The Venture Math of Biotech

The India essay made a claim and then refused to prove it. It showed that the same $2B exit returns 0.4× to a $300M fund and 2.4× to a $50M fund, held ownership flat to make the point unarguable, and wrote an IOU: the real model — dilution, ownership decay, the power law that turns a single multiple into a fund return — "is its own essay, and I will write it as one." This is that essay. It is the part of the argument that an LP actually underwrites.

The identity nobody writes down

Start with the thing most venture writing skips, because skipping it is how you get fooled. A fund's return is not a deal's return. The multiple on your best investment is a fact about one company; the multiple on your fund is a fact about a portfolio, and the two are related by an identity that looks trivial and is not:

Fund MOIC = Σ ( ownership at exit_i × exit value_i ) ÷ fund size.

Three variables hide inside that sum, and you control none of them once the cheque has cleared. The first is ownership at exit — not what you bought, but what survives after every subsequent round has diluted you. The second is the distribution of exit values — not the average outcome, which does not exist, but the shape of the tail. The third is fund size — the denominator, which is set the day you close the fund and never moves again.

Almost every confident sentence about venture returns is wrong because it reasons from one of these and ignores the other two. "We'll own fifteen per cent of a billion-dollar company" ignores dilution. "Our average exit is $400 million" ignores that there is no average. "We returned 0.4× on that position" ignores that the position is one draw from a distribution that is supposed to be brutal. The venture math of biotech is the study of how these three interact, and the punchline is that the third one — the denominator — quietly dominates the other two. That is the whole reason the India inversion works, and the only way to see it clearly is to build the other two variables first.

Dilution: the tax you pay for someone else's growth

You do not keep what you buy. Every financing round after yours sells new shares, and unless you write a follow-on cheque large enough to defend your percentage, your slice shrinks by roughly the fraction the new round sells. Dilution is the most predictable force in the entire model, and it is the one founders and first-time investors most consistently underestimate.

The benchmarks are not mysterious. Carta's 2025 cap-table data, drawn from tens of thousands of startups, puts median dilution at roughly twenty per cent at seed, eighteen per cent at Series A, fourteen per cent at Series B, and a little less at each round after. The median founding team holds about fifty-six per cent after seed, thirty-six per cent after Series A, twenty-three per cent after Series B, and somewhere in the mid-teens by Series C. Those are founder numbers, but the geometry is identical for an early investor who does not follow on: each round multiplies your stake by roughly one minus that round's dilution, and the multiplications stack.

Run it. An origination investor who buys eighteen per cent at seed and then sits through a full financing path — Series A, B, C, D — without defending the position holds, after the stack of dilutions and the option-pool top-ups each round demands, somewhere between six and eight per cent at exit. That is not a pessimistic case. That is the median case, and it is exactly why the India essay used six per cent as the diluted exit stake for a company that grinds all the way to a large outcome. The dilution is not a tail risk. It is the base rate.

Now change one variable and watch the whole picture move. Take the same eighteen per cent entry, but let the company exit after just one more round — acquired off a strong Series A, the way a de-risked biotech asset routinely is. Eighteen per cent times roughly 0.82 is about fifteen per cent. The investor walks in at eighteen and out at fifteen, because there was only one dilution event between entry and exit instead of four.

Here is the whole decay in one table — an eighteen-per-cent seed stake taken down two paths:

The two columns begin identical and end more than twice apart, and nothing in them is about the quality of the company. The only difference is how many times someone sold new shares between your cheque and the buyer's.

That contrast is the most important number in this essay, so state it plainly: the number of financing rounds between your entry and the exit is the master variable for retained ownership. Not the entry price, not the valuation, not the quality of the negotiation — the round count. A company that reaches its exit in two rounds hands its early investor roughly twice the ownership of a company that reaches a larger exit in five. This is the mechanism the India essay gestured at when it said capital-efficient companies "preserve ownership," and it is real, but it is conditional in a way that matters: it only holds if the capital-efficient company actually exits early. A company that needs five rounds to reach even a mid-band outcome has thrown the advantage away — it has eaten the full dilution stack and arrived at a small exit, the worst of both worlds.

And the round count is not a free choice; it is set by how much capital the science demands before someone will buy the asset, and biotech is unforgiving here. A clinical-stage company routinely raises three hundred million dollars or more before it reaches an exit at all: Lilly's 2025 purchase of Orna followed more than $320 million in venture funding, and AbbVie's $2.1 billion acquisition of Capstan followed roughly $340 million raised. Capital-to-exit, round count, and dilution are not three separate facts — they are one chain. The capital the asset needs sets the number of rounds; the rounds set the dilution; the dilution sets what your stake is worth when the buyer finally arrives. Which means a thesis that promises "preserved ownership" is really a thesis that promises capital efficiency — the only way to hold ownership is to need fewer dollars, and therefore fewer rounds, on the way to the door.

You can defend your ownership with pro-rata follow-on, buying enough of each new round to hold your percentage flat. But follow-on is not free: it consumes reserve capital, and on a small fund every dollar of reserve is a dollar not deployed into a new shot on goal. Defending ownership and maximising shots are in direct tension, and the smaller the fund, the sharper the tension. Hold that thought; it returns when the portfolio enters the picture.

The power law: why the average deal is a fiction

Here is the part of venture math that is genuinely counterintuitive, and the part that breaks every analogy to public-market investing. Venture returns are not normally distributed. They do not cluster around a mean with a few outliers on each side. They follow a power law: a tiny number of investments generate essentially all of the returns, and the bulk of the portfolio clusters near zero. There is no "average deal" to reason from, because the distribution has no meaningful centre.

The data is consistent across every large study that has looked. Correlation Ventures, examining more than twenty-seven thousand financings, found that roughly sixty-five per cent of venture deals return less than the capital invested, and only about four per cent return more than ten times — with the top fraction of a per cent returning more than fifty times and dragging the whole average up by themselves. Horsley Bridge, an institutional limited partner, found across seven thousand investments that six per cent of deals generated sixty per cent of all returns. VenCap, looking at more than eleven thousand companies across two hundred and fifty-nine funds, found that about half lost money, that roughly one per cent of companies returned the entire fund that backed them on their own, and — the single most important sentence in venture portfolio theory — that ninety per cent of funds which returned three times or more contained at least one such fund-returner. It is, in their phrase, quasi-impossible to build a top-quartile fund without one stellar investment.

Sit with the consequence, because it reframes everything. The fund's return is not the sum of a lot of decent outcomes. It is, almost entirely, the contribution of its top one or two positions. A portfolio with no fund-returner does not produce a slightly worse result; it produces a structurally capped one, because the middle of the distribution — the 1× to 3× outcomes — cannot, by arithmetic, carry a fund against the dead weight of the sixty-five per cent that returned nothing. This is why a single-position multiple, like the 2.4× the India essay computed, is not a fund return. It is one draw from a distribution where most draws are zeros and the fund lives entirely in the tail.

This also dissolves the intuition that diversification is the answer. In a normal distribution, adding more bets narrows your outcome toward the mean, which is good when the mean is positive. In a power-law distribution, the mean is dominated by outliers you mostly will not catch, so adding more average bets does not converge you toward a good outcome — it just buys more lottery tickets in a lottery where the prize is concentrated in a handful of winners you have to actually pick. The portfolio's job is not to reduce variance. It is to make sure you are holding the tail when it pays.

Make it concrete with a portfolio you can audit. Spread $50 million across twenty-five equal $2 million cheques and let the outcomes fall the way Correlation Ventures says they fall:

Read the bottom row, then read the tail row, and notice they are nearly the same number. The single tail position returns the whole fund on its own; the other twenty-four deals together return roughly one-and-a-bit times the fund. Strike that one name and the fund collapses from 2.4× to about 1.2× — a mediocre result assembled from the very same twenty-four "good enough" outcomes. That is not a rhetorical flourish; it is the arithmetic VenCap found in the field, where ninety per cent of funds returning three times or more had at least one position that returned the whole fund. The portfolio does not earn the return. One position earns it, and the portfolio's real job is to make sure that position is inside it.

Biotech makes the tail more binary, not less

Everything so far is general venture math. Biotech bends it in a specific direction, and the direction is more extreme, because biotech adds a source of binary outcomes that software does not have: the clinical readout.

A drug does not fail gradually. It passes a phase or it does not, and the value of the company gaps up or craters on the day the data reads out. Citeline's phase-transition data for 2014 through 2023 puts the likelihood of approval for a drug entering Phase 1 at about 6.7 per cent — an all-time low — with the Phase 2 efficacy readout as the executioner: only around twenty-eight per cent of programmes clear it. Phase 1 and Phase 3 sit near coin-flips; Phase 2 is where roughly three of every four candidates die. Each clinical asset, in other words, is a near-binary lottery whose worst moment is a single efficacy readout, and the equity value tracks that lottery almost discontinuously. Biotech's per-asset distribution is therefore even more bimodal than software's: fewer middling outcomes, more total wipeouts, and occasional violent multiples.

Two structural features push back against that binarity, and they are exactly the two this publication has spent its other essays on. The first is the platform. A single-asset company is the coin flip; a platform spreads the binary across many shots from one engine, which is the whole reason validated platforms produce more assets per research dollar than single-asset shops, and the whole reason the recursive discovery factory — the platform whose marginal cost of the next shot falls as it runs — is the architecture that survives. A platform does not change any single readout's odds; it changes how many independent readouts you own, which smooths the company-level outcome even as each asset stays binary.

The second is the exit timing. A biotech does not have to ride its asset to the approval lottery. The strategic acquirer — pharma buying a de-risked Phase 1 or Phase 2 asset to fill a patent cliff — routinely takes the company off the board before the worst binary resolves, and it does so at the mid-stage point where the company has raised only two or three rounds. That single fact links the two halves of this essay: the same early exit that truncates the clinical downside also truncates the dilution path, which is why the strategic mid-band exit is the structurally privileged outcome in biotech. It is more binary at the asset level and less diluted at the company level, and those two forces fight. But neither the binary nor the dilution sets the number on the exit. The buyer does — and the buyer is not a mystery.

What the buyer is paying for

The exit term in the identity is the only one set by someone other than you, and it is worth knowing who. The buyer of a biotech is, overwhelmingly, a large pharmaceutical company, and large pharma is not buying on sentiment. It is buying against a deadline already on the calendar.

Between 2025 and 2030, roughly $300 billion of branded drug revenue loses patent exclusivity — about one-sixth of the industry's annual top line, concentrated in some two hundred drugs and around seventy blockbusters, and three times the size of the 2016 cliff. The keystone is Keytruda: Merck's roughly thirty-billion-dollar-a-year cancer immunotherapy, more than half the company's revenue, with its core US exclusivity falling in 2028. Bristol Myers Squibb faces the steepest proportional cliff — Eliquis and Opdivo together are nearly half its revenue, a growth gap estimated near $38 billion. When a small-molecule blockbuster loses exclusivity, eighty to ninety per cent of its revenue can evaporate inside a year; biologics erode more slowly but arrive at the same place. This is not a soft headwind. It is a set of dated, multi-billion-dollar holes that have to be filled, and the fastest way to fill them is to buy something already de-risked.

The cliff is worth seeing with names attached, because the abstraction "$300 billion" hides which specific franchises are falling and onto whom:

Read down the right-hand column and the back half of the decade is a wall: roughly forty-nine billion in 2027 and sixty-seven in 2028, dominated by single franchises — Eliquis, then Keytruda — each larger than most companies' entire pipelines. These are the holes. The rest of this part is about the bar an asset must clear to be allowed to fill one.

That is the demand under every biotech exit, and the blockbuster math explains why pharma pays what it pays. A blockbuster is any drug clearing a billion dollars a year. Run that to a daily rate and it sharpens: a billion-dollar drug earns roughly $2.7 million for every day it holds exclusivity, and Keytruda, at thirty billion, earns on the order of $80 million a day. Against a clock like that — and an effective post-launch patent life of only about a decade, once development has burned the rest of the twenty-year term — an asset that can plausibly become a two-to-three-billion-dollar franchise is worth ten billion dollars to a company staring down a thirty-billion-dollar hole. The acquisition price is not the question; the revenue it defends is. This is why pharma pays a premium for cadence — the rate at which an engine throws de-risked, clinic-ready candidates — and it is why the venture investor's exit exists at all.

And the buying is not opportunistic — it is the engine. By McKinsey's accounting, more than seventy per cent of new-molecular-entity revenue since 2018 has come from externally sourced products: molecules pharma in-licensed or acquired rather than discovered in-house. Internal R&D has not kept pace with the cost of attrition, and the firms best at sourcing earn between 3.4 and 8.2 times the return on those external assets that their peers manage. The practical meaning of that statistic is the thing a founder or an early investor should internalise: the venture-backed startup is now where big pharma's pipeline actually comes from. You are not hoping to be noticed by a buyer who might prefer to build it himself. You are building the supply for a buyer whose own labs can no longer feed his revenue. That is what makes the exit reliable enough to underwrite in the first place.

The buyer's hurdle rate

Knowing that pharma buys, and against what, still leaves the sharpest question unanswered: how large does an asset have to be before a buyer shows up at all? Pharma does not acquire everything that de-risks — it acquires above a threshold, and below that line a company is effectively unsellable to the majors no matter how good its science. That threshold is computable. It lands around $2.8 billion in peak annual sales, and it is the bar every exit in the inversion below is quietly clearing. The derivation that follows is directional — laid out publicly by the analyst Andrew Pannu and re-run here in this essay's terms — and every input is a round number, because the point is the shape, not the decimal.

Start with why pharma buys rather than builds: operating leverage. A large pharmaceutical company is one of the most operationally geared businesses in the economy. Strip out the two GLP-1 outliers, Eli Lilly and Novo Nordisk, whose obesity-driven growth distorts every average, and across the other eight of the top ten a roughly 3.5 per cent rise in revenue translated into about an 8.2 per cent rise in net income — profit moving more than twice as fast as the top line, in both directions. The expensive machinery — global R&D, plants, regulatory affairs, and above all a standing sales force — is largely fixed, so the cheapest incremental growth is not a new capability but more volume pushed through the engine that already exists. And the slow growers are most of the group: with the GLP-1 pair removed, Johnson & Johnson's pharma arm grew 1.5 per cent in 2025, AbbVie 1.0, Novartis 1.6, Roche 1.9, Merck 3.9. These are enormous, slow-compounding annuities whose profit still jumps when revenue ticks up. A company growing its top line at one or two per cent while its valuation demands high-single-digit profit growth has exactly one lever it fully controls — add volume to the existing engine — and it cannot invent that volume fast enough internally. So it buys it. The market has already scored the game: the GLP-1 names trade near fifteen and eleven times net sales, the other eight cluster between four and six, and Pfizer — the most cliff-exposed major — trades at about 2.4 times, which is not cheapness but the market pricing revenue it expects to melt faster than the company can replace it.

Now size the appetite. Those eight companies, ex-GLP-1, carry an aggregate market capitalisation around two trillion dollars, and a market cap is a promise about growth; to roughly defend valuations like these the group needs to grow on the order of seven and a half per cent a year. Seven and a half per cent of two trillion is about $150 billion of new value the group must manufacture annually just to stand still against its own multiple. Some comes free, from existing products beating forecast — call it a fifth — leaving roughly $120 billion that must come from new assets. Value is bridged to revenue by a multiple, and the multiple is where it sharpens: a blended pharma trades near four-and-a-half times sales, but that blend hides a split — mature, eroding products are worth perhaps two times sales, while high-growth assets (new launches, recent acquisitions, anything compounding above ten per cent) carry far more. Back the mature bucket out at two times and the implied multiple on the high-growth bucket lands around nine times. So $120 billion of required new market value, at roughly nine times sales, needs about $13 billion of genuinely new revenue every year — the growth half of the bar.

Add the cliff as a recurring burden. The losses-of-exclusivity table above runs to some sixty-seven billion dollars by 2028, and at the current brutal running rate the cliff is taking roughly thirty-six billion of revenue off the table each year — but that is a peak; the long-run historical average is closer to ten billion. Normalise to roughly fifteen billion of recurring replacement, and the two halves combine: about $13 billion of new revenue to deliver growth, plus about $15 billion to replace what lapses, is roughly $28 billion of new revenue the group must source every single year.

Then divide by the launch rate. Revenue arrives in discrete units — approved drugs — and the recent record of FDA approvals across the top companies runs to about ten meaningful new products a year. Twenty-eight billion dollars spread across ten launches is $2.8 billion of peak sales per asset. That is the hurdle: for a new drug to be worth the acquisition, the integration, and the commercial attention of a major, it has to be able to reach roughly two-point-eight billion dollars in peak annual sales — not on average across a portfolio, but as a plausible ceiling, per asset, because that is the unit the income statement is built to absorb. Here is the whole chain in one place:

Every input is round on purpose, and the answer is robust to moving any one of them: halve the growth target and the cliff burden still puts the hurdle near two billion; use the $36-billion cliff rate and it climbs past four. There is no sane set of assumptions under which a sub-billion-dollar asset clears it. The hurdle is not a precise figure; it is a floor, and the floor is high.

Merck is the whole argument compressed into one company. It grows revenue under four per cent; only about fourteen per cent of its sales come from high-growth products; and its single largest piece — Keytruda, near thirty billion a year, more than half the company — loses US exclusivity in 2028. Its recent approval cadence rounds to about one new drug a year. Slow growth, a thin new-product engine, and a thirty-billion-dollar hole on a fixed date: there is no internal path out of that arithmetic, only an external one. Merck cannot be patient about the hurdle rate — it has to find several two-to-three-billion-dollar assets and buy them, soon, and every counterparty knows it. When people ask why a pharma "overpaid," the answer is usually a version of Merck's clock.

Two things the hurdle rate explains bear directly on everything below. The first is pharma's obsession with mega-markets: after discounting for competition, partial response rates, and the share a new entrant can realistically take, the only therapeutic areas that can yield a $2.8-billion asset are the genuinely enormous ones — oncology, immunology, cardiometabolic disease, the obesity frontier — which is why capital herds into the same few TAMs and then competes its own returns away. The second is the one the inversion has to reckon with: a company whose lead asset peaks below roughly a billion dollars is, to a major, structurally unsellable — not because the asset is bad but because it does not move the needle against the bar. Such a company has only two routes to a big-pharma exit — fit so cleanly onto existing commercial infrastructure that the cost to absorb it rounds to zero, or add scale until it does move the needle. Absent one of those, "we will be acquired by pharma" is not a plan but a hope the buyer's own model contradicts. Hold that, because the India inversion below is staked on a mid-band exit, and the hurdle rate is about to ask who, exactly, is paying for it.

How pharma pays

The buying has bifurcated, and the structure tracks exactly how much risk is left in the asset. At the late, de-risked end, the deal is a clean cash premium. When GSK agreed in June 2026 to buy Nuvalent for $10.6 billion — all cash, a forty per cent premium, its largest acquisition in eight years — it was buying two lung-cancer inhibitors already under FDA review, weeks from their approval decisions. There is little contingency in a deal like that because there is little risk left to share: you pay up front for near-certainty and for the speed of plugging the hole. GSK is a clean illustration of the whole pattern — it exited oncology in its 2014 swap with Novartis and is now rebuilding the franchise by acquisition, on the cliff's clock.

At the early, unproven end, the deal inverts into a milestone ladder — the structure that has quietly become the default. A small upfront payment buys the option; the bulk of the headline is contingent "biobucks," paid only as the asset clears clinical, regulatory, and commercial gates, with royalties layered on top. The China out-licensing wave is the clearest catalogue of the form: Pfizer paid 3SBio $1.25 billion upfront against up to $4.8 billion in milestones for a bispecific; GSK committed $500 million upfront to Hengrui in a package reaching toward twelve billion; AstraZeneca's deal with CSPC could reach the high teens of billions if everything pays. The press release quotes the sum of the upfront and every milestone. The cash that actually changes hands at signing is the small number at the front.

The same architecture runs through the Western dealmaking that opened 2025 — the "string of pearls" of focused mid-cap acquisitions announced in a rush at the J.P. Morgan conference. Lilly bought Scorpion's mutant-selective PI3-kinase programme for up to $2.5 billion; GSK took IDRx for up to $1.15 billion; Sanofi acquired the commercial-stage rare-disease company Blueprint for $9.1 billion in cash — and even that near-all-cash deal carried a contingent value right worth up to several hundred million more. The cleanest illustration of the structure, though, is Novartis's purchase of Anthos: a headline value of up to $3.1 billion, of which only $925 million was paid upfront and as much as $2.15 billion was contingent on regulatory and sales milestones. Read that split as the general rule — roughly two-thirds of the "price" is an option the buyer exercises only if the asset clears its gates — and note that the contingent fraction shrinks as the asset de-risks. A Phase 3 buyback like Anthos is heavily milestoned; a commercial-stage asset like Blueprint is mostly cash; an asset already at the FDA's door, like Nuvalent, is cash in full. The deal structure is a readout of how much risk is left in the molecule.

This is the milestone turn, and it is the same move as pharma learning to behave like a venture fund: explicit allocations across modalities and stages, terms that read like preferred-equity ladders, post-deal support that looks like a fund's value-add, and a portfolio mindset that prices each asset as a probability-weighted option rather than a purchase. As buyers have competed for the best early assets the upfront fraction has crept up — average China-deal upfronts roughly tripled between 2022 and 2026 — but the architecture holds: the riskier the asset, the more of its value is pushed into the future and made contingent.

For venture math, the milestone turn forces one correction that most cap-table arithmetic ignores. The exit value in the identity is not the headline biobucks number. It is the upfront plus the risk-adjusted, time-discounted value of the milestone ladder — and since most of that ladder is gated on the same clinical hurdles that kill roughly seventy per cent of programmes at Phase 2 alone, the realised expected value of an early-stage "two-billion-dollar deal" can be a fraction of two billion. An investor who marks the headline is marking a fiction. The realised exit term is therefore stage-dependent in a way the clean model hides: a late-stage acquisition delivers most of its value as cash now — the Nuvalent shape — while an early-stage out-licensing delivers a probability-weighted trickle that may never fully arrive. So when the inversion below uses a clean four-hundred-million-dollar mid-band exit, read it as the risk-adjusted, cash-equivalent realisation, not the press-release figure — and note that the India mid-band thesis has to price that contingent haircut, because mid-band strategic deals are precisely the ones that come milestone-structured.

Work the haircut once and it stops being abstract. Take the Anthos shape: $925 million paid on close, $2.15 billion in milestones spread across a regulatory approval and a ladder of sales thresholds. The upfront is certain and counts at face. The milestones are not — even a Phase 3 asset has gates left, and the sales tranches sit years out and depend on uptake. Apply a generous one-half realisation to the regulatory milestone, a steeper discount to the commercial ones, and discount the surviving stream back at a venture cost of capital, and the $3.1 billion headline resolves to something nearer $1.7–2.0 billion in expected, present-value terms. Run the same arithmetic on an early asset — a Phase 1 programme out-licensed on biobucks, where most of the ladder is gated behind the Phase 2 readout that kills seventy per cent of candidates — and a "$2 billion deal" collapses to a few hundred million of expected value. The headline is the ceiling of a distribution; the honest mark is its mean. An investor who books the ceiling is not optimistic. He is wrong.

The same cliff that creates the buyer also creates a second, simpler game off the identical calendar. When a blockbuster's exclusivity lapses, its revenue does not vanish — it transfers to whoever can manufacture the generic or biosimilar fastest and cheapest, which is the commoditised-end play India has run for decades. The semaglutide race is the live example: as Novo Nordisk's GLP-1 patents lapse market by market, Sun Pharma, Dr Reddy's, Natco, and Zydus are lined up for the generic opportunity. One cliff, two businesses — buy the novel asset that fills the hole, or make the cheap copy of the drug that fell into it — and the venture math of the first only makes full sense once you can see the calendar driving both.

Now the inversion, properly

The India essay asserted that the same exit inverts on fund size. Hold every variable from the previous sections in your hands at once, and you can finally see not just that it inverts but by how much, and under what conditions — which is the part that was missing.

Take two funds looking at the same opportunity set. The US fund is $300 million. The India fund is $50 million. Assume both are early, both get diluted on the same schedule, and both are trying to clear a 3× gross return — the rough bar for a credibly good fund.

The US fund needs $900 million back. The power law says that money has to come from a fund-returner, not a spread of decent exits. But here the platform-commoditisation ceiling bites with full force: discovery-layer platforms cap in the one-to-three-billion-dollar band, and six per cent of a $2 billion outcome — the diluted stake on a company that ground through a full financing path to get there — is $120 million, a 0.4× contribution against the fund. A capped discovery exit cannot be the fund-returner. To return a $300 million fund, the US investor needs the rare company that escapes the ceiling entirely: a vertical integrator that becomes a $10-billion-plus public company, where even a diluted six per cent is $600 million. The US fund is structurally forced to underwrite for the tail the ceiling makes rarest. That is the death sentence the India essay named, now shown in its mechanism: the fund needs the tail, and the modality the cohort is selling has had its tail amputated.

The India fund needs $150 million back, and the arithmetic of the mid-band is completely different on a $50 million base. A capital-efficient company that exits in the strategic mid-band — call it $400 million, reached in two rounds so the origination stake survives at roughly fifteen per cent — returns fifteen per cent of $400 million, which is $60 million. That single mid-band exit, the kind that returns essentially nothing to the US fund, returns 1.2× of the entire India fund. Two or three of them return the fund outright, with no fund-returner required. And if a tail does arrive, the small denominator turns even a capped outcome into a fund-maker: a $2 billion exit at a diluted eight per cent is $160 million — 3.2× of the $50 million fund, from one position, off an outcome that returned the US fund a fractional loss.

Run both as full portfolios, not single positions, and the inversion stops being a slogan. Give each the same twenty-five-name structure and the same brutal distribution; only the cheque size, the exits each fund can live on, and the denominator move.

The US fund — $300M, needs ~$900M for 3×:

The India fund — $50M, needs ~$150M for 3×:

The two tables are the whole argument. For the US fund the mid-band column is a rounding error and survival rests on the single rarest outcome in the set — the $10-billion integrator the commoditisation ceiling makes vanishingly unlikely. For the India fund the mid-band column is the return, and a tail is a bonus it never needed. But notice the India table's own knife-edge: it clears 3× on two or three mid-band exits and limps to 1.9× on one. Its survival does not depend on catching a giant; it depends on the mid-band machine firing two or three times a vintage. That is a lower bar than the US fund's — but it is still a bar, and it is the one the whole thesis is staking itself on.

So state the inversion at its true strength, with the false version stripped out. The India fund does not escape the power law — nothing escapes the power law, and ninety per cent of good funds still have a fund-returner, Indian or not. What the small denominator does is lower the bar at which the power law pays. The mid-band, which is a rounding error for the US fund, is a fund-fraction-returner for the India fund. The tail, which the US fund desperately needs and the commoditisation ceiling denies, the India fund collects as upside it did not require to survive. The US fund must catch the rarest outcome in the distribution to live; the India fund lives on the median outcome and treats the tail as a bonus. Same distribution, same dilution, same exits — and opposite survival conditions, because the denominator decides what counts as a win.

The dilution layer sharpens this rather than complicating it. The India advantage is two effects stacked, and they should be kept separate because they are not equally robust. The first is the denominator effect — pure fund size — which is unconditional: it holds even at identical ownership, and it is the version the India essay was careful to claim. The second is the ownership effect — that capital-efficient, fewer-round companies preserve the origination stake — which is real but conditional on the mid-band exit actually arriving early. When both fire together, the same exit returns more to the India fund because the fund is smaller and the stake is larger. When the second fails — when the "capital-efficient" company in fact needs five rounds to reach a mid-band outcome — the ownership advantage evaporates and only the denominator effect remains. Honest underwriting prices the denominator effect as the floor and the ownership effect as the upside.

Who buys in the gap

The inversion is staked on the mid-band exit — the $200-to-$700-million strategic acquisition that returns a fraction of a fund on a $50-million base — and the hurdle rate has just made that assumption uncomfortable. If an asset's peak-sales ceiling sits below a billion dollars, the buyer's own arithmetic rules out the global top ten; and if the ceiling clears $2.8 billion, the asset is not a mid-band company at all but a blockbuster-in-waiting that will command far more than mid-band money once its data matures. So the mid-band exit, to be real, has to come from somewhere specific, and the thesis owes a name. There are four, and a fund underwriting mid-band outcomes should be able to say which it is counting on, asset by asset.

The first is the mid-cap consolidator — Jazz, Ipsen, Supernus, Lupin, a Sun Pharma reaching upmarket — whose own hurdle is a few hundred million in peak sales rather than a few billion, and for whom a sub-blockbuster asset is genuinely needle-moving. These are a different population from the string-of-pearls acquirers above, and they are the natural buyers of the $200-to-$700-million company; the trade-off is that they pay less, and often in cash they do not comfortably have.

The second is the region-rights licence rather than a full takeout: an asset licensed for ex-US or ex-India territories to a regional commercial player, monetised in pieces rather than sold whole. This is the dominant shape of the China out-licensing wave catalogued above, and there is no structural reason an Indian asset cannot be carved the same way. It is the most realistic mid-band monetisation for a capital-efficient origination fund — and, exactly as the milestone turn warned, it arrives risk-adjusted and milestone-laden, not as a clean cheque.

The third is the roll-up: aggregating several sub-scale assets until the combined entity finally clears a major's bar — the "add scale or stay home" route the hurdle rate forces on no-man's-land companies, and the logic behind the recent run of mid-cap mergers of equals.

The fourth is the asset that was never really mid-band — a genuinely blockbuster-ceilinged molecule caught early and cheap, whose mid-band entry price reflected stage and geography, not ceiling. When its data matures it exits above the band, to a top-ten buyer at a premium. That is the single most lucrative outcome an India fund can have and the rarest, and the discipline is not to confuse it with the other three: it is a tail dressed as a mid-band, and it is caught by science diligence, not by exit-market hope.

Name which of the four, and the mid-band thesis is underwritable. Leave it unnamed, and "we exit in the mid-band" collapses back into the missing-buyer hope the hurdle rate exists to expose. The hopeful reading is that the gap itself — the whole population of valuable, sub-blockbuster opportunities a $2.8-billion-hurdle giant structurally cannot chase — is precisely the terrain a small, capital-efficient fund is built to farm. The denominator inversion was about a small fund needing a smaller exit to win; the hurdle rate hands it the hunting ground. A sub-$1-billion-peak asset that is a non-event for a major can be a perfectly good independent company, a regional licence, or a mid-cap takeout — outcomes a $50-million fund can build a return on and a $300-million fund cannot be bothered to chase. The hurdle rate does not just threaten the India thesis; it tells the India fund where to hunt.

Time is the variable the multiple hides

Everything to here has been counted in multiples, and multiples quietly lie, because there is no clock inside them. A 3× returned in four years and a 3× returned in ten are the same MOIC and completely different investments. The number an LP actually optimises is the internal rate of return, and IRR is just a multiple with time put back in: a 3× over four years is roughly a 32 per cent IRR; the identical 3× over ten years is about 12 per cent. One is a top-decile fund. The other barely clears the public index the LP could have held with no ten-year lock-up.

In biotech this is not a footnote, because biotech is slow and the science sets the schedule, not the fund. The median venture-backed company that reaches an exit takes something like six years to get there, and one that rides its asset toward approval can take far longer. The J-curve — the years a fund sits underwater paying fees and writing follow-on cheques before any cash comes back — is deeper and longer here than in software, so a biotech fund that returns 3× in year nine has earned a worse IRR than a software fund that returned the same 3× in year five, with an identical headline multiple.

Now lay the inversion over the clock, because time sharpens it exactly the way the denominator did. The US fund's survival outcome — the ten-billion-dollar vertical integrator — is not only the rarest, it is the slowest: building a company to a ten-billion public valuation is a ten-to-fifteen-year project, so even when the US fund catches its tail, it catches it late, and a late multiple is a poor IRR. The India fund's survival outcome — the mid-band strategic acquisition — is structurally earlier: it lands at the Phase 1/2 inflection, in year four or five, before the long grind. The same fact that preserves the India fund's ownership (fewer rounds) also shortens its time to cash (earlier exit), and sixty million dollars back in year five is a far better IRR than the same sixty million in year ten. So the India fund's edge is not two effects but three, stacked: a smaller denominator, a larger retained stake, and a shorter clock. The denominator decides whether the exit is a win; the clock decides whether the win is any good; the mid-band wins on both.

The flip side is the same conditional that haunts every lever in this essay. A mid-band machine that delivers in year eight instead of year four keeps the multiple and loses the IRR — and an India fund whose exits arrive late is just a small US fund with a worse cost of capital. Time is a multiplier on the thesis when the exits are early and a penalty when they drift; it is not a free advantage, it is a reason capital efficiency and exit speed are the same discipline.

Where the model breaks

A model you cannot break is a model you do not understand. Here are the four places this one fails, in the order an LP would press them.

The mid-band has to clear. The entire India engine rests on the assumption that capital-efficient companies actually get acquired in the $200-to-$700-million band, routinely, on a frequency that lets two or three hits return a fund. If that exit market is thin — if Indian-originated assets do not get bought at that size with that regularity — then the "1.2× per hit" engine stalls, and the fund is left needing a tail it was not built to catch. This is the single load-bearing empirical assumption, and the hurdle rate gives it a sharper form: the "Who buys in the gap" section named the only four buyers the buyer's arithmetic permits, and the thesis is exactly as strong as a fund's ability to say which of the four it is underwriting, asset by asset. Assert the mid-band exit without naming its buyer and you have assumed away the hardest part — which is why this remains the assumption with the least direct evidence and the likeliest reason the thesis is wrong.

The ownership effect is fragile. It survives only if the company reaches its exit in two or three rounds. The moment a company needs the full financing path — because the science took longer, or the capital efficiency was a pitch rather than a fact — the origination stake decays to the same high-single-digits the US fund holds, and one of the two India advantages is gone. Capital efficiency is the precondition, not a nice-to-have, and it is harder to engineer than to assert.

The power law still binds. The India fund's honest claim is "returns its capital reliably on the band, with the tail as the differentiator" — not "beats the US fund outright." Returning a fund on mid-band exits keeps you alive and probably above water; being top-quartile still almost certainly requires a fund-returner, because that is true of ninety per cent of top-quartile funds regardless of geography. The small denominator lowers the bar; it does not remove it. A fund manager who tells LPs the mid-band alone delivers top-decile returns is selling the same fiction in a new currency.

And reserves fight the portfolio. Defending ownership through pro-rata is how you stop the dilution decay, but every reserve dollar is a shot not taken, and the power law rewards shots — more independent draws at the tail. On a $50 million fund the tension between concentrating to defend winners and diversifying to catch them is sharper than on a $300 million fund, and there is no clean answer; it is a judgement that has to be re-made every quarter against which companies are actually pulling away.

The supply side: is the money even building hurdle-clearing assets?

There is a mirror to the buyer's hurdle on the side of where venture capital actually flows, and it confirms the same arithmetic from the opposite direction. In 2024, US biopharma therapeutics and platforms raised roughly twenty billion dollars across about two hundred and sixty rounds, and the money concentrated exactly where the hurdle rate predicts: oncology took close to half of it, with metabolic and immunology next — the mega-TAMs, the only places a $2.8-billion asset can plausibly live. Capital is not naive about the hurdle; it crowds toward the indications large enough to clear it, which is also why those indications are the most competitive and the returns hardest to keep. The herding the power-law section warned about is the hurdle rate expressed as a funding pattern.

Two structural facts sharpen the picture. The first comes from the deal data: platform and data-technology companies, when they are bought at all, are bought on heavily milestone-weighted terms — in one recent stretch, under a billion dollars of upfront cash against tens of billions of headline value — which is precisely the buyer pricing the distance between a platform's promise and a hurdle-clearing product. A platform is not a $2.8-billion asset; it is an option on one, and pharma pays for options in contingent paper, exactly as the milestone turn described. The second is the scarcity argument made by Atlas Venture's Bruce Booth: US biotech startup formation in early 2025 fell to its lowest level in a decade, and counterintuitively that is healthy, because a thinner cohort is a higher-quality one and the field's best names — Alnylam, Nimbus, Kymera — were repeatedly born in exactly these down cycles. Fewer shots, better aimed, into the few markets big enough to matter is not a contraction of the model; it is the model working.

For an India fund, the supply side is where the capital-geometry argument finally touches the ground. India's venture market reached roughly sixteen billion dollars in 2025; deep tech climbed to about fifteen per cent of all venture-and-private-equity activity from four per cent a decade earlier; fund-raising into the asset class roughly doubled to about $5.4 billion; and the government anchored a roughly $1.1-billion deep-tech fund-of-funds. The scaffolding a small, disciplined biotech fund would need is, for the first time, being built. But the buyer's hurdle is the discipline that scaffolding must be deployed under. A fund that hunts in the sub-blockbuster gap the majors structurally cannot enter is a coherent thesis with a named buyer at its end; a fund that hopes to sell sub-scale assets to the majors is the missing-buyer hope with a tricolour painted on it. The hurdle rate is indifferent to where the fund is domiciled. It cares only about peak sales.

The close

The venture math of biotech has two halves, and they meet at one number. The seller's half reduces to four facts: dilution is a predictable tax whose size is set by the round count, not the negotiation; returns follow a power law, so the fund lives in its top one or two positions and there is no average deal to reason from; biotech makes the per-asset outcome more binary still, which the platform smooths and the early strategic exit truncates; and time is the multiplier the multiple hides, so the same exit is a triumph in year four and a disappointment in year ten. The buyer's half reduces to one: a major will not acquire an asset that cannot plausibly reach roughly $2.8 billion in peak sales, because below that line the purchase does not move an income statement built on operating leverage and ground down by a patent cliff. And the two halves share a single variable — the asset's peak-sales ceiling — which simultaneously sets what the exit is worth to the fund and whether the exit happens at all.

That shared number is what makes the inversion both true and conditional. The denominator decides what counts as a win, so the same exit that is a rounding error for a $300-million fund can be a fund-returner for a $50-million one — not because the small fund is smarter, but because it changed the only variables in the identity it controls. But the exit only exists if a buyer clears the hurdle, and for the sub-blockbuster assets the small fund lives on, that buyer is rarely a major — so the inversion holds only where the fund hunts in the gap the hurdle rate leaves open and can name the buyer at the other end. The small fund does not repeal the power law or the hurdle rate; it lowers the price of admission to the first and learns to hunt in the shadow of the second.

So the inversion is discharged, and it is more modest than the triumphant version — arithmetic rather than optimism, with its conditions written on its face. Whether the Indian version of the fund actually clears — whether the mid-band exits arrive, whether the ownership survives, whether the named buyer shows up, whether one tail appears per vintage — is not a question the math can answer. The math only tells you what has to be true. The next three years of Indian biotech exits tell you whether it is. And the deepest of those conditions is the one the buyer's half added: most of the value destroyed in this industry is not destroyed in the lab. It is destroyed by building, beautifully, something that clears every scientific bar and not the one commercial bar that decides whether anyone has to own it.

This essay discharges the IOU in "Modality Commoditization and India's Right to Win," which defers its fund-inversion claim to this capital-math treatment. Return-distribution benchmarks are drawn from Correlation Ventures, Horsley Bridge, and VenCap; dilution benchmarks from Carta's 2025 cap-table data; clinical phase-transition rates from Citeline (2014–2023); time-to-exit (~6-year median) and capital-to-exit benchmarks (Orna ~$320M, Capstan ~$340M raised pre-acquisition) from Crunchbase and venture-exit analyses. The external-innovation figures (>70% of NME revenue externally sourced since 2018; outperformers earning 3.4–8.2× on sourced assets) are from McKinsey. The blockbuster and patent-cliff figures (the ~$300B 2025–2030 wave, the named loss-of-exclusivity ladder through 2028, Keytruda's 2028 loss, BMS's exposure) are from Evaluate, DrugPatentWatch, and company disclosures; the deal examples — GSK–Nuvalent ($10.6B, June 2026), the 2025 "string of pearls" (Sanofi–Blueprint, Lilly–Scorpion, Novartis–Anthos, GSK–IDRx), and the China out-licensing wave (Pfizer–3SBio, GSK–Hengrui, AstraZeneca–CSPC) — from primary announcements and trade reporting. The buyer's-hurdle derivation (operating leverage, the growth gap, the ~9× high-growth sales multiple, and the ~$2.8B peak-sales hurdle) follows a public walk-through by the analyst Andrew Pannu (@andrewpannu), re-expressed and extended here, with the operating-leverage, market-cap, multiple, and FDA-approval figures from his compiled tables and public disclosures. The supply-side data are from DealForma (2024 US biopharma venture totals), Bruce Booth / LifeSciVC (the 2025 startup-scarcity thesis), and Bain & Company's India Venture Capital Report 2026 (India VC ~$16B in 2025; deep tech ~15% of VC-PE; fundraising ~$5.4B; the government's ~$1.1B deep-tech fund-of-funds). The fund-mathematics worked examples are illustrative models, not audited comparables; they are built to show the geometry, and every input is a median a reader can move. The identity is the spine; the power law is the constraint; the hurdle rate is the gate; the denominator is the verdict — and peak sales is the one number they all turn on.