2026-09-24

Something interesting happened in structural biology in the last few years. A problem that had occupied researchers for 5 decades, predicting how a protein folds from its amino acid sequence, moved from “grand challenge” to “largely solved” faster than almost anyone expected. AlphaFold’s database now holds structure predictions for over 200 million proteins, effectively covering the catalogued protein universe. The 2024 Nobel Prize in Chemistry went, in part, to David Baker for computational protein design and to Demis Hassabis and John Jumper for protein structure prediction. Baker’s lab has demonstrated designs of novel proteins that do not exist in nature and do not resemble anything evolution produced. This is genuinely remarkable. It is also only half the story, the less difficult half.

That’s because once you have designed your molecule, you still have to make it, at scale, reproducibly—at a cost someone will actually pay. And that second half has not been solved by any algorithm.

The Bottleneck has Moved, It has not Disappeared

There is a pattern worth noticing across industries. When one constraint is relieved, throughput does not increase indefinitely. It increases until it hits the next constraint. Semiconductor design tools got extraordinarily good long before foundry capacity caught up, and the industry spent years discovering that the hard part was not designing the chip. Biology is arriving at a similar juncture. If computational tools let a lab generate a thousand candidate sequences in an afternoon, the question becomes: how many of those can you actually express, purify, and validate? And of those, how many can you manufacture at commercial volume?

The economics here are worth sitting with. The FDA’s Center for Drug Evaluation and Research (CDER) approved 50 novel drugs in 2024, of which 12 were biologics licensed through Biologics License Applications (BLAs). In 2023, the equivalent figure was 55 novel approvals with 17 BLAs.

Meanwhile, India’s biopharmaceutical sector accounted for roughly 62 percent of the country’s biotechnology revenue in 2023, and the Indian biotechnology sector overall has been reported at around USD 137 billion in 2022 with projections toward USD 300 billion by 2030. The demand signal or the discovery is no longer the constraint.

What Manufacturing Actually Costs

Consider what it takes to build biologics capacity: Between 2021 and mid-2024, Biocon Biologics deployed approximately USD 100 million in capital expenditure across 4 sites, and reported that constructing a greenfield biologics facility requires an estimated USD 200 to 500 million. The company operates roughly 130,000 liters of installed mammalian capacity and about 100,000 liters of microbial capacity.

That’s simply what conventional bioreactor infrastructure costs. Stainless steel, clean utilities, HVAC, validation, qualification, and the years of construction and commissioning before a single gram is produced.

Globally, biopharmaceutical firms committed over USD 20 billion to new manufacturing facilities in the 12 months to mid 2024, with many announcements clustering in the USD 1 to 2 billion range and single facilities reaching USD 4 billion. Single-Use Systems (SUSs) and modular construction have shortened build timelines, and India has been noted as a growth market for these approaches.

And, the industry is responding to this growth; but with more of the same architecture, faster.

The ‘Lab’ to ‘Scale’ Conundrum

Ask anyone who has run a process development group about the gap between “it expressed in the lab” and “it manufactures at scale,” post-translational modifications that do not behave the way the small-scale run suggested, aggregation appearing at concentrations one did not test, yields that are excellent per liter and irrelevant per kilogram, Glycosylation patterns that shift with the host system - none of this is a discovery problem. All of it is a manufacturing problem, and manufacturing problems are solved by people who have made things, repeatedly, and been wrong enough times to develop instincts.

However, the fact of the matter is that mammalian cell culture, microbial fermentation, plant-based expression, cell-free systems, all have a real place in the larger scheme of things. Mammalian systems are used for approximately 60 to 70 percent of therapeutic protein production, largely because of their capacity for human-like post-translational modification. E. coli remains the workhorse for simpler proteins, delivering high yields at low cost, and yeast systems offer both eukaryotic processing and rapid growth. These are not competing ideologies. They are tools with different foundational biology, and the mature view is that different molecules need different tools. While the industry has spent a great deal of its innovation budget on one architectural family, it has spent comparatively less on asking whether there are other ways to organise the same biochemistry.

Living Systems as Manufacturing Infrastructure

Firstly, sericulture already exists as industrial infrastructure. India is the second largest silk producer globally and the largest consumer. This means a rearing base, a farmer network, and a supply chain was built for textiles over centuries, that would otherwise take a decade and hundreds of millions of dollars to construct from scratch.

Secondly, silkworms are protein factories that evolution has already optimised for. Silkworm-based expression systems have been described in the literature as suitable for producing complex proteins with mammalian-type modifications, with reported advantages in cost and scalability. The technology has been noted as reaching commercial deployment for veterinary vaccine applications. Swapping the payload rather than inventing the chassis is the actual technical bet.

For 50 years, scaling protein manufacturing has meant scaling steel. A living production system introduces a new paradigm to the trifecta of fidelity, time and capital that de-risks protein production.

The Bottomline

Artificial Intelligence (AI) will keep accelerating discovery, and that acceleration is real and welcome. However, a designed molecule is a hypothesis. A manufactured molecule is a product.

If we want the protein revolution to actually reach patients and farmers, we need to treat manufacturing as an intellectual frontier in its own right, not as the downstream execution of clever science. That means valuing process engineers the way we value computational biologists. It means capital allocation toward architecture, not just capacity. And it means being genuinely curious about biological systems we have not yet tried to industrialise.

India, with its manufacturing base, its cost structure, and its existing biological infrastructure, is unusually well positioned to lead here. Not because our discovery is behind, but because our manufacturing instincts are ahead.