In April, Summit Therapeutics shed 24% of its market cap in a single session. No exploit. No depeg. No governance attack. The cause was an interim analysis that missed an early efficacy bar on ivonescimab, a PD-1/VEGF bispecific antibody. One readout. One trading day. Twenty-four percent.
Here is the anomaly I keep returning to: a single clinical data point can reprice roughly five billion dollars of enterprise value, and there is no on-chain venue that prices it at all. China's outbound drug licensing volume crossed $120 billion this year, up 36% year over year. Those deals close in law firms, not ledgers. Upfronts and milestones clear through commercial bank escrow. Royalties route through Delaware and Dublin holding structures, not through a settlement layer.
The DeSci sector has spent four years promising to put exactly this asset class on-chain. IP-NFTs. Tokenized royalties. Milestone escrows written in Solidity. On paper, a $120 billion pool of binary, event-driven payoffs is the ideal substrate for programmable money. On inspection, the architecture does not compile.
The distance between the biotech trade and the on-chain thesis is not a marketing gap. It is an oracle problem, a legal-wrapper problem, and a settlement problem โ and they compound in that order.
Context: What Is Actually Being Sold
Strip the Goldman framing and the mechanics are simple. Big pharma faces a patent cliff: roughly $440 billion of branded revenue loses exclusivity before 2032. The industry needs replacement molecules, and it needs them faster than internal research can deliver. China, historically the world's supplier of active pharmaceutical ingredients and intermediates, now exports finished novel drugs.
The numbers are not subtle. Year-to-date, licensing deals for Chinese-origin assets have exceeded $120 billion in headline value, up about 36% year over year. China's five-year pharmaceutical plan, published on September 18, sets two directional targets: at least 25% of global first-in-class drugs by 2030, and 50 domestic companies with annual revenue above 10 billion RMB. AstraZeneca placed a $2 billion investment into Summit Therapeutics. Goldman analyst Richter framed China as both a sourcing ground for multinationals and a potential competitor that compresses product cycles.
The anchor asset is ivonescimab. Akeso originated it. Summit licensed it in December 2022 for roughly $500 million upfront against a total headline package in the $5 billion range. Mechanically, it is a single molecule that blocks PD-1 and VEGF simultaneously โ immune checkpoint plus anti-angiogenesis in one construct, designed to capture the efficacy of a combination while avoiding the additive toxicity of co-administered drugs.
The data point that matters: HARMONi-2, a head-to-head Phase III in first-line PD-L1-positive non-small-cell lung cancer, reported a progression-free survival hazard ratio around 0.51 for ivonescimab monotherapy versus pembrolizumab monotherapy. That is the first time a PD-1-class agent has beaten Keytruda head-to-head on PFS in a Phase III. The global confirmatory trial, HARMONi-3, reads out squamous data in the second half of 2026. Keytruda is a roughly $25 billion annual franchise. First-line NSCLC is its core.
Now place that next to crypto. A binary, dated, high-information event โ a readout โ that moves billions of dollars. This is the trade DeSci said it would host. This is the trade that, so far, has not moved an inch on-chain.
Core: Where the On-Chain Version Breaks
I want to be precise about the failure modes, because the DeSci pitch is coherent enough that it deserves a real audit rather than a slogan.
Failure mode one: the clinical oracle.
A smart contract cannot observe a Phase III readout. There is no trustless data feed for "did HARMONi-3 hit overall survival in the squamous cohort." The result is generated by the sponsor, locked in a statistical analysis plan, reviewed by a data monitoring committee, then disclosed in a press release and a conference abstract. The only parties who can attest the truth are the sponsor and the regulators. That is a centralized oracle with a single reporter.
DeSci projects paper over this with optimistic attestations, multisig committees, or "verified" research marketplaces. All of them collapse the same way: the attestation is only as credible as the entity signing it, and the entity signing it has a direct financial interest in the outcome. Compare this to a price oracle, where an aggregator pulls dozens of independent venues and manipulation becomes expensive. Here there is one venue, one reporter, and the reporter owns the asset. That is not an oracle. That is a press release with a signature.
When I audited a tokenized-royalty project in 2024, the milestone escrow logic was clean. The issue was upstream. The contract released tranches on a boolean flag named milestone_met, and the flag was set by an admin key controlled by the issuer. The bytecode didn't lie โ it did exactly what it said. But what it said was "trust the issuer," dressed in the syntax of trustlessness. That is the DeSci pattern in miniature. A perfectly auditable function that verifies nothing.
Failure mode two: the legal wrapper.
A drug royalty is a contractual cash-flow right, governed by the law of a specific jurisdiction, denominated in fiat, and enforceable only in a court that recognizes the assignment. Tokenizing it does not move the cash-flow on-chain. It creates a token whose value depends on an off-chain legal claim that the token holder may or may not actually own.
This is the part DeSci marketing skips. When you hold a tokenized royalty, you usually do not hold the royalty. You hold a token issued by a special-purpose vehicle that holds the royalty. Your enforcement path runs through the SPV's governing documents, not through the token contract. If the SPV is structured in a jurisdiction that does not recognize token holders as beneficial owners, the token is a claim on a claim. Two layers of abstraction between you and the cash, and the second layer is a law firm in a time zone you cannot reach.
I spent part of 2024 auditing a Layer 2 for MiCA compliance, reviewing 200-plus contract functions to confirm KYC/AML logic sat at the protocol level rather than the gateway. The lesson generalizes: regulatory-aware architecture is not a feature you bolt on. It is a constraint that determines what can be built at all. A tokenized drug royalty must answer, at the contract level, who the beneficial owner is, which regulator has jurisdiction, and how a court order to freeze the asset would be executed. Most DeSci contracts answer none of these. They are not illegal. They are just unenforceable in the direction that matters.
There is a second-order problem. Clinical trial data that crosses borders is governed by human genetic resources rules in China and by privacy regimes elsewhere. A global Phase III generates patient-level data that cannot simply be piped into an oracle. Even the inputs to a milestone verification may be legally constrained from leaving the jurisdiction. The contract cannot read what the law will not let it see.
Failure mode three: settlement.
Even granting a credible oracle and a clean legal wrapper, the cash itself is fiat. Upfronts clear in dollars. Milestones clear in dollars. Royalties accrue quarterly from distributors and get remitted in dollars. To get that value to a token holder, someone has to convert fiat to stablecoins and push it through a bridge or a custodian. Each hop adds latency, counterparty risk, and a compliance surface.
Here is the structural point that the RWA boom obscures. Tokenized Treasuries worked โ and they did work โ because the underlying asset is already a settlement instrument inside a system with deterministic, near-instant finality at the institutional layer. A T-bill is money-market plumbing. A drug royalty is a legal claim on a future contingent event that may never occur. The first tokenizes cleanly because it was already digital in everything but wrapper. The second does not, because the contingency is a human biological outcome that no chain can adjudicate.
Failure mode four: the metric trap.
China's plan targets 25% of global first-in-class drugs by 2030. Read that carefully. First-in-class is defined by mechanism novelty โ a new target, or a new way of hitting an old target. It is not defined by clinical superiority. A PD-1/VEGF bispecific like ivonescimab is arguably a new combination of validated targets, not a new target. Depending on how you draw the line, the 25% target is either ambitious or achievable through a wide statistical aperture.
This is the same pathology as TVL. A headline metric that can be maximized without improving the underlying substance. The number goes up. The system does not get more robust. When I stress-tested Balancer V2 vaults during the 2020 liquidity mining frenzy, I watched pools inflate total value locked with recursive incentives while real depth stayed thin. The number and the liquidity were different things. First-in-class counts and therapeutic value are different things in exactly the same way.
Failure mode five: the event-timing gap.
A tokenized asset needs continuous price discovery to be useful as collateral, as a hedge, or as a market. A biotech asset has almost none. Between readouts, its value is a function of an undisclosed data monitoring committee's confidential view. Summit's 24% single-day move was not a repricing of a liquid market โ it was the release of a discrete information event into a thin book. If you tokenize that, you tokenize a market that is flat for months and then gaps. That is not a market. That is a series of jumps. And jumps are the one thing that liquidators and lending protocols are worst at handling.
I have watched this failure mode play out in DeFi stress tests. A collateral asset that gaps 24% on a single oracle update is a liquidation cascade waiting to happen. Now imagine that oracle is a press release. Now imagine the liquidation engine is levered against it.
Core, Continued: What the DeSci Thesis Gets Right
I do not want to be unfair. There is a real, narrow case where on-chain rails add something biotech does not already have.
Capital formation at the long tail. Traditional biotech venture capital funds a few hundred assets a year and ignores the rest. A platform that lets a thousand researchers fund a thousand early-stage hypotheses โ with tokens representing a claim on downstream licensing revenue โ could expand the funnel. The failure is not the idea. It is that the funnel has not produced a single tokenized asset that cleared a major licensing event. The deals are happening off-chain, in the same law firms that have always done them, and the on-chain layer has captured none of the $120 billion.
Transparency of milestone structures. Licensing terms are usually confidential. The $120 billion headline figure is a sum of mostly undisclosed packages. DeSci's genuine contribution could be standardized, machine-readable deal structures โ upfronts, milestone tiers, royalty rates โ that let anyone price the asset class. But standardization requires disclosure, and disclosure is the thing sellers refuse. The entire competitive advantage of a Chinese biotech selling into a patent cliff is the information asymmetry it holds over the buyer. Asking it to publish its milestone ladder on-chain is asking it to surrender the asymmetry.
Secondary liquidity for illiquid claims. Royalty interests are famously illiquid. A secondary market could let holders exit before a readout. In practice, the secondary market for tokenized royalties is thin, wide-spread, and dominated by a handful of wallets โ the same whale concentration that shows up everywhere else in crypto. The few IP-NFT projects that launched secondary markets discovered the obvious: nobody wants to buy a tokenized claim on a drug that might fail, from a seller who knows more than they do.
Programmable waterfall splits. This is the strongest technical case. A licensing deal has a waterfall: upfront, then milestone tiers, then royalty percentages, split between originator, developer, and academic inventors. Encoding that waterfall in a contract makes it auditable and automatic. But "automatic" only holds if the fiat actually arrives on-chain, which returns us to the settlement problem. The waterfall is programmable in the same way a spreadsheet is programmable. The money is not.
I keep landing on the same conclusion: the on-chain version of biotech IP is not wrong. It is early in a way that the timeline does not forgive. The off-chain deals are compounding now. The on-chain layer is still writing whitepapers about the deals it intends to host.
Contrarian: The Blind Spot Nobody Prices
The DeSci narrative assumes that tokenization democratizes access to biotech upside. The data says the opposite. Value capture in this cycle is happening in two places: multinational pharma buying Chinese assets, and a handful of public equities like Summit that reprice on single readouts. The on-chain layer is not capturing either.
Look at the actual flows. The $120 billion is headline value โ total consideration including milestones. Upfront cash is typically 5% to 15% of the headline. So the "cash value" of the wave is somewhere between $6 billion and $18 billion, not $120 billion. Anyone modeling tokenized royalties off the headline number is modeling a mirage. This is the biotech equivalent of counting TVL that is double-leveraged and double-counted โ a number that exists on a slide and nowhere in a bank account.
Then there is the governance parallel, and it is uncomfortably exact. DeSci DAOs promise community-directed science funding. Voter turnout in on-chain governance sits below 5% across the board, and the decisions that matter are made by the wallets with the most tokens โ usually the founding team, a venture fund, or an exchange. A DeSci DAO that funds research is a small committee with a token wrapper. The "community" is a cap table. The rhetoric of decentralized science and the reality of concentrated control are the same distance apart as the rhetoric of decentralized finance and the reality of a dozen market makers.
And the competitive dynamic that Richter flagged deserves more weight than it got. China is simultaneously the supplier and the rival. Multinationals are buying Chinese assets while building next-generation internal pipelines. That is a customer who is also a competitor โ the worst possible counterparty structure, and it does not map onto any token model. Merck and Eli Lilly are both buyers of external innovation and owners of incumbent franchises. When Keytruda's patents expire, biosimilars will compress the very market that makes ivonescimab valuable. Nobody is pricing that on-chain either.
There is a further blind spot: the obesity lane. The report flags Chinese developers targeting GLP-1, the growth engine for Eli Lilly. This is where the next wave of licensing headlines will come from, and it is where the demand is most obviously unmet. A tokenized-IP thesis that ignores the single largest demand pocket in pharma is not a thesis. It is a category error dressed in crypto vocabulary.
Volatility is noise. Architecture is the signal. The architecture here says: the value is in clinical data and legal claims, not in tokens. The tokens are downstream of the data, and the data is downstream of the sponsor. Until an on-chain system can verify a readout without trusting the sponsor, the on-chain layer is a wrapper around a phone call.
We didn't get the settlement layer. We got the press releases. And a press release, no matter how you hash it, is not a proof.
Takeaway
Watch two dates. The first is the HARMONi-3 squamous readout in the second half of 2026. That single event decides whether the Chinese-data narrative travels to Western patients, and whether the $120 billion headline converts into cash. It is also the event any serious on-chain IP project must eventually settle against โ and the one it cannot, because no contract can read a data monitoring committee's confidential minutes.
The second is softer. Watch whether any tokenized-IP project ships a milestone escrow that settles against a real, externally verified readout rather than an admin flag. Watch whether the oracle is the sponsor's own key, or something that survives the sponsor's incentive to lie.
The question for the next cycle is not whether biotech assets can be tokenized. It is whether the token adds a single enforceable right that the underlying contract did not already grant. If it does not, then the $120 billion stays off-chain, DeSci stays a demo, and the only thing that moves on a readout is a stock โ not a settlement layer.
The chain doesn't price biology. It prices the promise of biology. Those are not the same asset, and the spread between them is where the entire DeSci thesis lives or dies.