Industries / Health / Pharmaceuticals / Contract drug manufacturing
Biologics
Contract makers grow protein medicines, such as antibodies, in tanks of living cells for drug companies that have no plant of their own or need more room than they have.
5
Milestones achieved, from 1996 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1996
Cell-grown medicines for hire
Lonza, a Swiss chemicals company, bought Celltech Biologics, with plants in Slough, England, and Portsmouth, New Hampshire. Two Lonza scientists wrote in 1997 that the unit was "the world's leading contract manufacturer" of antibodies and other proteins from animal cells. It had over 300 staff and culture tanks of up to 2,000 litres. It offered a drug company the whole job: taking a product "from the laboratory through to full scale manufacturing" and preparing the data that regulators ask for. It also had its own GS System, a set of tools for getting cells to make a chosen protein. Lonza's history says both plants are still hubs of its cell-culture work.
Credit
Celltech Biologics; Lonza
2022
Paying for an empty plant
Novavax hired Fujifilm Diosynth Biotechnologies to make its first COVID-19 vaccine in Texas, under agreements signed in June 2020 and August 2021. On 30 September 2022 Novavax agreed to pay up to USD 185.0 million to cancel that work. In October 2023 it refused the last two payments, USD 68.6 million, arguing that Fujifilm had not tried hard enough to fill the empty space with other work. Fujifilm went to arbitration. In March 2024 the two settled for USD 42.0 million. The figures are from Novavax's own quarterly filing.
LessonReserved plant time becomes a debt when demand falls, and the contract decides who pays for idle space.
Credit
Novavax; Fujifilm Diosynth Biotechnologies
2024
A 2011 newcomer at scale
Samsung Biologics was founded in 2011 and built four plants at Songdo, Incheon, South Korea. In 2024 its sales were KRW 4.5 trillion on a consolidated basis, and its contracts signed that year passed KRW 5 trillion for the first time. By its own figures a fifth plant of 180,000 litres, due to open in April 2025, brings its capacity to 784,000 litres, which it calls the largest of any maker, and 17 of the world's 20 largest drug companies are customers.
Lessona contract maker can reach the front rank in about a decade by building very large plants ahead of orders.
Credit
Samsung Biologics
2024
Roche sells a plant
Lonza bought Genentech's biologics plant in Vacaville, California, from Roche for USD 1.2 billion. The deal was announced in March 2024 and completed by October 2024. The site has about 330,000 litres of bioreactor capacity, and about 750 Genentech staff moved to Lonza. Lonza keeps making Roche's products there at committed volumes that shrink as other customers move in. It plans to spend a further CHF 500 million on upgrades.
LessonA drug company can sell a large plant and keep its own supply by contract.
Credit
Lonza; Roche and Genentech
2025
Many bags instead of tanks
A single-use bioreactor is a plastic bag in a support frame, used once and thrown away, so no steel tank has to be cleaned and checked between batches. In 2021 WuXi Biologics started a plant in Wuxi, China, with nine 4,000-litre bags, 36,000 litres in all. WuXi called it the world's largest GMP plant built on single-use bioreactors (GMP, good manufacturing practice, is the set of rules regulators enforce in medicine plants). In 2025 the European Medicines Agency (EMA) approved WuXi's plant in Dundalk, Ireland, as a commercial maker of a client's medicine. WuXi says its qualification runs there included a 16,000-litre batch made by combining four 4,000-litre bags.
Credit
WuXi Biologics; EMA
Watch for
Copying a plant exactly
Fujifilm Biotechnologies, the contract-making arm of Fujifilm, opened a USD 3.2 billion antibody plant in Holly Springs, North Carolina, on 24 September 2025. It has eight 20,000-litre bioreactors, with eight more planned by 2028. Fujifilm says the site is a near-copy of its plant in Hillerød, Denmark, that copying cut design time by about 70%, and that it can move production between the two sites if trade problems such as tariffs require it. Regeneron (10 years, more than USD 3 billion) and Johnson & Johnson (USD 2 billion over 10 years) have signed on.
Achieved whena regulator approves Holly Springs as a maker of a marketed medicine that is also made in Hillerød.
Claimed by
Fujifilm Biotechnologies, September 2025
Larger suites for linked antibodies
An antibody-drug conjugate (ADC) is an antibody chemically joined to a cell-killing drug, so the drug goes where the antibody binds. Lonza says it has made more than 1,000 batches of such linked products for over 70 programmes since 2006. In October 2026 it said a plant under construction at Visp, Switzerland, will have three multi-product suites working at up to 1,200 litres from 2028, and that existing suites will gain 500-litre single-use equipment in 2027.
Achieved whenthe new suites make commercial batches of an approved ADC, confirmed by a regulator's approval of the site.
Claimed by
Lonza, October 2026
Open
01
Who makes each antibody
A public table of every antibody medicine authorised in the European Union. Each row names a plant listed as a maker of the medicine's active substance, says whether that plant belongs to the drug's owner or to a contract maker, and quotes the source line.
The briefCredit · unclaimed02
A shared test for schedules
An open library of scheduling problems for a plant that grows several clients' medicines in shared tanks. Each problem gives the products, batch times, cleaning time between products, storage limits, shelf lives and delivery dates in one file format. A checker tests any proposed schedule against every rule and scores it. The problems are copied from published case studies, so no company's plant data are needed.
The briefCredit · unclaimed
Open 01 · the brief
Who makes each antibody
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of every antibody medicine authorised in the European Union. Each row names a plant listed as a maker of the medicine's active substance, says whether that plant belongs to the drug's owner or to a contract maker, and quotes the source line.
Why it is still open
The plant names are public but scattered, one PDF per medicine. Keytruda's EU product information, for example, names AstraZeneca's plant in Frederick, Maryland, and Boehringer Ingelheim plants in Biberach, Germany, and Fremont, California, alongside MSD's own plant in Oss, the Netherlands. The closest study, by B. Vezér and colleagues in 2016, counted 404 approved manufacturing changes across 29 antibodies from EMA reports, but published no table of plants. Paid market databases track plant capacity, not which plant makes which medicine.
Built with what exists
The EMA's product-information PDFs; pdftotext; Python; a language model such as Claude to pull plant names and addresses, with a person checking each one. The note "Seven more flagged" covers Claude reviewing drug-making records.
First step
Pull the plant names from 20 antibody product-information PDFs and publish the table with its rules for telling owner plants from contract plants.
Done looks like
Every monoclonal antibody authorised in the EU listed with at least one plant, a quoted source on every row, and a second person's check of 50 random rows finding at least 95% of them correct.
Open 02 · the brief
A shared test for schedules
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open library of scheduling problems for a plant that grows several clients' medicines in shared tanks. Each problem gives the products, batch times, cleaning time between products, storage limits, shelf lives and delivery dates in one file format. A checker tests any proposed schedule against every rule and scores it. The problems are copied from published case studies, so no company's plant data are needed.
Why it is still open
Each paper on planning such plants tests its model on its own examples or on a few printed in earlier papers. In 2024 Vaibhav Kumar and Munawar Shaik found that schedules from earlier published models broke storage and shelf-life limits once those limits were checked over time, and their corrected model gave up to 36% more profit on four examples from the literature. A 2026 study of a tablet plant by Ioannis Balatsos and colleagues used three instances from one site. Kim van den Houten and colleagues published an open simulator of one biomanufacturing site in 2023. I found no downloadable library with a shared checker.
Built with what exists
Python; open-source solvers such as Google OR-Tools and HiGHS; the SimPyManufacturing simulator to test schedules when batches fail or run late; an AI agent that writes and runs its own solver code, entered and scored like any other method.
First step
Copy one of the literature examples that Kumar and Shaik used into the file format and write the checks for storage and shelf life.
Done looks like
At least 20 problems copied from at least five papers, each paper's reported result reproduced or the gap explained, the checker published, and at least three methods, one of them an AI agent, scored on every problem.