The Harvest Illusion

In 2025, Chinese biotech out-licensed roughly $136 billion of novel drugs — nearly triple the year before — and a single Chinese antibody beat Keytruda, the best-selling medicine on Earth, in a head-to-head trial. The consensus wrote the headline immediately: China has arrived as an originator of medicines, a second United States, a renaissance. The consensus is reading the wrong number, and then reading it wrong. What you are watching is not a renaissance. It is a harvest — the cashing-in of a crop planted years ago — and three things about it are invisible from the headline: the harvest is far smaller in cash than the announced value; the field it came from has not been replanted; and the engine that does the replanting collapsed from $15.7 billion a year to $4.2 billion while the harvest was being sold. A harvest is the most visible event in a farm's year and the least informative about next year's yield. This essay is about that gap — and about the empty field next door, which is the most interesting thing in the picture.

I. The boom everyone is celebrating

Start with the numbers, because they are genuinely staggering and the argument fails unless you take them at full strength. In 2024, cross-border out-licensing by Chinese biopharma reached $51.9 billion in total deal value — a record, a ninety-per-cent jump on 2023. Then 2025 nearly tripled it, to roughly $135.7 billion. The first quarter of 2026 alone did about $60 billion — more than all of 2024, compressed into three months, a seventy-three-per-cent jump on the prior-year quarter. Average deal size crossed $1.3 billion, up seventy-six per cent year on year. The count of cross-border deals more than doubled, from 42 in 2022 to over 90 in 2025. And the composition moved up the value chain: by deal count, Chinese assets now make up roughly half of global out-licensing transactions, with US-originated deals falling to under thirty per cent; by deal value, China is about a third of the world's out-licensing, up from low single digits a few years ago. On the supply side of the global pipeline, China now holds about 30.5 per cent of the world's innovative drug candidates against roughly 33 per cent for the United States — a gap that was thirteen points two years earlier.

The marquee deals make it concrete. In a sixteen-month stretch into early 2026, Chinese-originated assets accounted for a large share of big pharma's biggest transactions, and the buyer list reads like the entire industry.

And it is not only money. In September 2024, the bispecific antibody ivonescimab — discovered by Akeso, partnered to Summit — beat Merck's Keytruda head-to-head in first-line non-small-cell lung cancer in the HARMONi-2 trial, cutting the risk of progression or death by forty-nine per cent. Keytruda had been standard of care in that setting since 2017 and is the highest-grossing drug in the world. A Chinese-originated molecule beating it in a randomized trial was the symbolic moment the narrative had waited for; Summit's stock rose roughly six hundred per cent. Underpinning the volume is a real concentration of capability: ADCs and bispecific antibodies dominate the deal flow, and China is now estimated to originate something close to ninety per cent of global ADC licensing. The consensus read writes itself. China is no longer the world's generics factory or its contract lab; it is an originator of first-in-class, best-in-class medicines, and Western pharma — facing a patent cliff that puts $200–230 billion of annual revenue at risk between 2026 and 2030 — is lining up to buy.

Every fact in that section is true. The conclusions drawn from them contain three errors, and they stack.

II. The first illusion — the harvest is smaller than the headline

The most-quoted number, the "$136 billion," is a measure of announced potential value, not cash. A modern licensing deal has three layers: an upfront payment at signing, milestone payments that pay out only if the asset hits specified clinical, regulatory, and sales targets, and royalties on eventual sales. The headline figure sums all of these at their maximum — it is the number you reach if every milestone, however distant and improbable, is eventually met. The cash that actually changes hands at signing is the upfront, and for the China cohort the upfront is a sliver. Across 2025's deals, upfronts averaged on the order of four per cent of total announced value. In aggregate, the total upfront cash across all of China's cross-border deals was about $5.6 billion in 2025 — against a headline of roughly $137.7 billion (the trackers range from $135.7 billion to $137.7 billion; PharmCube, the most-cited, reports the higher figure). The harvest, measured in money received rather than money promised-on-success, is roughly one-twenty-fifth of the number on the marquee.

Look at the structures and the pattern is uniform. AstraZeneca's up-to-$18.5 billion obesity deal with CSPC carries $1.2 billion upfront — about six per cent. AstraZeneca's earlier ~$5.3 billion AI-discovery pact with the same company carries $110 million upfront — about two per cent, with the rest in development and sales milestones. GSK's ~$12 billion package with Hengrui is $500 million upfront. These are milestone-heavy, "biobucks" structures, and the structure itself tells you something the celebration ignores: the buyer is also buying an option, not expressing a conviction. If a Western pharma believed an asset was a sure best-in-class winner, it would pay up to own it outright. Paying two-to-six per cent upfront and loading the rest onto success milestones is precisely how you buy a lottery ticket on someone else's science — you pay a little now, and you pay the rest only in the world where it works. The deal terms encode the buyers' real estimate of the odds, and that estimate is "cheap option," not "proven medicine." So the illusion compounds at the very first number: the field is being cleared for far less actual money than the headline implies, and the people clearing it are paying like people who know most of these assets will never trigger their milestones.

This does not make the assets worthless or the science fake — the upfronts are real, several billion dollars of real, and a few of these molecules will be enormous. It makes the renaissance framing wrong. A renaissance is when the world pays full price, with conviction, to own your originated science. A harvest sold for a small upfront and a basket of contingent milestones is what a buyer's market looks like when the seller needs cash and the buyer knows it.

III. The second illusion — output is not formation

Here is the number the celebration leaves out entirely. Venture investment into Chinese biotech peaked in 2021 at $15.7 billion. By 2024 it had fallen to $4.2 billion — a seventy-three-per-cent collapse, to roughly four per cent of global biotech venture funding. The engine that forms new Chinese biotech companies — that turns scientists and capital into clinical assets, that plants the pipeline — ran down to a quarter of its peak in three years, during exactly the window in which the licensing harvest set its records.

That is the whole argument in two data points, and it repays sitting with why they are causal rather than contradictory. Output and formation are different stages of one pipeline. Out-licensing measures what is being sold — assets that already exist, already have data, already cleared enough risk to be worth a Western cheque. Venture formation measures what is being planted — the new companies and programs that become sellable five to seven years later. A record harvest beside a collapsed planting season is not a paradox; it is a field being cleared faster than it is sown, viewed from the two ends at once.

Read the boom through that lens and its character changes. The most-cited summary of why this is happening does not describe triumph; it describes triage. As one trade headline put it, big pharma is throwing a lifeline to Chinese companies struggling to survive. The mechanism is explicit in the financing data: with venture funding collapsed and the public markets shut, license-out agreements became, for many Chinese startups, not a strategic choice but the last open channel to raise money. When you cannot raise equity and cannot IPO, you sell your best asset to whoever will write an upfront — not because the timing is right, but because payroll is due. Forced selling at scale looks, from a distance, exactly like a boom.

The public-market half of the collapse is just as stark. Chinese biotech IPOs peaked at 89 in 2021. The window snapped shut in the 2022 biotech winter; by 2023 only 21 biotechs listed on China's A-shares — offerings down fifty-seven per cent, fundraising down seventy. Hong Kong's Chapter 18A, the route built in 2018 specifically to let pre-revenue biotechs raise public equity, became a trap: of the 56 companies listed under it, only 13 still traded at or above their IPO price by the end of 2022 — and fewer still a year on. Picture that cohort — roughly sixty pre-revenue biotechs, each holding a pipeline from plausible to genuinely world-class, each locked out of equity by its own collapsed share price and out of every other channel by geopolitical risk. What does such a company do with a world-class asset and no way to fund it? It sells it. The harvest is not the renaissance's first fruit; it is the liquidation sale of a planting season that already ended.

IV. The biology — harvesting the crop while the seed bank empties

The metaphor is not decoration; it is the mechanism, and biology supplies the exact version. A seed bank is a store of variation — many genotypes held in reserve, most never planted, a few carrying the trait that matters when conditions change. Its value is not any single seed; it is optionality across an unknown future, which is why you keep the diversity and keep replenishing it, because seeds lose viability with time. A field of standing grain is the opposite: variation already resolved into one committed crop, valuable now, worthless as a store of the future. You can eat a field. You cannot plant next year from it unless you held seed back.

China spent roughly 2015 to 2021 filling a seed bank, and it was one of the great planting seasons in the industry's history.

Ivonescimab was planted in that window. So was almost everything in the licensing tables. The assets are real and many are excellent — that was never the question. The question is what happens to a seed bank you stop refilling.

Then the window shut, and it shut on the planting, not the harvest. The IPO market closed. The redemption wave began — the buy-back rights that Chinese venture deals carry, which convert, when no exit arrives, into an obligation on the founder to repurchase the investor's stake, draining the company of exactly the cash it would have used to start the next program. US–China decoupling, formalized in the BIOSECURE impulse, made Western public capital flee. Every channel that funds formation closed at once, while the one channel that monetizes existing assets — out-licensing — stayed wide open and even widened. So the system did the rational thing under those constraints: it harvested. It sold the standing crop into the one open market, used the proceeds to survive, and stopped planting, because the instruments that fund planting were gone.

This is what it means to optimize on the macro signal alone. The signal said sell assets to Western pharma; that is where the money is. Obeying it produced a spectacular year and ate the seed corn. A system that maximizes this year's harvest by drawing down the store of variation that produces future harvests is not growing. It is converting a stock into a flow and reporting the flow as income.

V. Why the model self-destructs

Lay this against the canon's account of how drug discovery compounds, and the self-destruct mechanism turns specific. The venture engine of biotech is a formation machine whose entire output is governed by a power law: value lives not in the average company but in the rare extreme winner, and the only way to reach that winner is to start enough independent, varied bets that one lands. Two things must compound for that to work — a steady stream of new formation (the variation) and a population of investors and operators who learn, across cycles, which bets to tend (the judgment). Kill formation and you starve the variation. The redemption wave does something worse than starve it: it discards it. When the 18A cohort is forced to liquidate assets and wind down rather than continue, the specific knowledge inside each team — what worked, what failed, the judgment that only compounds by surviving into the next program — is dispersed and lost. The harvest sells the assets; the bust scatters the people who would have planted the next ones. Variation destroyed, judgment reset to zero.

You can run a farm this way for several seasons. The standing crop from a great planting year is large, and selling it generates enough cash and enough headlines to look like prosperity. But the arithmetic is unforgiving on a lag, because an asset takes roughly five to seven years to travel from formation to licensing stage. So each year's harvest is a function of the planting one half-decade earlier.

The 2025 harvest was enormous because the 2018 planting was enormous. The 2031 harvest will be a function of the 2024 planting — and 2024 was $4.2 billion against $15.7 billion. The thin harvest is not a risk on this picture; it is a schedule. And the modality concentration sharpens it: what is being sold is heavily weighted to the few hot classes the golden window planted — ADCs, bispecifics — so when that specific inventory is harvested, the cupboard is not merely smaller, it is barer in exactly the categories the buyers want.

It is worth being precise about the canon claim underneath this, because it is the same claim that makes "The Biomanufacturing Bet" reach the opposite verdict about the same country. Biomanufacturing is infrastructure: a plant, once built, compounds — it runs for twenty years, climbs its cost curve, and the capital sunk into it keeps producing. China's manufacturing dominance is therefore durable, and the canon says so plainly. Drug discovery is not infrastructure. It is a formation engine whose product is optionality, and optionality is perishable — renew it continuously or it decays. So the same country can be genuinely, durably winning one game and harvesting-toward-exhaustion in the other, because the two games have opposite capital physics. The cost curve compounds. The binary does not. That is not a contradiction in the China story. It is the two halves of the canon, confirmed in one nation.

VI. The optimist's turn — China's bust is India's opening

None of this is schadenfreude, and the point of seeing it clearly is not to be right about China; it is to notice what the harvest illusion obscures — that a seed bank just emptied next door, in the largest pharmaceutical labor market on Earth that is not China, at the precise moment the West has decided it needs a source of drug innovation that is not China. That is the opening, and it is India's to take or miss.

The tailwind is real, if wobbly. The same decoupling impulse that helped freeze Chinese formation is, from India's side, a demand signal: Western pharma and capital are actively trying to de-risk dependence on China across the drug value chain, and India is the obvious diversification destination — large, English-speaking, IP-respecting, democratic, with a deep clinical and chemistry base and a government pushing its bioeconomy toward $300 billion. The honesty caveat: the BIOSECURE Act itself stalled in the US legislature, so the policy tailwind is an intention, not yet a law, and Western money has kept flowing to China regardless. The decoupling is real as a direction and unreliable as a guarantee. But the demand for a non-China originator base is structural and growing, and while Chinese venture funding collapsed, India's overall venture market actually grew modestly through 2024. The field next door is being abandoned just as the buyers go looking for a new one.

The opening is a trap, though, if India reads the harvest illusion as a how-to guide. The temptation will be to copy the model that produced the headline — flood the zone, start hundreds of fast-following ADC and bispecific programs, optimize for the license-out to Western pharma, spray and cull. That is the fish strategy: produce enormous numbers of offspring, invest almost nothing in each, let the environment kill nearly all, win on volume. It works in a stable, capital-flooded environment, and it is exactly the strategy that just hit its wall, because spray-and-cull discards variation and never compounds judgment — every cycle starts over. India does not have the capital depth to out-spray China even at China's diminished pace, and chasing it means entering the same self-consuming model one cycle late.

The strategy that compounds is the mammal strategy, and it is the one the canon has described all along under other names: produce few offspring, invest heavily in each, cap the downside of every bet so failure is survivable and the team and its knowledge persist into the next attempt, and let judgment accumulate across a small number of deeply-tended programs rather than resetting every cycle through mass death. This is the Recursive Discovery Factory's logic and the studio's logic — own a few bets you actually built, tend them, compound the learning. Concretely, for India, that means a short list of disciplines:

First, build for the binary, not the bench. The Venture Math of biotech says value is made or destroyed at a small number of clinical readouts; a mammal strategy spends to de-risk those readouts on a few assets, rather than spreading thin capital across many that all stall pre-clinical. Second, capitalize for survival, which means patient, concentrated capital — the structural thing India has historically lacked and the precise reason the opening is not an outcome. Third, grab the vacated base deliberately: target the modalities and indications where China was the supplier the West now wants to re-source — ADCs, bispecifics, the chemistry-heavy and manufacturing-adjacent assets where India's existing strengths overlap the demand. Fourth, hold seed back — keep formation running through the cycle so the 2030s harvest exists at all, which is the one thing China structurally failed to do. Fifth, sell with conviction terms, not desperation terms: a mammal-strategy company licenses from strength, keeping more economics and more control, rather than dumping its best asset for a two-per-cent upfront because payroll is due.

India's structural advantages — a low cost base, a deep talent pool, a government willing to fund early stages, and now a West that wants exactly this — fit the mammal strategy precisely. They do not fit a spray-and-cull race against a larger, if stalling, neighbor. So the future-cast is specific and falsifiable: if India spends the next several years planting the mammal way — few, well-capitalized, deeply-tended originator bets, built to survive their failures and accumulate judgment, aimed at the base of Western demand China is vacating — then the harvest of the early 2030s could be India's, grown from a seed bank that China was emptying while India was filling. Not because India out-sprayed anyone, but because it planted while the giant next door sold its standing crop and called the sale a renaissance.

I should say where this is wrong if it is wrong, because the excited-realist discipline requires it. China's formation engine may be restarting: Hong Kong listings roughly quadrupled in 2025, biotech stocks there more than doubled, and venture funding ticked up late in 2024. A country that planted once can plant again, and a single strong year could begin refilling the bank — though one good year does not undo a seed bank drawn down for three, and the recovery so far is concentrated in late-stage IPOs and the licensing reflexivity itself, not in early formation. Ivonescimab's global trials, still running outside China, could validate it as a true best-in-class drug rather than a China-only data point — its overall-survival benefit has not yet reached statistical significance, and "King Keytruda still reigns" until it does, but if it clears, the renaissance reading gains real ground. And India's structural advantages have been true for two decades without producing an originator industry; the binding constraint there has always been execution and patient capital, not opportunity, and an opening is not an outcome. The thesis is not that China is finished or India destined. It is narrower and harder to dodge: a harvest is not a planting, the two have been moving in opposite directions, and the country that confuses the first for the second will be surprised by the thin years a stalled formation engine has already scheduled.

Close

China's harvest is real, and it is the largest the industry has ever seen, and it is not a renaissance. It is the crop of a planting season that is over — sold for a small upfront and a basket of contingent milestones, into the one open market, by companies that had no other way to survive, while the engine that grows the next crop ran down to a quarter of its peak. Renaissances replant. Harvests end. The lesson for everyone watching the $136 billion and feeling envy or alarm is the same: do not study the harvest. Study the seed bank — who is filling one and who is emptying one — because that is the number that tells you where the next decade's medicines come from. The seed bank just emptied next door. The buyers are looking for a new field. The only question that matters now is whether India plants it the way that compounds — few seeds, deeply tended, in the ground the giant just abandoned — or sprints to copy the harvest it should have read as a warning.

A contrarian foil for Atoms and Cells, and the therapeutics sibling of "The Biomanufacturing Bet" with the China read inverted: China's manufacturing win is durable infrastructure; its discovery boom is a depleting harvest — because the cost curve compounds and the binary does not, which is the canon's two constitutions in one country. It applies "The Venture Math of Biotech" (formation, the power law, the binary, the buyer's option-pricing of milestone-heavy deals) and the Recursive Discovery Factory / Studio Math logic (few deeply-tended bets, judgment that compounds; the mammal over the fish). Licensing and pipeline figures ($51.9B 2024 → $135.7B 2025; Q1'26 ~$60B; avg deal $1.3B; ~32% of global out-licensing by value and ~50% by count; China ~30.5% of the global pipeline; patent cliff ~$200–230B 2026–30 (analyst estimates; Morgan Stanley cites ~$171B); upfronts ~4% of value; ~$5.6B total upfront cash 2025) are from PharmExec, CIRSD, PharmCube/PharmaSource, Jefferies (via CKGSB, C&EN), Evaluate (via Fierce Biotech), Vision Life Sciences, SynBioBeta (via Fierce), and GlobalData. Named deals are from Fierce Biotech, Drug Discovery Trends, Life Science Daily, BioBucks, and Pharmaceutical Technology. Ivonescimab / HARMONi-2 (PFS win over Keytruda, 49% risk reduction, China-only, OS not yet significant) is from Fierce Pharma, STAT, BioSpace, and Clinical Trials Arena. The formation collapse — venture funding $15.7B (2021) → $4.2B (2024), IPOs 89 → 21, the Chapter 18A cohort (13 of 56 above IPO price by end-2022, fewer still a year on), and the "lifeline" framing — is from CKGSB, BioSpace, and Owlposting. India venture and bioeconomy figures are from Crunchbase, Bain, and DBT/BIRAC. The BIOSECURE Act is, as of this writing, stalled in the US legislature — the tailwind is a direction, not a guarantee. All figures are dated and directional; the counter-case is stated in the text. The thesis is one line: a harvest is not a planting, and the country that confuses them will be surprised by a thin decade it has already scheduled.