Modality Commoditization and India's Right to Win

The drawdown that wasn't a mistake

Through the first half of 2026, AbCellera Biologics traded at a market capitalisation hovering between roughly one billion and one-and-a-half billion dollars. At its December 2020 IPO-era peak it was worth $15.66 billion. The intervening five years were not a story about execution. The execution was competent. The science held. The balance sheet stayed strong enough that the company kept advancing its own clinical pipeline — ABCL635 into Phase 2, ABCL575 into Phase 1 — rather than surviving as a pure contract-research shop. The drawdown of roughly ninety per cent from peak was not a failure of the company. It was the market repricing the category.

This is the uncomfortable thing about AbCellera, and the reason it is the right place to begin. Nothing went wrong. A company built one of the best antibody-discovery engines in the world, took it public at the high-water mark of a once-in-a-century demand shock, and then watched the demand shock recede and the valuation settle back to the structural level that an antibody-discovery services business is worth. The $15.66 billion was the anomaly. The billion-and-change is the truth. The peak was the breach; the floor is the rule.

Every wave of biotech innovation since recombinant DNA has ended at roughly the same place. The first cohort captures the imagination and the option value. The second cohort competes on architecture. The third cohort builds the tooling, and by the time the broader market understands a modality well enough to underwrite it confidently, the modality has commoditised. By the time the regulator has metabolised it, the next modality has arrived. The intervals between waves are getting shorter. The capital deployed into each wave is getting larger. And the lessons the last wave's investors learned have, to a striking degree, failed to transfer.

This is the argument Elliot Hershberg made in On Modality Commoditization in March 2025. It is a clean argument and the receipts are unambiguous. What this essay extends — from a vantage Hershberg's piece gestures at but does not fully occupy — is what the commoditisation curve means once you stop pricing it inside the capital structure that valued AbCellera at fifteen billion dollars in the first place. Hershberg's essay ends on geography: the last chapter of commoditisation, in his telling, is geographic arbitrage, and the protagonist of that chapter is China. He is right about the chapter. He is looking at the wrong country for the part of the story that has not yet been written.

From an Indian vantage, almost everything Hershberg correctly identifies as a structural endpoint is instead a structural opening. The ceiling that closes the Western fund is the floor that opens the Indian one. The same exit value that returns a fraction of a Western fund returns a multiple of an Indian one. The mathematics is not opinion; it is geometry, and the geometry inverts when the denominator changes. That inversion is the entire essay.

What a modality actually is

"Modality" entered the biotech lexicon as a portfolio-engineering term, and it is worth slowing down on it, because the word is doing more work than it appears to. A large pharmaceutical company structures its discovery effort across modalities: small molecules; biologics, meaning recombinant proteins and monoclonal antibodies; cell therapies; gene therapies; messenger RNA; oligonucleotides and other nucleic-acid medicines; antibody-drug conjugates; bispecific and multispecific antibodies; and now AI-designed novel chemistry. The word lets a business-development executive describe a transaction as a cell-therapy modality acquisition without referring to a single fact about the underlying biology.

That sounds like bureaucratic shorthand. It is in fact load-bearing. The modality designation determines manufacturing class, regulatory pathway, clinical-trial design, intellectual-property strategy, sales-force specialisation, and — most consequentially for the argument here — the financial-architecture template that a deal inherits before the lawyers arrive. A 2024 cell-therapy deal is structured differently from a 2024 small-molecule deal not because the cancers are different but because thirty years of precedent shaped the contract terms, the milestone ladders, the royalty stacks, and the comparable-company tables that the bankers reach for. The modality carries its own financial physics.

Modalities also age, and they age on a schedule you can read off the historical record. Small molecules dominated pharmaceutical value creation from the 1940s onward and still account for the majority of prescriptions written. Recombinant-DNA biologics arrived with Genentech's founding in 1976 and reached commercial saturation by the late 1990s. Antibody-discovery platforms — the services layer that compressed the time from target nomination to a developable clinical candidate from years to months — were the cohort that matured between roughly 2007 and 2020. Cell and gene therapies arrived commercially in 2017 with Kymriah and Yescarta, the first two CAR-T approvals. Messenger RNA arrived at civilisational scale with the COVID-19 vaccines in 2020 and 2021. And AI-designed drug candidates — Insilico's rentosertib, Recursion's clinical pipeline, the molecules emerging from Isomorphic's partnership stack — are arriving as legible, peer-reviewed commercial assets between 2024 and 2026.

Each of these modalities has had its own peak of commercial value. Each has commoditised, or has begun to. The intervals between the waves have compressed. The capital deployed into each successive wave has grown. And — this is the part that should worry any investor who learned their craft in the prior cycle — the lessons do not carry forward cleanly, because the thing that changes between waves is not just the biology but the cost of iteration, and iteration cost is the variable that quietly governs everything else.

The order-of-magnitude law

Hershberg's central observation is best stated as a law, because that is how it behaves. Across each generation of drug-discovery business, the magnitude of the businesses falls by roughly one order of magnitude. The first movers grew into hundred-billion-dollar companies. The leading service providers that followed became ten-billion-dollar companies. The newest entrants in the discovery market are billion-dollar companies. Same function, one-tenth the value, each generation.

The receipts line up. Genentech, founded in 1976, was the canonical first-mover; when Roche completed its acquisition of the shares it did not already own in March 2009, it paid roughly $47 billion for that remaining stake, against a total enterprise value that had been well north of a hundred billion at points earlier in the decade. The services generation that followed — Adimab the clearest exemplar, with secondary-market valuations reported in the ten-billion-dollar range — captured roughly a tenth of that. And the third cohort, the publicly traded antibody-discovery platforms, settles into the one-to-three-billion band: AbCellera at roughly a billion through 2026, FairJourney Biologics valued near nine hundred million in 2024, OmniAb at a few hundred million. Three generations of essentially the same business — find a good antibody faster than the client could — and a tenfold value compression at each step down the ladder.

The pattern is older than antibodies, and the shape repeats regardless of the molecule. The first commercial player captures the imagination and the option value because nobody can yet price the thing it does. The second captures the architecture, because the thing is now legible enough to systematise. The third captures the tooling, because the architecture is now standard enough to sell as a service. The fourth captures very little, because by then the tooling is open-source and the problem is, in the relevant sense, solved. What flattens the curve is not that the science gets easier in some absolute sense. It is that each generation's tooling lets the next generation rebuild the same capability faster and cheaper, until the capability itself is no longer a moat but a commodity input that anyone can rent by the month.

This is the lesson that matters, and it is not "biology is hard." It is the opposite. Biology, considered as a software-and-architecture problem rather than a mystery, is being solved faster every generation, and the value-capture window for any specific solution is collapsing toward zero. A company that needed fifteen years to build a defensible antibody-discovery moat in 2005 would have something closer to a six-month head start if it tried to build the equivalent in 2027, because the tools it would assemble are now sitting on the shelf. Hershberg named this the modality commoditisation curve, and he was correct about the curve. The disagreement is not about whether the curve exists. It is about what the curve means once you change the capital structure standing underneath it.

The compression of the intervals

The compression of the gaps between waves is the second-order fact, and it is the one that turns the curve from an academic observation into an investment clock.

From recombinant biologics in 1976 to the first mature antibody-services platforms was roughly three decades. From those first services platforms to the second cohort — AbCellera spun out of the University of British Columbia in 2012 — was a handful of years. From the services-platform peak to messenger RNA being a globally deployable modality, manufactured at the scale of billions of doses, was less than a decade, and the final stretch of it happened in about eleven months under pandemic pressure. From mRNA scale to the first peer-reviewed Phase 2a readout of a drug whose target was discovered and whose molecule was designed by generative AI — Insilico's rentosertib, published in Nature Medicine in June 2025 — was four years.

The clearest single illustration of the compression is what happened to protein-structure prediction. DeepMind published AlphaFold 3 in May 2024, initially behind a restricted-access interface, with the full code following later in the year. Open-weight and open-source competitors and reproductions — Chai-1, Boltz-1, and others — appeared within months, not years. A capability that would have been a company-defining moat a decade earlier was a commodity, available to any competent lab, before its originators had finished deciding how to license it. The moat evaporated faster than the legal department could build the toll booth.

The compression has a consequence that is easy to state and hard to internalise: the half-life of a modality's defensibility is now shorter than the fund life of the vehicle that backs it. A traditional venture fund deploys over three to four years and harvests over seven to ten. If the defensibility of the underlying modality decays faster than that — if the moat is gone in three years and the fund needs the moat intact in year eight to justify the exit — then the fund is, by construction, betting on a structure that the curve has already condemned. This is not a risk that diversification fixes, because it is not idiosyncratic to any one company. It is the category decaying underneath the whole portfolio at once.

Why biology resisted the word "platform" — and what changed

For most of biotech's first three decades, the word "platform" was treated with active suspicion by serious investors, and the suspicion was earned. The argument against platforms was structural. Each drug had to be discovered, manufactured, tested, and approved against its own molecular target, in its own clinical population, under its own regulatory pathway, against its own failure modes. The capital intensity per asset was enormous and the failure rate was punishing. A company that called itself a platform was betting that its platform would generate enough asset-level wins to amortise the platform's overhead — and the empirical record, until very recently, was that "platform" companies tended to deliver exactly one asset of real value and then become operationally indistinguishable from a single-asset company that happened to have an expensive origin story.

This was not paranoia. Through the 2015-to-2018 window, the platform companies that delivered shareholder returns delivered them through single-asset wins, and the platform narrative was, in the working assumption of most biotech investors, a story a founder told until they had a clinical asset, after which they quietly stopped telling it. Moderna's pre-pandemic valuation was, on most honest analyses, almost entirely a bet on the platform's ability to deliver one specific product — a better flu vaccine — and the COVID-19 vaccine collapsed that uncertainty in about ninety days and revealed, all at once, what the platform had actually been worth the whole time.

What changed after 2020 was not the underlying biology. What changed was the cost and the speed of the iteration loop, and that single change rewired the economics of the word.

Consider the three companies that have made the change most legible. Recursion runs its discovery engine at a scale the industry had never seen: up to 2.2 million experiments per week across its automated wet labs, feeding a platform — Recursion OS — that the company says generates more than a hundred million candidate molecules a year, backed by one of the largest supercomputers in the pharmaceutical industry, built with NVIDIA. Insilico took a target its own algorithms nominated — TNIK, a kinase not previously prosecuted in idiopathic pulmonary fibrosis — and carried an AI-designed molecule against it from discovery to a developable preclinical candidate on a timeline and a budget that have no analogue in traditional medicinal chemistry, then into a randomised, placebo-controlled Phase 2a trial whose top dose improved lung function by 98.4 millilitres of forced vital capacity against a 20-millilitre decline on placebo. Isomorphic, the Alphabet drug-discovery company built on the AlphaFold lineage, signed strategic partnerships with Eli Lilly and Novartis whose combined headline value approaches three billion dollars — and signed them before its first molecule had entered a clinical trial, on the strength of the inference engine alone.

These are platforms in the software sense, which is the sense that finally makes the word mean something. The architecture itself generates candidate assets faster than the assets can fail. That is the property the older biotech platforms never had, and it is the property that changes who should own them, on what time horizon, and at what price.

The one-to-three-billion-dollar ceiling

So the empirical answer, in the US public market, to the question "what is a services-shaped discovery platform worth?" is now reasonably well established. It is a one-to-three-billion-dollar market-capitalisation band, and that band behaves as a structural endpoint rather than a way station. AbCellera's settling near a billion through 2026 is not the market punishing AbCellera. It is the market pricing the modality. The pandemic-era peak above fifteen billion was a demand-shock breach of the ceiling, and when the demand normalised, the ceiling reasserted itself with the indifference of a physical constant.

Now do the arithmetic that makes this a verdict rather than a curiosity, because the arithmetic is the whole point.

Take a US biotech venture fund of three hundred million dollars. Assume it holds six per cent of a company at exit — a reasonable post-dilution ownership for an early-stage investor who has been diluted through several subsequent rounds. A two-billion-dollar exit, at six per cent, returns one hundred and twenty million dollars to the fund. Against three hundred million dollars of committed capital, that is a 0.4x on that position — and a single 0.4x outcome, even on a clean exit with no write-down, is a contributor to a loss, not a fund-returner. To return a three-hundred-million-dollar fund at the multiple its limited partners were promised, that fund needs exits in the ten-billion-dollar class and above. Applied to a three-hundred-million-dollar US fund, the one-to-three-billion ceiling is not a disappointment. It is a death sentence written in the language of fund mathematics, and no amount of scientific excellence at the company level can repeal it.

This is the framework Hershberg built, generalised. The same ceiling that has settled over antibody-discovery platforms will settle over mRNA platforms once the pandemic premium has fully bled out. Cell-therapy platforms have already begun to meet a version of it through the manufacturing-cost wall — the very wall that the Carvykti generation of autologous therapies ran into, and the wall that in-vivo CAR-T is now trying to demolish. AI-designed-drug platforms will meet a more porous version of the ceiling, because their iteration loops are genuinely more durable than a services platform's, but they will meet it.

What the framework does not address — because, priced inside a Western fund, it does not need to — is what the identical ceiling means inside a different capital structure. And that is the question on which everything turns.

The durable archetype, and the case against it

Before the inversion, one necessary detour, because not every platform is condemned to the ceiling, and a serious essay has to say which ones escape and on what evidence.

The most useful taxonomy of platform businesses is older than the AI cohort and comes from Steven Holtzman, Millennium's former chief business officer — Elliot Hershberg and Patrick Malone revived and extended it in On Biotech Platform Strategy in 2023. Holtzman's cut is the one that matters: the deep divide is between modality platforms, whose core technology is a way of making a class of drug — antibodies, an RNA chemistry, a cell-therapy process — and insight platforms, whose product is not a molecule at all but the generation of new biological understanding: targets, pathways, mechanisms. Millennium did this with genomics; it did not own a modality, it owned a way of finding out what to make. The modern reading adds two operating postures on top — the services platform that only ever sells its capability to others, and the vertical integrator that turns the platform inward to build its own pipeline — but the load-bearing distinction is Holtzman's. Modality platforms face a commoditisation pressure that scales with how copyable their one modality is. Insight platforms, in principle, do not — and the reason is worth stating precisely, because it is also the reason the claim is dangerous.

The distinguishing property of an insight platform is that the asset it owns is path-dependent on its own accumulated data. Recursion's millions of weekly experiments are not valuable as experiments; they are valuable because each one is an embedding that conditions the next inference, and a competitor starting clean cannot re-derive the quality of that inference without replaying the entire history that produced it. Insilico's generative chemistry has the same property: the generator's quality is a function of cumulative learning that cannot be reconstructed from a cold start. Call the modern, compute-scaled, wet-lab-in-the-loop version of Holtzman's insight platform the recursive discovery factory. In the available record it is the only platform shape that has so far resisted the commoditisation pressure that flattened the services cohort, because the thing it sells is the one thing on the curve that does not standardise — the proprietary, path-dependent data loop. (The caveat Holtzman's own framework supplies: insight platforms historically struggle to capture the value they create, because pharma will not pay much for pre-clinical information and a pathway cannot be patented the way a molecule can — which is exactly why these companies are forced to integrate forward into their own pipelines to monetise the loop.)

That is the bull case, and it is a real one. Here is the bear case, which the bull case usually omits and which an honest essay must put on the same page.

The recursive discovery factory has not yet produced a commercial drug. Recursion has spent more than a billion dollars over more than a decade and, as of this writing, its pipeline's most advanced clinical validation is an early signal — REC-4881 in familial adenomatous polyposis, with a majority of patients showing polyp-burden reduction — not an approval, while its 2025 accounts show roughly seventy-five million dollars of revenue against a net loss of more than six hundred million, and its stock has fallen by more than half from recent highs. NVIDIA, which had been a totemic investor, exited its position. The broader cohort offers a graveyard alongside the flywheel: BenevolentAI lost roughly three-quarters of its market value between 2022 and 2024; Exscientia, which had put the first AI-designed candidate into Phase 1 as far back as 2020, terminated that lead programme before its Phase 2 readout and was absorbed by Recursion at a steep discount to its former valuation. The skeptical case against the entire archetype is simple and not yet refuted: that mastering protein structure, or cellular imaging, or generative chemistry, does not transfer to the genuinely hard problem, which is clinical efficacy in a human body, and that the data flywheel spins beautifully right up to the clinic and then stalls.

So the durability of the recursive discovery factory is not a fact. It is a hypothesis with a deadline. If two or more of the leading recursive discovery factories — Recursion, Insilico, Isomorphic, Genesis, Iambic, Generate, Xaira — post a clean, positive, mid-stage clinical readout on a self-originated programme by the end of 2027, the archetype's durability is on the record and the ceiling does not apply to it. If none does, the archetype thesis collapses back to a smaller and much less interesting claim about supervised pre-training accelerating early discovery, and the ceiling reasserts at a lower band than even the services platforms occupy, because the burn was higher. I hold the bull case, but I hold it as a position with a stop-loss, not as a faith.

For the Indian operator, this is the one archetype where the honest answer is "not yet, but deliberately." Recursive discovery factories require compute, frontier talent density, and a wet-lab data-loop at a scale India does not possess today. That gap is real. It is not, however, fundamental — and the clearest evidence that it is closeable is that China closed a version of it inside fifteen years, starting from a weaker base than India holds now, by sequencing the build: manufacturing and services first, then clinical-trial scale, then talent repatriation, then originated science, then the data-and-compute layer on top. There is no contradiction in saying India cannot build a recursive discovery factory on a single fund cycle and prescribing that India start building toward one. The two statements describe the same plan on different clocks. The move is not to clone Recursion next year; it is to back one or two credible Indian insight platforms now, capitalise them with the one input a non-frontier ecosystem supplies more cheaply than a frontier one — patience, which is just the cost of carrying a company, and the cost of carrying a company is precisely what India compresses — and let the compute-and-talent gap close on the same fifteen-year arc China just walked, instead of pretending it is already closed.

The inversion

Here is the move the whole essay has been walking toward, and it is almost embarrassingly simple once the arithmetic is laid side by side.

Take the same two-billion-dollar exit that was a death sentence for the three-hundred-million-dollar US fund, and hold everything constant except the size of the fund. Six per cent of two billion dollars is one hundred and twenty million. Against three hundred million dollars of committed capital that is a 0.4x; against fifty million dollars of committed capital it is a 2.4x. Identical company, identical exit, identical ownership — and the same position is a write-down for one fund and a fund-returner for the other. Nothing changed but the denominator. That, on its own, is the whole argument, and I have deliberately kept ownership pinned at the diluted six per cent to make it unarguable: the inversion does not require the India fund to hold more of the company, only to be smaller.

In practice the India fund should hold more than six per cent, because a capital-efficient company that exits in the mid-band reaches its exit through fewer dilutive rounds than a company grinding toward a ten-billion-dollar public outcome — origination ownership survives better when the road to the exit is shorter. But that is an argument I am not going to make properly here, because doing it honestly requires modelling dilution paths, ownership decay through each round, loss ratios across the funnel, and the portfolio-level power law that turns a single-position multiple into a fund return. That is its own essay, and I will write it as one — the venture math of biotech, as a dedicated follow-up. For this piece the conservative, ownership-held-constant version is enough, and it is the version I want on the record, because it cannot be dismissed as a spreadsheet flattered by an optimistic cap-table assumption.

Same exit value. Inverse outcome. The science did not change. The company did not change. The molecule, the trial, the approval, the buyer, even the ownership stake — all identical. The only thing that changed is the denominator: the size of the fund, which sets what a given exit is worth to the people who funded it. The one-to-three-billion ceiling that closes the US fund is the desired endpoint of the India fund. The decision rule inverts not because anyone got smarter but because the geometry inverted.

This is why importing the Western playbook into India is worse than useless. The Western playbook is optimised to chase the ten-billion-dollar outcome, because the Western fund mathematics requires it, and chasing the ten-billion-dollar outcome means concentrating capital, taking large ownership in few companies, and underwriting the long, expensive, winner-take-most path to a category-defining public company. Run that playbook on an Indian cost base and you have simply built a smaller, worse version of an American fund. The playbook that transfers is the one the American fund mathematics cannot support but the Indian fund mathematics requires: a wider portfolio funnel at a far lower per-company cost, entering at origination, targeting the two-hundred-to-seven-hundred-million-dollar strategic-exit band — the band where global pharma actually buys, routinely, without a competitive auction — and closing the fund-returner mathematics on that band rather than betting the fund on a tail outcome that the cost structure was never built to reach.

The Hershberg ceiling, in other words, is not a constraint on the Indian architecture. It is an input to it.

India's receipts

A thesis this neat invites the obvious objection: that it is a spreadsheet fantasy, a geometry trick with no companies behind it. So here are the companies. Each on its own is suggestive. Stacked, they describe an environment the commoditisation framework, priced for the West, does not predict.

In April 2026, Sun Pharmaceutical Industries signed a definitive agreement to acquire Organon — the women's-health and biosimilars company that Merck had spun out in 2021 — in an all-cash transaction valued at $11.75 billion in enterprise value, $14 a share, expected to close in early 2027. It is the largest acquisition ever made by an Indian pharmaceutical company, and it vaults Sun toward the top tier of global women's health and biosimilars and into the ranks of the world's twenty-five largest pharma companies by revenue. (Worth correcting a tempting misreading: this was a full acquisition of Organon, not a carve-out of a women's-health portfolio, and it was announced in 2026, not 2025. The detail matters, because the thing being demonstrated is not frontier discovery — it is that an India-domiciled company can now be the lead party in a deal of this size, financed and operated from an Indian cost base. The capability on display is capital architecture, not chemistry.)

Eighteen months earlier, and from the opposite end of the value chain, ImmunoACT had begun to demonstrate the cost story directly. NexCAR19 — talicabtagene autoleucel, India's first indigenous CAR-T therapy, developed by ImmunoACT as a venture out of IIT Bombay and Tata Memorial Centre and founded by the immunologist Rahul Purwar — was approved by India's CDSCO in October 2023 and launched commercially in April 2024. By mid-2025 it had been administered to more than 350 patients across some seventy to eighty hospitals. It launched at about ₹42 lakh per dose and has since fallen toward ₹30 lakh — roughly thirty-six thousand US dollars — against Western CAR-T list prices in the three-hundred-and-seventy-thousand to five-hundred-and-thirty-thousand-dollar range. That is close to a tenfold compression, and in the case of the most expensive comparators, more. The compression was not labour arbitrage. It came from a different cost stack entirely: autologous manufacturing organised close to the bedside, a humanised CAR construct that the company associates with a favourable safety profile, an academic research foundation that did not have to be paid for out of venture capital, and a CDSCO regulatory pathway that priced the modality on different assumptions than the FDA pathway does. And — the detail that should make an investor sit up — ImmunoACT reported roughly ₹62 crore of revenue and about ₹12 crore of profit before tax in its first full year, which is to say it was profitable, a sentence one almost never gets to write about a first-year cell-therapy company anywhere in the world.

The third receipt complicates the story in a way that strengthens it, which is why I am including rather than hiding it. In June 2025, Insilico published in Nature Medicine the first peer-reviewed Phase 2a proof of concept for a drug whose target was discovered and whose molecule was designed by generative AI: rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis. The cost structure that produced that result has no analogue in Cambridge or Boston. But the trial ran across twenty-one sites in China, not India — Insilico is, in the relevant sense, a China-cost-structure company, not an India one. I include it precisely because it is the honest boundary of the thesis. The capital-efficiency advantage that the argument rests on is real, and China is currently expressing it at a scale and a level of clinical maturity that India is not. The Indian thesis is not that India already leads. It is that the structural conditions that let China express this advantage are reproducible in India, and that the window to build the capital architecture for it is open now and will not stay open.

The exit pattern is real. It just isn't Indian yet.

The honest objection to everything above is that the strategic-exit band the fund mathematics depends on — global pharma routinely buying originated assets in the two-hundred-million-to-low-billions range — has no Indian precedent. There is no example, yet, of an Indian-originated drug asset being acquired or out-licensed into that band. The fund math rests on a mechanism India has not demonstrated.

The mechanism has been demonstrated, comprehensively, next door. China's cross-border out-licensing went from roughly $51.9 billion in 2024 to about $135.7 billion in 2025 — nearly a tripling in a single year — and by 2025 something like thirty-eight to forty per cent of large-pharma licensing deals originated from Chinese partners, accounting for roughly a third of all upfront payments made across the industry. These are not theoretical structures. Pfizer paid 3SBio $1.25 billion upfront, took an equity stake, and layered on as much as $4.8 billion in milestones for the ex-China rights to a PD-1/VEGF bispecific. GSK committed $500 million upfront to Jiangsu Hengrui for a respiratory candidate plus options on eleven further programmes, a package reaching roughly $12 billion in milestones. AstraZeneca's obesity licensing deal with CSPC could reach the high-teens of billions if everything pays out. The dominant deal structure — ex-domestic rights with an option back into the home market — is precisely the shape an Indian originator would use, and it is now the most active sourcing channel in global pharma.

This is the proof that a non-Western ecosystem can originate science cheaply and exit it, repeatedly, into exactly the band the Indian fund mathematics requires. It is also a clock. Chinese assets historically carried upfronts sixty to seventy per cent below Western comparables, with total deal sizes forty to fifty per cent smaller — that discount was the entire opportunity, and it is being competed away in plain view: the average China-sourced upfront has climbed from about $52 million in 2022 to roughly $172 million in early 2026. The arbitrage that built China's out-licensing engine is closing as the buyers learn to price it. India's window is the same window, one cost-curve behind, and it is narrowing for the same reason.

Where does that leave the Sun–Organon transaction, which I cited above? Correctly understood, it is adjacent evidence, not a direct receipt. It proves an Indian company can now be the buyer in an eleven-figure global deal — that Indian capital operates at that scale and with that sophistication. It does not prove the thing the fund math needs, which is Indian-originated assets being bought in the mid-band by others. China supplies that proof; India supplies, so far, only the buyer-side half. Closing the originator-side half is the entire bet, and the absence of an Indian precedent is not a refutation of the thesis — it is a precise statement of what the thesis is wagering on, with a working template sitting one country to the north-east.

Borrowing Flagship's discipline, not its cost base

If the inversion is the strategy, the operating discipline already exists in a fully industrialised form. It just exists at the wrong cost base, in the wrong country, optimised for the wrong fund mathematics — and it can be ported.

Flagship Pioneering spent two decades turning venture creation from an art into a manufacturing process. The funnel, in its publicly described shape, runs roughly like this: scores of exploration starts a year, of which a fraction become protocompanies, of which a fraction become true new companies, of which a fraction raise meaningful external capital, of which a fraction file an investigational new drug application, of which the occasional one reaches a public-market event. The funnel is wide at the top and ruthless at every gate, and the unit economics at a US cost base run to tens of millions of dollars per company carried through the middle of the funnel. The financial logic underneath it is the blockbuster logic of all venture creation: out of any cohort, a single category-defining outcome carries the entire vehicle. For Flagship, that outcome was Moderna, whose position is reported to have accounted for more of one fund's gross return than the rest of the portfolio combined.

Now port the funnel to the Indian cost stack and hold the discipline constant. The front of the funnel is where the capital intensity lives — the protocompany and early new-company stages, where senior scientific salaries, lab space, and reagent costs dominate, and where an Indian operator's costs run a meaningful multiple lower than a Boston operator's. A protocompany that costs a Boston engine a six-figure sum to run for a season costs an Indian engine a fraction of that. A new company that consumes tens of millions of dollars to reach a credible Series A in Cambridge can reach the equivalent milestone in India for a small fraction of the figure, because the most expensive input — senior scientist time — compresses several-fold. The capital intensity at the front of the funnel does not shave by twenty per cent. It collapses by something closer to an order of magnitude. That single number is the entire opportunity, because it means the same disciplined funnel that requires a multi-hundred-million-dollar fund to run in the United States can be run on a fifty-million-dollar fund in India — and a fifty-million-dollar fund, as the inversion showed, is exactly the vehicle for which the commoditisation ceiling is a floor rather than a guillotine.

I want to be precise about the status of these numbers, because precision is the whole currency of an essay like this. The Indian cost figures are modelled estimates, not audited comparables; the Flagship funnel ratios are drawn from the firm's own public descriptions of its process and should be read as illustrative of a discipline rather than as a forecast of any specific Indian vehicle's yields. The claim that survives every reasonable haircut to those numbers is the directional one: the front-of-funnel capital intensity of venture creation compresses dramatically at the Indian cost base, and that compression is what makes the small-fund, wide-funnel, mid-band-exit architecture viable in India and unviable in the US. You do not need the exact multiple to be right for the architecture to hold. You need only the order of magnitude, and the order of magnitude is not seriously in dispute.

The two games India already knows how to play

There is a version of this essay that reads as a leap into the unknown — a country with no frontier-discovery record proposing to originate biotech. That version is wrong, because the architecture is not a leap. It is two games India has played for forty years, with a third bolted on top.

The first game is biosimilars. India has the deepest biosimilar manufacturing base in the world outside China, built on the same process-chemistry and regulatory-arbitrage muscle that made it the world's pharmacy for small-molecule generics. This is not frontier science and it is not supposed to be — it is the late-stage, fully-commoditised end of the modality curve, where the molecule is known, the patent has expired, and the entire competition is cost, scale, and manufacturing reliability. It is exactly the part of the curve where the Indian cost stack is structurally dominant, and the Sun–Organon transaction is in large part a bet on it: the deal vaults Sun toward the top tier of global biosimilars, and Sun launched a generic GLP-1 the moment semaglutide lost protection in India. The GLP-1 and peptide wave now cresting is, for India, not a frontier opportunity but a biosimilar and large-population opportunity — the one modality where the country's existing advantage and the world's largest patient pool point in the same direction at the same time.

The second game is population scale. India is simultaneously one of the largest disease markets on earth and one of the cheapest places to run high-quality clinical operations — the combination that let NexCAR19 reach three hundred and fifty patients and turn a profit while charging a tenth of the Western price. Scale is not a consolation prize behind frontier science; in a commoditising industry it is a primary source of advantage, because the late stages of every modality compete on volume and cost, and that is the competition India is built to win.

The third game — the new one, and the one this essay is really about — is a timing arbitrage on the commoditisation curve itself. The curve has a sweet spot. Very early, a modality is all technical risk and no proven value; very late, it is all proven value and no margin, fully commoditised. In between there is a window where the science has been de-risked by the first cohort but the value has not yet fully commoditised — where the technical question is largely answered but the cost curve has not yet collapsed. Entering a modality platform in that window is the structurally favourable trade: low residual technical risk, still-meaningful value to capture, and an Indian cost base that lets you build the de-risked version for a fraction of what the first cohort spent proving it was possible. This is the venture-grade version of India's traditional advantage — not "be first," but "be the cost-efficient fast-second into a modality the moment it is proven but before it is free." And the single best place to run that trade is the insight platform with a wet-lab loop built in — Holtzman's durable archetype — because it is the one shape where being a cost-efficient fast-second still compounds, since the data loop you build on the way in does not commoditise even after the underlying technique does.

Stacked, the three games are a barbell. Biosimilars and population scale are the low-variance cash engines that an Indian biotech ecosystem can run today, at known margins, with no frontier dependency. The timing arbitrage is the higher-variance origination bet that the cash engines can fund and the fund mathematics can carry. The mistake would be to treat the third game as a substitute for the first two. It is the apex of the same pyramid, and the base is already built.

What India does not have

An honest pre-cycle call has to spend at least as long on what is missing as on what is promising, because the failure modes are where the thesis actually lives or dies. Four constraints bind, and none of them is solved by the geometry.

The first is compute and frontier talent density. The recursive discovery factory — the one platform shape that escapes the ceiling — is exactly the shape India is least equipped to build, because it depends on concentrated compute and a deep bench of people who have shipped frontier models, and both are scarcer in India than the optimistic version of this essay would like. The honest move is not to pretend the gap is small. It is to concentrate India's scarce frontier resources behind one or two credible attempts rather than spreading them thin, and to compete on the dimensions where India has structural advantage — capital patience and cost-efficient wet-lab throughput — rather than on the dimension where it does not.

The second is capital patience itself, which I have described as India's structural advantage and which is, just as accurately, India's structural absence. The advantage is latent. Indian biotech alternative investment funds have generally been structured on consumer-technology timelines — five-to-seven-year horizons, portfolios of fifteen to twenty companies, average cheques in the one-to-two-million-dollar range — because that is what the available limited-partner base understands and underwrites. The biology does not fit those timelines, and the architecture this essay proposes requires limited partners who will accept a different shape of return on a different clock. The patient capital is a possibility, not a fact, and someone has to go raise it from people who have never been asked to hold biology for a decade.

The third is regulatory portability. The CDSCO pathway that lets NexCAR19 price a CAR-T at a tenth of the Western figure is a genuine advantage inside India, but an asset approved only in India is an asset with a capped buyer universe. The mid-band strategic exit the whole thesis targets assumes a global pharma buyer, and a global pharma buyer underwrites against FDA and EMA pathways, not CDSCO alone. The architecture has to be built so that the assets it originates are legible to a global regulator and a global acquirer from the start, or the exits it is counting on will not materialise at the prices the model assumes. This is a design constraint, not a deal-breaker, but it is a constraint that an India-only mental model will quietly ignore until it is fatal.

The fourth is exit-market depth, and I have given it its own section already because it is the load-bearing assumption of the whole fund mathematics. To restate it as a constraint rather than a hope: the two-hundred-to-seven-hundred-million-dollar strategic-exit band must clear, routinely, for Indian-originated assets. China has proven the band clears at scale for a non-Western originator; India has proven only that it can be the buyer. Until an Indian-originated asset actually transacts in that band, the geometry can be entirely correct and entirely irrelevant at the same time. This is the single most important thing to watch, and the thing most likely to be the reason this essay is wrong.

The architecture nobody has built yet

Put the pieces together and the prescription writes itself, which is usually a sign either that the argument is sound or that it has been over-fitted. I think it is the former, and I have tried to load enough disconfirming weight onto it in the preceding two sections that a reader can judge for themselves.

The architecture is an origination-first venture-creation engine, domiciled in India, running on a fund small enough — call it fifty million dollars to start — that the one-to-three-billion-dollar commoditisation ceiling is its floor rather than its ceiling. It runs a Flagship-shaped funnel at the Indian cost base: wide at the top, ruthless at the gates, capital-efficient through the expensive middle, with per-company costs an order of magnitude below the US equivalent. It enters at origination and holds enough ownership that a mid-band exit returns the fund several times over. It builds its assets to be legible to global regulators and global acquirers from the first experiment, so that the exits it is counting on actually clear. It targets the two-hundred-to-seven-hundred-million-dollar strategic-exit band as the base case and treats the tail outcome as upside it does not need. And it places one or two patient, concentrated bets on a recursive discovery factory — the one platform shape that escapes the ceiling — while refusing to pretend that the compute gap between India and the frontier is anything other than what it is.

None of the modalities arriving in 2026 — AI-designed small molecules, in-vivo cell therapy, GLP-1 and peptide biosimilars, the recursive discovery factories themselves — is yet priced into an Indian capital structure. All of them are priced, exhaustively, into the US-axis structure, which is why the US-axis structure is staring at a ceiling. The window to set the Indian pricing closes the moment the next two or three Indian biotech funds deploy capital against a thesis that names the inversion out loud. After that, the architecture is consensus, and consensus does not earn the return that the first correct mover earns. The right to win belongs to whoever internalises that the ceiling that ends the Western platform is the entry-cost discipline the Indian platform inherits for free — and acts on it before the sentence stops sounding strange.

The dated, falsifiable close

A thesis that cannot be proven wrong is not a thesis; it is a mood. So here is the test, with a date and a scoring rule, on the record.

By the fourth quarter of 2028, an India-domiciled venture-creation engine of the shape described here — on the order of fifty million dollars under management, running a wide origination funnel, with fund-returner mathematics calibrated to the two-hundred-to-seven-hundred-million-dollar exit band, and with at least one credible recursive-discovery-factory position — will either have materialised, with its first cohort of new companies visible at or near the investigational-new-drug stage, or it will not.

If two or more such engines are visibly operating and deploying against a thesis that explicitly names the capital-architecture inversion by the close of 2028, this essay is on the record as correct. If fewer than two have appeared, then the binding constraints — the talent gap, the absent patient capital, the regulatory-portability problem, or the shallowness of the mid-band exit market for Indian-originated assets — have bound harder than the geometry suggests, and this essay should be cited as the version that was confident and wrong. I would rather be cited as wrong with a date attached than be right in a way that could never have been checked.

The Indian biotech capital architecture has, until 2026, treated the Hershberg ceiling and the Flagship funnel as imports — useful to study, not yet relevant to deploy. The argument here is that the imports are now the natives. The ceiling is the desired endpoint. The funnel is the operating discipline. The recursive discovery factory is the one position the patient capital should be built around, held with a stop-loss and not a prayer. And the window to set the Indian architecture before the consensus arrives is open today and will not be open in thirty months.

Refresh in ninety days. Score at the end of 2028.

This essay extends Elliot Hershberg's "On Modality Commoditization" (The Century of Biology, March 2025) and the modality-platform vs insight-platform distinction articulated by Steven Holtzman and elaborated in Hershberg & Malone's "On Biotech Platform Strategy" (2023). Receipts across AbCellera, Genentech–Roche, the Sun Pharma–Organon transaction, ImmunoACT's NexCAR19, Insilico's rentosertib, Recursion, Isomorphic, and China's out-licensing wave are drawn from primary and trade sources logged in this essay's research dossier; market-capitalisation and deal figures are as of the first half of 2026 and will drift. The Indian cost estimates and the fund-mathematics worked example are the author's models, presented as illustrative geometry rather than audited comparables; the dilution, ownership-decay, and portfolio-level treatment is deferred to a dedicated follow-up, "The Venture Math of Biotech." India is the vantage; the modalities are the subject; the inversion is the verdict.