The fifty-year compression curve that makes platform commoditization one process, not three episodes
The 2023 piece named the genus. The 2025 piece named the commoditization. This piece names the lineage.
In November 2023, Elliot Hershberg and Patrick Malone published On biotech platform strategy in The Century of Biology. The essay did two things at once. It organized fourteen years of partnership data into a four-genus typology --- Modality, Insight, Services, Vertical Integrator --- that it built out from the two-genus Product-versus-Insight split Steven Holtzman, then chief executive of Decibel Therapeutics, had sketched a decade earlier. And it gave the rest of us a frame stable enough to argue inside. In March 2025, Hershberg returned with On modality commoditization, which pointed out, in the cleanest language anyone had used, that the moat shape of a Modality Platform decays faster than the platform's own clinical timelines. Those two essays are the parent works. Everything that follows is written inside their scaffold.
This essay is the first of a two-piece arc. It is the foundation. It names the lineage --- the fifty-year compression curve that makes Hershberg's 2023 and 2025 essays both necessary readings of the same underlying process --- and it states the structural claims the companion piece, On Biotech Platform Strategy, 2026: The Index, then takes up against current-period data. The Lineage refreshes annually. The Index refreshes quarterly. The Lineage is what The Index stands on.
The claim we will defend across nine sections runs as follows. Biotech platforms have been getting commoditized for fifty years, the curve has compressed in each successive era, and each successive cohort of platform builders has captured roughly a tenth of the value of the cohort that preceded it. The 1980s recombinant-DNA cohort built $100 billion companies. The 2000s services cohort built $10 billion companies. The 2020s AI-discovery cohort, by every public data point currently available, is capping at $1 billion. The Hershberg 2025 commoditization thesis is the descriptive surface of this structural fall. The 2023 typology is the load-bearing classification beneath it. The Lineage is what makes the fall legible as one process, not three episodes.
A second register sits underneath the first. Hershberg's frame classifies platforms by defensibility. Holtzman's earlier two-genus split --- Product Platforms and Insight Platforms --- classified them by what the platform generates. The lineage from Holtzman to Hershberg to The Index's fifth archetype is one continuous taxonomy, refined under selection. The companion piece extends the genus tree by one archetype. This piece states why the tree had room.
We are not here to be quantitative about the slope. The data is too messy, the cohorts too small, the public marks too few. We are here to state that the slope exists, that it points the way it points, and that the curve's mechanism --- open-source closing the gap to closed-source, in months rather than decades --- is the engine of all three eras. The exact multiples are contested. The direction is not.
The compression curve runs five eras, and each era halves the next era's lead time
The cleanest way to read the fifty years is as a sequence of five eras, each one shorter than the last, each one ending when a closed-source pioneer's edge is matched by an open-source rival.
1976--1985. The recombinant DNA era. Genentech is founded in 1976. The IPO in 1980 prices at $35, opens at $88. Amgen incorporates in 1980, goes public in 1983. The shared secret is the recombinant-DNA technique --- Cohen and Boyer's 1973 paper at Stanford, licensed widely, but operationalized first by a small cohort that knew how to scale a bacterial fermenter and purify a folded protein at clinical-grade purity. The IPO frenzy is the visible surface. The architectural fact underneath is that the technique was secret in execution, not in publication. The decade ends with several dozen biotech companies in clinical development and a handful of products approved.
1985--1995. Bubble burst and technique maturation. The recombinant-DNA cohort hits the wall every technology cohort hits: more capital deployed than near-term products can absorb. The shared secret stops being a secret as the technique matures and the operators move between companies. Genentech and Amgen survive because they have product cashflow; the long tail does not. The era ends with the surviving cohort firmly capitalized --- Genentech and Amgen on durable footing, Regeneron founded in 1988 --- and the technique now in widespread industrial use.
1995--2010. Platform development and the first service providers. The locus of innovation rotates from the protein to the antibody-discovery platform, the high-throughput screen, the structural-biology software. Genentech is acquired by Roche in 2009 for $46.8 billion. Adimab is founded in 2007 --- antibody discovery as a service rather than as a vertically integrated drug company. ChemDraw, the closed-source cheminformatics suite from CambridgeSoft, dates to 1985; RDKit, its open-source rival, ships in 2006. Twenty-one years. That is the gap that will be compressed in every subsequent generation.
2010--2020. Development-as-a-service emerges. The contract research organization stops being a back office and becomes a discovery partner. AbCellera is founded in 2012. Recursion in 2013. Schrödinger Suite (1990) is matched by OpenMM and AutoDock around 2005. Pipeline Pilot (2001) is matched by KNIME (2006) --- a five-year gap, a quarter of what the recombinant-DNA era took. Benchling launches in 2012; the open-source equivalents --- Galaxy, the R-Bioconductor and Python notebook stack --- are essentially contemporary. The closed-to-open gap has fallen from twenty-one years to under one. AI is not yet the protagonist; the cloud and the open scientific stack are.
2020 onwards. AI compresses what is left. AlphaFold 3 is published by Abramson et al. in Nature in May 2024, initially behind a restricted-access interface. Within months --- not years --- open-weight competitors and reproductions in the same performance class arrive: Chai Discovery's Chai-1, then Boltz-1 and others. The closed-to-open gap is now a single quarter. The 1980s recombinant-DNA cohort had a decade of monopoly on a technique. The 2024 structural-prediction cohort had a quarter. The compression is not a metaphor. It is a measured halving, era by era, until the gap is gone.
What the sequence describes is not a story about AI. It is a story about how the same process --- closed-source pioneer, open-source rival, gap closes, value migrates --- has run five times, and each time the lead time the pioneer holds is roughly half of what the previous pioneer held. The 2020s AI cohort is reading the same script the 1980s recombinant-DNA cohort read, only on fast-forward. This is the Lineage.
A small honesty marker, because the data is noisy. The era boundaries are not crisp. Some platforms span two eras. The exact closed-to-open gap depends on which closed product and which open rival one picks. The point is not the precise number. The point is that across five eras, every fair comparison shows the gap halving. Directionally certain, calendar-loose.
The mechanism underneath the curve --- foundation-model spillover, cloud-native collaboration, academic-industry permeability, composability beats suites --- sits in Section 7. The empirical consequence --- the 90 percent value-destruction rule --- sits in Section 3, immediately below.
Each era captures roughly a tenth of the value of the era before it
The compression curve has a price. Plot the median peak market capitalization of the platform leaders in each era and the chart is monotone in one direction. Each successive cohort of platform builders captures roughly a tenth of the value of the cohort that preceded it. We will call this the 90 percent rule, which is not a law but a striking enough regularity that it deserves the name.
First Movers, 1980s through 1990s. $100 billion band. Genentech is acquired by Roche in 2009 for $46.8 billion. Amgen's market capitalization sits at roughly $168 billion at the time of writing. Regeneron at roughly $78 billion. These are the platform builders of the recombinant-DNA era, and the rough order of magnitude is unmistakable. The platform was a technique-plus-infrastructure stack --- protein expression, purification, characterization, scaled fermentation, regulatory navigation --- and the value captured at the platform level dwarfed the value of any single asset.
Service Providers, 2000s through 2010s. $10 billion band. Adimab raised at roughly $10 billion private valuation in its last priced round. AbCellera went public in December 2020 at a peak intraday market capitalization of $15.66 billion. Twist Bioscience has traded in a band capping below $4 billion. The platform here is the antibody-discovery service, the high-throughput synthesis service, the DNA-writing service. The companies are real businesses with real revenue and real customers, and the cohort still caps an order of magnitude below the first-movers.
New Entrants, 2020s. $1 billion band. Of the public AI-discovery pure-plays, none has crossed $5 billion at the time of writing, and most cap below $2 billion. The private cohort --- including the $1 billion Xaira seed and Isomorphic, which raised a $600 million first external round in 2025 and a $2.1 billion Series B in May 2026, both at undisclosed valuation and with zero clinical assets --- has more upside still resident, but the public marks set the ceiling the market is currently willing to pay for the discovery layer in isolation.
Three eras, three orders of magnitude. The cohort that built the first wave of biotech platforms captured roughly a hundred times what the cohort building today appears positioned to capture, on current marks. Even granting that the 2020s cohort is young, the AI-discovered drugs are not yet approved, and the eventual outcomes may surprise --- the gap is structural enough that the question to ask is not whether the rule holds but why.
The why is the compression curve. When the closed-to-open gap shrinks from twenty-one years to under two months, the window in which a single platform builder can compound by selling exclusive access shrinks with it. The 1980s recombinant-DNA cohort had a decade to capitalize the platform inside one company. The 2024 structural-prediction cohort had a quarter. A company cannot build a $100 billion platform inside a quarter. It can build a useful tool, a partnership, an asset --- but the durable monopoly that compounded into Genentech is not architecturally available on a sub-annual closure cycle.
This is the empirical answer to Hershberg's 2025 modality-commoditization thesis. The commoditization is not a one-modality story. It is the latest expression of a fifty-year compression. We have priced the platform layer down by a factor of a hundred across two generations, and the third generation looks set to extend the run by another factor of ten. The receipt that this is happening now, not in some future state, is the AbCellera mark --- $15.66 billion in December 2020, roughly $1 billion through the first half of 2026, a drawdown of more than ninety per cent across the most carefully built services platform of its era. Section 8 returns to AbCellera in full.
The honest qualifier sits in the next sentence. Public marks can mislead. The cohort is small. The eventual market capitalization of an AI-discovery pure-play that ships a Phase 3 readout in 2028 is not knowable from 2026 marks alone. What we can say with confidence is that the platform-layer premium that built three $100 billion companies in the 1980s does not appear to exist for the 2020s cohort at the same architectural position. The platform-layer premium has migrated. Where it has migrated to is the subject of Sections 4 and 5.
Holtzman, Hershberg, the fifth archetype --- the lineage is a refined classification under selection
The first stable classification of biotech platforms was Steven Holtzman's. Holtzman, then chief executive of Decibel Therapeutics, split the universe into two genera. Product Platforms pioneered new modalities, partnered out discovery as a service, and capitalized the chemistry. Moderna (~$61.9 billion market capitalization), Selecta, Dynavax, Alnylam (~$24.1 billion) were the canonical Product Platforms. Insight Platforms generated novel insights into pathways, targets, and disease relationships, and capitalized the dataset. Holtzman further split the Insight genus by scope: broad-applicability Insight Platforms (Grail at ~$8.1 billion, Peptris, Ginkgo at ~$5.1 billion, Tempus at ~$5.0 billion) and disease-specific Insight Platforms --- Alnylam appearing in this column too, alongside Ionis (~$7.5 billion), Agios (~$2.1 billion), Dyno (~$500 million).
Holtzman's split was generative because it asked a structural question. What does the platform produce? --- a molecule of a particular kind, or a piece of biological knowledge of a particular kind? The genus is defined by the output, not by the moat shape. This is the move Hershberg later refined.
Hershberg and Malone's 2023 paper kept the spirit of Holtzman's classification and extended it from two genera to four. Modality Platforms pioneer new chemistries and defend with composition-of-matter intellectual property. Insight Platforms generate biological knowledge and defend with data exclusivity or model weights. Services Platforms scale up shared infrastructure and defend with switching costs and process power. Vertical Integrators carry an asset through to a commercial product and capitalize the cashflow. The classification by defensibility is what makes the 2023 essay tractable as VC underwriting --- every genus has a different 7 Powers shape and therefore a different financing pattern.
Hershberg's 2025 commoditization essay is best read as a partial revision of the 2023 typology. The Modality Platform's defensibility --- the patent on the chemistry --- is the weakest of the four moats on the compression curve, because the chemistry is what the open-source rivals replicate fastest. Modality is genus number one in 2023; it is the most exposed genus in 2025. The 2025 essay does not retract the typology. It adds a temporal dimension to it.
The Index, the companion piece to this Lineage, names the fifth archetype. The Recursive Discovery Factory --- the closed-loop, retroactive-improvement, dataset-on-the-balance-sheet, modality-agnostic platform whose unit of learning is the dataset rather than the molecule. Recursion, Isomorphic, Xaira, Insilico, Iambic, Genesis, Valo. The genus is defined neither by output (Holtzman) nor by moat (Hershberg) but by what the platform does --- by architecture. The Index treats the fifth in full and splits it into five sub-architectures by substrate.
The lineage Holtzman → Hershberg → The Index is one continuous refinement. Holtzman asked the output question. Hershberg added the defensibility question. The Index adds the architectural question. Each step in the lineage subsumes the previous classification without invalidating it. A Recursive Discovery Factory is an Insight Platform that has closed the loop on its own substrate. A Modality Platform with a data flywheel is a Recursive Discovery Factory of the Generative-Chemistry sub-type. The genera do not displace each other. They cumulate.
This is what we mean when we say the Lineage is the parent piece. The reader who has internalized Holtzman and Hershberg has the prerequisite vocabulary. The fifth archetype is not an exotic claim; it is the next refinement in a forty-year-old taxonomy that has been improving on a roughly decade cadence. The companion piece carries the load of stating the fifth in detail. This piece carries the load of stating why a fifth was structurally available.
A small note on credit. Holtzman's framework is the most under-cited piece of biotech platform vocabulary in current use. Most writing that uses the Product-versus-Insight split does not name him. We are naming him. The lineage is improved by being made visible.
Platforms compound because seven assets per $100 million of R&D beats four
Why has the platform premium existed at all? The question is not rhetorical. The 1980s recombinant-DNA cohort captured $100 billion of value because the market believed that a platform compounds R&D in a way a single-asset drug company does not. The belief was empirically warranted. ZS Associates, in 2023 secondary-source research, reports that validated platform companies generated approximately seven subsequent assets per $100 million of R&D, against approximately four for non-platform companies. The ratio is roughly two to one. The mechanism is synergies --- shared infrastructure, shared technique knowledge, shared regulatory experience, shared chemistry-and-manufacturing-controls work, shared clinical-trial machinery, shared biomarker development.
A platform that has run one programme to approval has paid the fixed costs of building the protein-expression line, validating the analytical methods, qualifying the contract manufacturers, training the regulatory team, and writing the chemistry-manufacturing-and-controls documentation. The second programme inherits all of that. The third programme inherits it again. The marginal cost of programme number five is, on the industry's own data, roughly half of what the marginal cost of programme number one was --- which is what produces the seven-versus-four ratio.
The compounding is not just in unit economics. ZS's same 2023 work shows that validated platforms accelerate the clinical advance of the second-through-fifth asset relative to the first. The platform's second programme reaches first-in-human roughly faster than the first did. The third faster again. The acceleration flattens out around the fifth asset, by which point the platform's diminishing returns on synergy start to look more like a normal pharma pipeline. The compounding window is real but bounded.
This is the structural justification for the platform premium. A platform that throws seven assets at the wall on the same R&D budget that throws four for a single-asset shop is a different financial object. The financing pattern follows. Platforms can sell partnerships --- upfront payments, milestone payments, royalties --- because the buyer is paying for access to a productive system, not for one molecule. The platform retains the optionality on the other six assets. The financing pattern in turn extends the platform's runway, which lets it run more cycles, which generates more assets. The flywheel is structural; the premium is its capitalized form.
The honest qualifier is that the seven-versus-four number is a median across a noisy cohort. Individual platforms vary. The number rests on ZS's classification of which companies count as platforms --- a judgement call. And the ratio does not control for selection: companies that build platforms are the companies that had the capital to do so, which is endogenous to having more assets in the first place. We have read the ZS work and judged the methodological adjustments not large enough to overturn the directional claim. The ratio is roughly two to one. The platform premium is empirically warranted, on data the industry generates against itself.
What is interesting in 2026 is that the Recursive Discovery Factory cohort --- Recursion alone runs ten clinical programmes post-Exscientia merger, on a single platform --- appears to be pushing the ratio further. If the ratio for an RDF turns out to be three or four to one rather than two to one, the structural argument for a fifth archetype strengthens considerably. That is the empirical bet The Index names and times.
Three revenue streams diversify three valleys of death, and the diversification is the business model
The platform business model has three revenue streams, and the structure is more load-bearing than the language. An upfront payment when a partnership is signed. Milestone payments when programmes hit pre-specified clinical, regulatory, or commercial gates. Sales royalties when a partnered programme reaches the market. The platform builder, having capitalized the engine, sells access to the engine in this three-part form. The buyer --- typically a large pharma --- pays for the option to develop the platform's outputs, on terms that share risk along the development path.
The diversification matters because biotech has three valleys of death. The translational valley: a platform's outputs work in mice and fail in patients. The financing valley: the platform's burn outruns its capital before its first partnership scales. The regulatory valley: the platform's lead asset receives a complete response letter or a clinical hold that suspends the entire pipeline. A single-asset company cannot diversify across these. A platform structurally can. Upfront cash mitigates the financing valley. Milestone payments mitigate the regulatory valley. A portfolio of partnerships, each at a different clinical stage, mitigates the translational valley.
The mechanism is the same one a venture-fund portfolio uses. The platform builder is running a portfolio of bets on its own engine. Some partnerships pay out at the milestone gate and never reach royalty. Some reach royalty and underperform consensus. A small number compound into the durable royalty streams that justify the platform's market capitalization. The portfolio shape is what makes the platform's economic profile different from a single-asset company's, even when the underlying success rates per asset are similar.
What changes in 2026 is the mix. Pharma deals are now 70 percent contingent --- the upfront fraction has fallen, the milestone fraction has risen, the structure has rotated toward post-event payment. This compresses the platform builder's near-term cashflow and pushes more of the value to the back of the deal, where the buyer has more control. The platform that survives this shift is the platform whose engine is productive enough that the milestone density keeps the financing valley closed. The 2026 platform builder is being underwritten on its ability to clear gates, not on its ability to sign deals. The financing pattern has rotated to match the commoditization of the chemistry layer.
The 2025 Hershberg commoditization thesis lands precisely here. When the chemistry layer is commoditizing, the upfront payment falls, because the buyer is no longer paying a premium for access to a scarce technique. The platform's business model survives the commoditization only if the milestone-rich portion of the deal can be earned at scale. Which is to say: only if the engine throws enough quality candidates, fast enough, that the milestone schedule keeps the platform funded. Section 8 returns to the empirical question of which architectures can clear this bar.
The closed-to-open curve compresses because open-source closes faster than corporate secrecy can earn back its capital
We named the closed-to-open compression in Section 2. The mechanism deserves its own section because it is the engine driving the entire fifty-year curve. The table is short and the trend is monotone.
ChemDraw, the closed-source cheminformatics suite, launches in 1985. RDKit, its open-source rival, ships in 2006. Twenty-one years.
Schrödinger Suite (1990) is matched by OpenMM and AutoDock around 2005. Fifteen years.
Pipeline Pilot (2001) is matched by KNIME Analytics Platform (2006). Five years.
Benchling (2012) is matched by Galaxy plus R-Bioconductor and Python notebook workflows, essentially contemporary in research-grade use cases. Roughly zero to one year.
AlphaFold 3 --- Abramson et al. in Nature, May 2024, initially access-restricted --- is matched in the same performance class by Chai Discovery's Chai-1 and the wave of open structure models that followed (Boltz-1 and others) within months. A single quarter.
The lead time the closed-source pioneer holds has fallen from two decades to two months across forty years. Four mechanisms explain it.
The first is foundation-model spillover. A protein-language model trained at one institution generates an embedding space that any downstream group can fine-tune. The pretraining is the expensive step; the fine-tuning is comparatively cheap. When the pretraining is published --- or when an open-source rival pretrains a comparable model on public data within months of the closed release --- the closed pioneer's lead time collapses. This was not true in the 1980s, because the equivalent of pretraining was the physical fermentation infrastructure, which did not transfer between groups by uploading a paper.
The second is cloud-native collaboration. A 2024 graduate student in São Paulo runs the same compute stack as a 2024 graduate student in Cambridge. The cloud has flattened the infrastructure layer that, in the 1990s, took a regional cluster to build. The open-source rival no longer has to wait years to assemble the operational substrate the closed pioneer had. The substrate is rented by the minute, in any country with payment rails.
The third is academic-industry permeability. The people who built AlphaFold are also the people who, after a few years, start the spinout, leave for the open-source rival, or publish the next-generation method as a preprint. The talent layer transfers faster than the corporate secrecy regime can contain it. Non-compete clauses are imperfect, retention bonuses are expensive, and the median structural-biology PhD now expects to switch institutions every three to five years.
The fourth is composability. Modern open-source tools are designed to fit together. KNIME plus RDKit plus AutoDock plus an open chemistry transformer plus a notebook is a workflow assembled in days. The closed suite --- ChemDraw plus its proprietary database plus its custom plugin layer --- is a system built over years. The composable open stack can match a closed suite's specific workflow without matching its specific feature set. Composability beats suites, on the workflow buyer's actual margin.
The four mechanisms compose. Foundation models spill; the cloud removes the infrastructure moat; the talent migrates; composability eats the suite's product surface. The compression curve we drew in Section 2 is what these four mechanisms produce when run together for forty years. The 1980s recombinant-DNA cohort had a decade of monopoly because none of these mechanisms operated at scale. The 2024 structural-prediction cohort had a quarter because all four operate at full force.
The consequence for platform builders is direct. Corporate secrecy no longer buys a multi-year lead. The platform that depends on its chemistry being closed-source is depending on a moat that is now measured in weeks. The platform that capitalizes its dataset, its talent stack, its physics priors, its closed loop --- capitalizes something that the four mechanisms compress more slowly. Which is the same point The Index makes from the architectural side. The Lineage states it from the mechanism side. Both are the same claim about which moat shapes survive the curve.
AbCellera is the empirical confirmation of the $1 billion services ceiling
We named the 90 percent rule in Section 3. The single most carefully prosecuted test of whether the rule is structural or accidental is AbCellera Biologics. AbCellera went public in December 2020 at the peak of the pandemic biotech surge, on the back of a covid-19 antibody --- bamlanivimab --- discovered and developed in roughly ninety days against an active outbreak. The peak intraday market capitalization on the opening days post-IPO was $15.66 billion. The company had, on the technical merits, one of the strongest antibody-discovery platforms ever built --- a microfluidic single-cell screening system that compressed the antibody-discovery cycle from months to weeks.
By March 2026, AbCellera's market capitalization was roughly $1.03 billion, and it has traded in the $1--1.5 billion band through the first half of the year --- a drawdown of more than ninety per cent from the December 2020 peak. (The $1.97 figure that circulates is the fifty-two-week-low share price, not the market capitalization; the market cap is the right denominator, and it is the harsher number.) The technical merits did not change. The platform still works. By the end of 2025 it had reached 104 partner-initiated program starts carrying downstream economics, with nineteen molecules in the clinic, and its partners --- Eli Lilly, Regeneron, Pfizer among them --- still use it. What changed is the market's pricing of the services-platform genus. The 2020 mark capitalized the platform as if it were a hybrid services-and-vertical-integrator that would convert its discovery engine into wholly owned assets at a Vertical-Integrator multiple. The 2026 mark capitalizes it as a services platform, full stop, capped in the $1-2 billion band.
The AbCellera trajectory is what the 90 percent rule looks like inside one ticker. Twist Bioscience tells a similar story at a similar order of magnitude --- the DNA-writing platform peaked above $9 billion in early 2021, traded between $1.5 and $4 billion across the subsequent five years, and is now best understood as a services-platform-with-cashflow capped under $4 billion. Adimab, the antibody-discovery service that predates AbCellera, has not gone public; the last priced round signal we trust in the $5 to $10 billion band, which extends rather than contradicts the rule.
Hershberg's 2023 essay flagged the services-platform-converting-to-therapeutics archetype as the genus most likely to break out of the $1 billion band. The empirical record across 2024, 2025, and 2026 falsifies that hope. Services platforms have remained services platforms, on the marks the market is currently willing to pay. The conversion mechanism --- using the platform's discovery output to build a wholly owned pipeline --- has worked technically. It has not worked financially. The market discounts the wholly owned pipeline of a services platform at a steeper rate than it discounts the pipeline of a pure-play biotech. The reason is structural: the buyer of a services-platform-with-pipeline is uncertain whether the platform's best customers will leave when the platform becomes a competitor.
The conclusion from AbCellera is not that the services platform is a bad business. It is a good business. AbCellera generates real revenue, has a real customer base, and runs at meaningful margin. The conclusion is that the ceiling on the services platform's market capitalization is real and structural --- the market has decided, across multiple test cases in multiple sub-types, to cap the genus in the $1-3 billion band. This is the empirical confirmation of the 90 percent rule at the services-platform level.
What the ceiling means for the next cohort is the work The Index picks up. If services platforms cap at $1-3 billion, and the next cohort --- the AI-discovery cohort --- is currently capping at the same band, then the question is whether the new cohort's architectural advantages let any individual platform punch through. The Lineage states the ceiling exists. The Index takes up what the ceiling means for the architectures that follow it, including the implications for India, where the services-platform-with-pipeline model has been the default playbook for a decade.
China is the structural threat, not the headline --- 30 percent of Western licensing in five years is a force, not a story
We can name China as a structural force or we can name it as a quarterly headline. The Lineage names it as a force. The Index names the headlines.
In 2020, Chinese-origin assets accounted for roughly 3 percent of Western pharma licensing deals. In 2025, by EY Firepower data and matched against BioCentury's deal database, the share is approximately 30 percent. A factor of ten in five years. Pfizer has committed $1 billion across China over five years. AbbVie and BMS have hosted partnering days in Shanghai. Roche, Bayer, and Eli Lilly are operating innovation incubators inside China. Hengrui Medicine has cumulatively out-licensed more than $16 billion of preclinical and clinical assets to Western buyers across the period.
The structural reasons are visible if one is willing to read them as substrate rather than as narrative. China's 2015 regulatory reforms compressed the timeline from preclinical-to-IND for the average asset, in part by reorganizing the National Medical Products Administration along lines that match FDA timelines more closely than the legacy CFDA process did. The labour cost of running a clinical trial in China is materially below the cost of running the same trial in the United States, on a per-patient and per-site basis. The domestic pharma buyer market in China is real, growing, and has its own out-licensing logic to the rest of the world. The Chinese biotech founder, in 2026, has buyers, capital, and regulatory speed that the 2015 Chinese biotech founder did not.
The 30 percent share is not a story about geopolitical surprise. It is a story about a substrate --- regulatory speed plus labour cost plus domestic-buyer depth --- producing an output that Western pharma is willing to buy on the same milestone-heavy terms it would buy a Boston asset. The asset moves through the same partnership structures the Lineage has described in Section 6 --- upfront plus milestone plus royalty --- and the milestone density is structurally higher because the Chinese platform has compressed the early-development timeline. The Western buyer pays the same dollar for an asset that arrived in clinical readiness in less time and at less capital. This is the rational response to a structural shift.
The implication for the Lineage is that the closed-to-open compression curve we have described, and the platform commoditization Hershberg named, are both running in parallel with a geographic compression of who can supply the front of the drug-development funnel. The 1980s recombinant-DNA cohort had a geographic monopoly --- the platform technique existed in roughly a dozen American and European institutions. The 2020s AI-discovery cohort does not. The same compute stack, the same open-source models, the same talent layer is now resident in Shanghai, Hangzhou, Seoul, Bangalore, and São Paulo. The geographic monopoly that buttressed the value capture of the first three eras is not architecturally available to the fourth.
This is the threat Hershberg's 2025 commoditization essay flagged as a quarterly headline. The Lineage frames it as a structural force. The two readings are not in conflict. They are the same observation made at different time-scales. The Index will carry the current-period deal flow and the specific 2026 transactions. The Lineage states the longer claim: the geographic substrate has flattened, and the value-capture ceiling that depended on geographic monopoly is being repriced accordingly.
A small note on India, because we will not let it sit as an appendix. The 30 percent share is overwhelmingly Chinese. The Indian share of Western licensing is materially smaller. The substrate reasons --- regulatory speed, domestic-buyer depth, talent-stack composition --- are not yet aligned in India the way they are in China. The Index's India section treats this in detail. The Lineage notes here that the geographic compression is real, that it is currently disproportionately Chinese, and that the architectural question for the Indian platform builder of 2026 is whether the same substrate can be assembled on a five-to-ten-year horizon. The honest answer in this register is: not yet, and the timing is uncertain.
Leading pharma now operates like a venture-capital portfolio, and the platform's price is set by buyer behaviour
The demand side of the Lineage has its own evolution, running on the same fifty-year clock as the supply side. The buyer of the platform --- large pharma --- has rotated its dealmaking posture three times in the period the Lineage covers.
The first rotation, roughly 1995 through 2010, was capability-focused. Pharma bought platforms for the technique. Novartis acquired Chiron in 2006 in part for the vaccine platform. Roche acquired Genentech in 2009 for the antibody platform. The deal logic was: own the technique, own the next decade of products. The capability acquisition closed the moat on the buyer's side at the cost of paying the platform's full strategic premium.
The second rotation, roughly 2010 through 2020, was asset-focused. The Bristol-Myers-Squibb acquisition of Celgene in 2019 was, at its core, an acquisition of three late-stage assets --- Revlimid, Pomalyst, ozanimod --- and the cashflow they generated. Pharma stopped paying the platform premium and started paying the asset premium. The buyer's view became: own the product, share the risk on the next product. The platform was no longer the prize. The molecule was.
The third rotation, roughly 2020 onwards, is in-house plus targeted asset. Eli Lilly, having acquired Prevail Therapeutics in 2021, built the Institute for Genetic Medicine to internalize its own gene-therapy platform rather than continue to source it externally. Bayer's reorganization around Asklepios BioPharmaceutical's adeno-associated-virus platform follows the same logic. The early-rotator pharmas (Novartis, Roche, Lilly) have rotated from capability to asset to in-house plus targeted asset, in that sequence. The late-rotators (Amgen, certain phases of Pfizer) are still catching up via in-house build. Bristol-Myers-Squibb and Pfizer continue the targeted-asset strategy. The dispersion across pharma's eight-to-ten major buyers is now real.
What unifies the third rotation is that pharma is now operating like a venture-capital fund. EY Firepower's 2025 work flags this explicitly: the median large-pharma business-development team runs its outside dealmaking on a portfolio-construction logic, with explicit allocations across therapeutic areas, modalities, and stages; with milestone-heavy terms that look like preferred-equity structures; with post-deal operational support that looks like a venture fund's value-add; and with portfolio rebalancing on roughly a venture-fund cadence. The headline acquisitions of 2025 --- Sanofi acquiring Blueprint Medicines for $9.1 billion, Lilly acquiring the Scorpion Therapeutics PI3K-alpha programme for $2.5 billion, GSK acquiring IDRx for up to $1.15 billion, Novartis acquiring Anthos Therapeutics for $3.1 billion --- read more cleanly as a venture portfolio's later-stage exits than as the legacy mega-merger logic.
The headwinds for 2025 and 2026 are material. Tariffs could rise from approximately $0.5 billion to $63 billion annually on the assumptions buried in the 2025 US administration's pharma-pricing policy. The most-favoured-nation drug-pricing reforms compress the pharma buyer's expected revenue per approved product. FDA regulatory uncertainty raises the cost of capital on every late-stage asset. Pharma's response has been to deepen the contingency in deal structure --- 70 percent of 2025 deals include contingent payments, against approximately 50 percent five years earlier --- and to lean into the venture-portfolio posture rather than retrench into mega-mergers. Deal value is concentrating: M&A excluding megamergers was approximately flat between 2018 and 2023, while total deal value rose to $191 billion in 2024 on the back of a small number of large transactions. The shape of the buyer's behaviour is the shape of a fund that allocates patiently and rebalances ruthlessly.
The implication for the platform builder is direct. The price a platform commands is now set by a buyer whose default posture is venture-style. The platform builder is being underwritten as a portfolio company by a pharma that runs a portfolio. Which means the platform's milestone-payment density, its capital efficiency per asset, and its rate of throwing IND-stage candidates matter more than its named modality. The buyer is no longer paying for the chemistry. The buyer is paying for the cadence.
This closes the loop with Section 6. The platform's three-stream business model --- upfront, milestone, royalty --- has rotated under the buyer's repricing. The upfront has shrunk. The milestone density has become the load-bearing variable. The royalty tail is increasingly contingent on commercial outcomes the buyer controls. The platform that survives this configuration is the platform that throws fast enough, on a substrate productive enough, that the milestone schedule keeps the engine funded. Which is, again, the architectural argument The Index picks up in detail.
What this Lineage cannot tell you is what 2026 changed
The Lineage closes here because the structural claims it can make have been made. The fifty-year compression curve has been drawn. The 90 percent value-destruction rule has been documented. The Holtzman-Hershberg-fifth-archetype lineage has been named. The mechanism --- closed-to-open compression driven by foundation-model spillover, cloud-native collaboration, academic-industry permeability, and composability beating suites --- has been stated. The services-platform ceiling has been confirmed empirically. The geographic flattening, the buyer-side venture-portfolio rotation, and the rotation of platform value capture from chemistry to cadence have all been priced into the same structure.
What this Lineage cannot tell you is what 2026 changed. There is a new archetype that needs the full Index treatment --- the Recursive Discovery Factory and its five sub-architectures by substrate. There are four BioCentury data series --- the AI-discovery deal trace, the China-licensing rotation, the Vertical-Integrator median market-capitalization shift, and the Modality-Platform deal-value collapse --- that need to be threaded against the structural claims this piece has made. There are three years of post-pandemic platform data, six clean cases that satisfy the fifth-archetype definition, and an Indian cohort whose architectural maturity needs to be read against the same framework the rest of this Lineage has used. None of that is the work of an annual foundation piece. It is the work of a quarterly Index.
That is the next piece in this arc. On Biotech Platform Strategy, 2026: The Index --- ships one month after this Lineage. Read in order if reading both. Read The Index alone if the structural claims here are already internalized; the Index makes its argument against the parent works directly. Read this Lineage alone if the question is why the fifty-year curve produced the platform-shape questions the 2026 cohort is being asked to answer.
We will be wrong on the slope. We will be wrong on the timing. We will probably be wrong, in some sub-section we cannot yet identify, on which 2026 marks turn out to anchor the next cohort and which turn out to be the noise around the median. The architectural claims --- that the compression curve runs, that the 90 percent rule is structural, that the lineage from Holtzman to Hershberg to the fifth archetype is one continuous taxonomy, and that the buyer's rotation to a venture posture is real --- we are willing to write down before the next year of data lands. The framework is the gift. The future is open.