies Designation Program." FDA, final guidance December 2025, under authority established by the Food and Drug Omnibus Reform Act of 2022. Accessed August 10, 2026: designation triggers expedited review for drug applications manufactured using the designated technology. (Primary -- agency guidance page) ↩ ↩²

63. National Institute for Innovation in Manufacturing Biopharmaceuticals. "NIIMBL Announces Project Call 10.1 to Advance Biopharmaceutical Manufacturing Technology and Workforce Capabilities." NIIMBL, August 3, 2026. Accessed August 10, 2026: a combined $8 million of project activities under Project Call 10.1; concept submissions due September 9, 2026 at 5pm ET; lead technical topic "Intelligent Biomanufacturing Through Artificial Intelligence, Digitalization, and Advanced Process Control." (Primary -- the institute's own announcement) ↩ ↩²

64. US Patent and Trademark Office, patent file-wrapper data, applications with "biomanufacturing" in the invention title; the by-year series recomputed here from the 76 returned application records' own filingDate fields rather than from a summary table. USPTO Patent Public Search; a reader without an API key can reproduce the shape of the search at Google Patents. Queried August 10, 2026. Title-text only, not classified by CPC code -- a narrower proxy than a field-wide patent count. (Primary -- patent-office data, narrowly scoped) ↩

65. Moran L. "Onshoring Is Here. The Hiring Wave Is Still a Decade Out." Genetic Engineering and Biotechnology News, August 4, 2026. Accessed August 10, 2026: over $370 billion committed to domestic US pharmaceutical manufacturing plants since 2025; decade-scale technical hiring ramp. (Secondary -- trade press) ↩ ↩² ↩³

66. Mirasol F, Smith AK. "How AI-Driven Batch Review May Reduce Biomanufacturing Labor Burden." BioPharm International, April 23, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: AI and machine learning handling routine batch-record review under a review-by-exception model. (Secondary -- trade press) ↩ ↩²

67. "The End of Business as Usual: How CDMOs will Evolve Operations in 2026." Contract Pharma, April 1, 2026. Accessed August 10, 2026: "some CDMOs are regionalizing and specializing, others are doing the opposite -- aggressively expanding their service breadth to become integrated, 'one-stop-shop' partners"; a move toward late-stage complex commercial work in cell and gene therapies and antibody-drug conjugates; "Local strategy will become a quality attribute," with the BIOSECURE Act named as signalling "a shift toward geographically constrained sourcing and manufacturing models." A Section 232 tariff framing carried by earlier coverage was not present in the article text this pass and is not repeated. (Secondary -- trade press) ↩ ↩²

68. Anthropic. "Novo Nordisk" customer case study. claude.com. Accessed August 10, 2026: NovoScribe combining Claude Code, Claude models on Amazon Bedrock and MongoDB Atlas with retrieval-augmented generation; clinical study report production "10+ weeks to 10 minutes"; writing time cut by 90%; a stated baseline of "2.3 CSRs per year" by hand. A vendor describing its own customer's use of its own product, not independent verification. (Secondary -- vendor case study) ↩

69. Anthropic. "Bluenote" customer case study. claude.com. Accessed August 10, 2026: three agent types for life-sciences clients including Guardant Health and Eledon Pharmaceuticals; Guardant Health cited as having boosted a specialised workflow's efficiency by 40-50%; broader "10x faster analysis" and "50-75% acceleration" figures self-reported and unaudited. (Secondary -- vendor case study) ↩

70. AgenticBricks. "i3 Pharmaceuticals custom SaaS." agenticbricks.com. Accessed August 10, 2026: a specialty generic pharmaceutical company in Warminster, Pennsylvania, 150,000 square feet under full cGMP oversight; a production system delivered in three months from an empty repository; "Claude Code proposes, engineers verify, tests validate, reviewers approve. No generated code reached production without human review and automated verification." Published by the engineering firm that did the build, about its own client. (Secondary -- vendor-authored case study) ↩

71. Scale AI Labs. "Coding agents for drug discovery." labs.scale.com, June 6, 2026. Accessed August 10, 2026: 66 domain-expert-curated tasks run through the Biomni framework against Claude Code (Opus 4.7), Codex (GPT-5.5) and Gemini CLI (Gemini 3.1 Pro); Claude Code declined 5 tasks on content-policy grounds; "on 43 of the 66 tasks at least one of the three agents produced a correct answer"; the bottleneck identified as high-level planning rather than domain knowledge. Scale's own research blog, benchmarking competing products rather than its own. (Secondary -- independent industry benchmark) ↩

72. Boltz. "Boltz and Claude." boltz.bio. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: a CD3-by-CD19 bispecific designed across a 20,000-call budget reaching minimum binding confidence 0.62 across three interfaces; a smaller model skipping the antibody-numbering step and invalidating all its designs; a hit-to-lead predictor reaching log-scale Pearson correlation 0.38 against 0.23 for the raw model and 0.14 for a fixed non-agentic regressor. A vendor blog post, with specific and reproducible-style methodology. (Secondary -- vendor blog) ↩

73. "Eli Lilly taps AI, digital twin technology to boost manufacturing capacity." Pharma Manufacturing, June 11, 2026. Accessed August 10, 2026: Scot Lindsey, senior vice-president and information officer for manufacturing and quality -- "We don't want any black box. We want to fully understand why our AI solution is recommending what it's recommending"; GLP-1 products kept off the FDA shortage list since late 2024; the October 2025 NVIDIA partnership. (Secondary -- independent trade journalism) ↩

74. "Adaptive agent-oriented control for biomanufacturing systems." Genetic Engineering and Biotechnology News, April 29, 2026. Accessed August 10, 2026: Seyed Soheil Mansouri of the Technical University of Denmark on decentralised "agent hives" of "rule-based, mathematically informed agents"; agentic AI "is not yet fully ready for complete, independent control in biopharmaceutical manufacturing"; "any AI that directly affects medicine quality still needs strong human oversight and full approval"; the shadow-mode adoption path. (Secondary -- independent trade press) ↩

75. WuXi Biologics. PatroLab digital-twin platform press release. wuxibiologics.com, January 12, 2026. Accessed August 10, 2026, retrieved live after the page returned 403 to scripted fetchers: Raman-based process-analytical-technology monitoring of "more than 40 key process performance and product quality attributes" in real time, with data density per batch increased by "nearly 1,000 times" against traditional methods, framed around enabling real-time release testing. (Secondary -- vendor press release) ↩

76. "Top CDMO Companies 2026: Largest Contract Development and Manufacturing Organizations Ranked." IntuitionLabs, August 8, 2026; cross-checked against Slabodkin G, "4 Contract Development and Manufacturing Organizations to Watch in 2026," Pharma Manufacturing, January 8, 2026. Accessed August 10, 2026: consistent top tier of Lonza, Samsung Biologics, WuXi Biologics, Catalent and Thermo Fisher (Patheon). (Secondary -- commercial ranking sources, cross-checked across two independently verified pages; individual methodologies not audited) ↩

77. No source located this pass verified the specific 2025-2026 market positioning of Cytiva, Sartorius, Thermo Fisher's instruments business, or Merck's MilliporeSigma. Flagged as not established rather than asserted from general knowledge. (Not found -- no source) ↩

78. US Office of Information and Regulatory Affairs, Unified Agenda entry RIN 0910-AJ31, "Amendments to 21 CFR Parts 210 and 211; Current Good Manufacturing Practices; Advanced Manufacturing, Distributed, and Point of Care Manufacturing." Reginfo.gov. Accessed August 10, 2026: proposed rule stage; the abstract describes science- and risk-based compliance approaches, flexible batch definitions, real-time monitoring control strategies and lifecycle-based validation, building on Section 3213 of the 2022 Food and Drug Omnibus Reform Act. Not finalised. (Primary -- the regulatory agenda itself) ↩

79. US Food and Drug Administration. "Advanced Manufacturing," Office of Clinical Policy and Programs. FDA. Accessed August 10, 2026: the Advanced Manufacturing Innovation Hub and regulatory-science benchmarks for platform technologies including closed-loop controls; ICH Q13 on continuous manufacturing finalised March 2023. (Primary -- agency page) ↩

80. Executive Order 14293, "Regulatory Relief to Promote Domestic Production of Critical Medicines." whitehouse.gov, May 5, 2025. Accessed August 10, 2026. (Primary -- the order itself) ↩

81. BIOSECURE Act, enacted via the National Defense Authorization Act for Fiscal Year 2026, signed December 18, 2025; analysis via K&L Gates, "BIOSECURE Act: What You Need to Know." K&L Gates, January 20, 2026. Accessed August 10, 2026. (Primary for the statute; Secondary for the law-firm analysis of implementation timing) ↩

82. openFDA drug-shortage database, queried directly at api.fda.gov/drug/shortages.json with search=status:"Current". open.fda.gov. Queried August 10, 2026: 1,179 currently active US drug shortages against 1,651 records unfiltered, with the API's own meta.last_updated reporting August 8, 2026. Spans the whole pharmaceutical supply chain, not biologics specifically; not broken out by product type this pass. (Primary -- FDA data) ↩

83. Regulation (EC) No 1394/2007 of the European Parliament and of the Council on advanced therapy medicinal products, Article 28, read with Article 3(1) of Directive 2001/83/EC. EUR-Lex. Accessed August 10, 2026: the hospital exemption for custom-made advanced therapy medicinal products prepared on a non-routine basis, under individual prescription, for an individual patient, within a single member state. (Primary -- EU legislative text) ↩

84. "EU Pharma Legislation Review Series: Pharmacy and Hospital Exemptions." Inside EU Life Sciences, April 28, 2023. Accessed August 10, 2026: "we are likely to see pharmacies compounding product in a way that increasingly threatens the market for approved innovator products." (Secondary -- specialist legal analysis of primary legislative text) ↩

85. International Bar Association. "The hospital exemption pathway for advanced therapies." ibanet.org, June 2, 2025. Accessed August 10, 2026: fragmented and inconsistent application across member states, with Spain flagged as treating exemption products almost as de facto approved drugs. (Secondary -- independent legal analysis) ↩

86. Contemporaneous reporting of 60 million doses' worth of drug substance discarded following the March 2021 event. The Washington Post, June 11, 2021. Congressional oversight finding that "nearly 400 million doses" were destroyed across Emergent's 2020-2021 quality failures, against a $628 million federal contract. House Committee on Oversight and Accountability, Democrats. Accessed August 10, 2026. A 75 million upper bound carried by earlier coverage is not supported by any source located this pass and is not repeated. (Secondary -- contemporaneous press and congressional oversight) ↩

87. WuXi AppTec's addition to the Department of War's statutory 1260H list, June 8, 2026, per Holland and Knight client analysis; the complaint and the August 7, 2026 preliminary injunction on the docket of WuXi AppTec v. Department of War, No. 1:26-cv-02069, US District Court for the District of Columbia, via CourtListener. Accessed August 10, 2026: the opinion finds the department's case rested on a misreading of the evidence, specifically a stake in a fund reported as a stake in a company and a third party's laboratory work reported as the company's own. Reported here as an unresolved, actively litigated matter rather than as settled in either direction. (Primary for the docket and opinion; Secondary for the law-firm analysis) ↩

88. Single-use equipment lead times of 10-12 months during COVID-driven demand. BioProcess International, May 2021, citing ABEC. And the combined Cytiva and Pall commitment: Pall Corporation press release, July 2021. Accessed August 10, 2026: $1.5 billion over two years -- over $600 million to chromatography resins, $400 million to cell-culture media, $300 million to single-use manufacturing and $200 million to other capacity, across thirteen sites with 2,000 hires. (Secondary -- trade press and a vendor press release) ↩

89. Moderna, Inc. Annual report and quarterly results releases. SEC EDGAR. Accessed August 10, 2026: a strategic resizing initiative launched in the third quarter of 2023 incurring $1.6 billion in charges that year; full-year 2024 cost of sales including $495 million in inventory write-downs and $105 million of unutilised manufacturing capacity; a $238 million non-cash charge disclosed with fourth-quarter 2024 results to terminate a contract-manufacturing agreement. (Primary -- filed financial statements) ↩