AI Vendors Want Drug Royalties: Outcome-Based Licensing, IP Mechanisms, and the Case for Royalty Financing

OpenAI has publicly floated a revenue model in which it would subsidize AI compute for drug discovery firms in exchange for royalties on any resulting products. No deals exist yet. But the proposal — better understood as outcome-based licensing than simple pricing — has surfaced structural questions that matter to anyone in the pharmaceutical royalty market: what happens when a general-purpose AI vendor tries to capture downstream value from drug R&D, and what contractual and IP mechanisms would be required to make that work?

This piece examines the proposal in detail, maps it against existing AI-biotech deal structures, and considers how the pharmaceutical royalty financing market — already scaling rapidly — could provide a more proven framework for capturing AI-generated value in drug discovery.

What OpenAI Has Actually Proposed

In January 2026, OpenAI CFO Sarah Friar published a blog post outlining the company's intent to move beyond subscriptions and API pricing. The key passage: "As intelligence moves into scientific research, drug discovery, energy systems, and financial modeling, new economic models will emerge. Licensing, IP-based agreements, and outcome-based pricing will share in the value created."

The framing was deliberately broad, positioning OpenAI not as a software vendor but as economic infrastructure — with licensing models that evolve alongside the value its models help generate. Friar described OpenAI's strategy as a "Rubik's Cube" of business models, with each face representing different combinations of technology, pricing, products, and markets.

On February 3, 2026, CEO Sam Altman elaborated at a Cisco AI conference, suggesting OpenAI could cover compute costs for drug companies and then "get some royalty" from discoveries made. He was explicit that no partnerships exist: "This is not something we're doing now, but I think the frontier of scientific discovery with AI will require so much capital that maybe we think of ourselves as an investor in some of those cases." He also clarified that standard API customers would not be affected — their work remains their property.

What Altman described is not outcome-based pricing in the SaaS sense. It is outcome-based licensing: the grant of a right to use OpenAI's technology under terms that include a contingent royalty obligation triggered by commercial success. From a deal-structuring perspective, this more closely resembles a platform technology license in biotech — where the licensee pays a running royalty on net sales of products developed using the licensed technology — than it does a subscription pricing model.

The distinction matters. Licensing creates legal encumbrances on IP. It establishes obligations that survive corporate transactions. It introduces audit rights, disclosure obligations, and definitions of "licensed products" that can follow an asset through its entire commercial lifecycle. If OpenAI is serious about this model, it is not adjusting its pricing — it is entering the licensing business.

IP Attribution Under Current Law

The fundamental legal barrier to any AI vendor claiming inventorship is well established. Patent systems in the US, UK, and EU do not recognize AI as an inventor. The DABUS cases settled this definitively — the U.S. Federal Circuit ruled in 2022 that "only a natural person can be an inventor, so AI cannot be." UK and European courts reached the same conclusion.

This means OpenAI cannot be named as an inventor or co-inventor on any drug patent arising from use of its models. The inventors of record will be the biotech's scientists who selected, refined, and validated AI-generated outputs. The USPTO's guidance on AI-assisted inventions further clarifies that patents "reward human ingenuity" and that simply prompting an AI to solve a problem does not constitute inventive contribution — a person using an AI must have significantly contributed to the conception of the invention to be a true inventor.

The consequence is that any economic stake for an AI vendor must be established entirely through contract. Without an agreement, the default legal position is clear: the customer owns the IP, and the AI vendor has no claim. This is no different from any other software tool — a chemistry supplier does not receive royalties on drugs synthesized with its reagents, and a computational platform does not automatically acquire rights to discoveries made using it. The customer's scientists are the inventors, the customer's company is the assignee, and the AI vendor has zero standing.

There is, however, an important subtlety for OpenAI to consider. If an invention were generated so autonomously by AI that no human could credibly claim to have conceived it, the resulting compound might not be patentable at all — no human inventor, no patent. OpenAI therefore has a structural interest in ensuring that its customers' scientists are meaningfully involved in the inventive process, because a patented drug is far easier to monetize via royalty than an unpatented one. This creates an unusual alignment: the AI vendor needs the human contribution to be real enough to secure a patent, but wants the AI contribution to be acknowledged enough to justify a royalty.

The evidentiary challenge is also significant. Proving that a particular drug was "enabled" by an AI model requires detailed record-keeping: timestamps of prompts, logs of AI-generated outputs, documentation of which outputs were incorporated into the R&D program. In principle, OpenAI's API could retain records showing a particular molecular structure was suggested by the model. But R&D firms are typically unwilling to share invention logs with an external vendor, and the confidentiality implications are considerable. Any contractual framework would need audit rights, disclosure obligations, and carefully defined triggers — all of which introduce complexity and negotiation friction.

One additional nuance: the global conversation on AI inventorship is evolving. If any jurisdiction were to move toward allowing AI or AI owners to be co-inventors — or requiring disclosure of AI involvement in patent filings — the contractual dynamics would shift. For the next 1–2 years, however, the legal framework is settled: inventors must be human, and any AI vendor stake must be contractual.

Contractual Structures for Outcome-Based Licensing

If both parties agree to proceed, several established contractual structures could tether economic rights to the AI vendor. None are novel in pharma licensing — what would be novel is their application to a general AI platform provider.

Royalty license agreements represent the simplest approach. The contract stipulates that if the customer commercializes any invention developed with the aid of the AI, they pay a royalty to the AI vendor — say, X% of net sales. This functions like a tech transfer license from a university or a platform technology license in biotech. The AI vendor does not need to own the patent; the royalty is a contractual obligation binding the licensee. The agreement would define the scope of "AI-enabled invention," require the biotech to report any such inventions, and include covenants to pay agreed royalties on sales or sublicensing deals. The royalty obligation could be recorded as an encumbrance on the IP to put third parties on notice.

IP assignment or co-ownership clauses represent a more aggressive approach. The contract could require that any patent arising from use of the AI must be co-assigned to OpenAI, or that OpenAI be granted a perpetual exclusive license. Under U.S. law, patent co-ownership is complex — each co-owner can typically exploit a patent without consent of the other unless otherwise agreed — so OpenAI might prefer an exclusive license or a conditional assignment that triggers upon a specific event such as regulatory approval. Most customers would resist automatic IP assignment, and for good reason: it clouds ownership and complicates downstream transactions. But conditional arrangements — where an interest transfers only upon commercial success — are more feasible and have precedents in royalty financing structures where a fund receives a patent assignment as security for payment.

Joint R&D collaboration agreements shift the relationship from tool provision to formal co-development. If OpenAI contributed not just the model but personnel, expertise, or custom model training for a specific project, it could have legitimate human co-inventors on patents — strengthening its IP position considerably. The collaboration agreement would spell out how resulting IP is shared: jointly owned, or owned by the biotech with a royalty-bearing license to OpenAI. This model closely resembles how domain-focused AI biotech firms already operate. OpenAI would be acting as an AI-driven drug discovery partner, not a software vendor, which would justify an IP split similar to two companies collaborating on R&D.

Equity and synthetic royalty models provide indirect upside capture. OpenAI could take equity stakes in companies using its AI, negotiate warrants tied to R&D milestones, or structure synthetic royalty agreements — contracts where the company promises to pay a fraction of future revenue in return for the AI resources provided upfront. This approach may be more palatable to startups than a perpetual product royalty. Sam Altman's language about OpenAI seeing itself "as an investor" aligns with this framing. The synthetic royalty structure in particular has direct parallels to the mechanisms used by established royalty financing firms, which we discuss later.

Regardless of structure, several contract design elements are critical:

These are all standard provisions in biotech licensing, but implementing them in the context of a general AI platform introduces new friction — particularly around the definition of AI contribution and the audit mechanisms required to verify it.

How Existing AI-Biotech Companies Capture Value

While OpenAI's approach is novel for a general AI platform, companies at the intersection of AI and biotech are already capturing downstream value through IP rights and royalties. The critical distinction: they do so by being deeply involved in the R&D, not by licensing a tool.

Isomorphic Labs (Alphabet/DeepMind)

Isomorphic Labs entered strategic collaborations with Eli Lilly and Novartis in January 2024, with a combined deal value approaching $3 billion in upfront and milestone payments, plus tiered royalties up to low double digits on net sales. Isomorphic received $45 million upfront from Lilly (with up to $1.7 billion in performance milestones) and $37.5 million from Novartis (with up to $1.2 billion in milestones). The Novartis collaboration was expanded in February 2025 to include up to three additional programs, reflecting what Novartis described as "exploration of new chemical spaces that would be unavailable to probe through traditional methods."

In March 2025, Isomorphic raised a $600 million Series A, signaling substantial institutional conviction. The company leverages AlphaFold 3 and related AI technologies built on DeepMind's protein-folding expertise to design small molecule drug candidates for specific targets. Crucially, Isomorphic's scientists are actively co-inventing the drugs — the company is not licensing its platform for pharma to use independently. The pharma partner obtains exclusive rights to the resulting drug candidates, while Isomorphic retains rights to its underlying AI platform. First molecules from these partnerships are expected to enter Phase I trials by late 2026.

The deal structure closely parallels traditional biotech out-licensing: Isomorphic discovers molecules, hands them off with rights to develop and commercialize, and retains a contractually defined economic stake (milestones plus royalties). The royalties "tether" Isomorphic to the drug's success even after the compound is transferred. This is outcome-based licensing in the fullest sense — but critically, it is grounded in Isomorphic's role as co-inventor.

Recursion Pharmaceuticals (including Exscientia)

Recursion acquired fellow AI-biotech Exscientia in late 2024 in a ~$688 million all-stock deal, creating one of the largest vertically integrated AI drug discovery platforms. The merger combined Recursion's biology-first phenomics approach with Exscientia's precision chemistry design and automated small molecule synthesis capabilities.

The combined entity has over 10 partnered programs with Sanofi, Roche/Genentech, and Merck KGaA, with more than $450 million in upfront and realized milestone payments received to date out of more than $20 billion in aggregate potential deal value. As of mid-2025, Recursion had six active development programs — four in oncology and two in rare diseases — after strategically deprioritizing three clinical-stage programs to focus resources. The company had over $500 million in cash as of mid-2025 and backing from NVIDIA.

Recursion's monetization model is instructive: it discovers compounds internally using its AI platform, files patents with its own scientists as inventors, and then licenses or assigns IP to partners in exchange for milestones and royalties. The Sanofi partnership, originally penned by Exscientia in 2022 covering the discovery of up to 15 small molecules in oncology and immunology, has already generated a $7 million milestone payment. This is a clear example of the AI-as-inventor model, where the IP position — and therefore the royalty claim — flows naturally from the company's direct role in discovery.

Insilico Medicine

Insilico Medicine achieved what is arguably the most significant clinical milestone in AI-driven drug discovery to date: the first proof-of-concept for a drug where both the target and the molecule were discovered using generative AI. Its TNIK inhibitor rentosertib (formerly ISM001-055) showed positive Phase IIa results in idiopathic pulmonary fibrosis, published in Nature Medicine in June 2025.

The trial enrolled 71 IPF patients across 22 sites in China. Patients on the highest dose (60 mg QD) showed a mean FVC improvement of +98.4 mL compared to a -62.3 mL decline in the placebo group — suggesting not just slowed progression but potential disease modification. Exploratory biomarker analyses further validated the mechanism: profibrotic proteins were significantly reduced and anti-inflammatory markers increased. Insilico's generative AI platform Pharma.AI identified TNIK as a novel target and designed the small molecule inhibitor, achieving preclinical candidate nomination in just 12–18 months per project versus the typical 2.5–4 years.

Insilico owns the IP outright. It is both the platform creator and the drug developer. The rentosertib result represents the strongest evidence to date that AI can contribute meaningful value beyond just hit generation — and, importantly, that the AI company itself can capture that value through direct IP ownership rather than through outcome-based licensing of a tool.

Xaira Therapeutics

Xaira launched in 2024 with $1 billion in initial funding — one of the largest initial funding commitments in biotech history — and has since raised approximately $1.3 billion total. Co-founded by Nobel laureate David Baker and led by former Genentech CSO Marc Tessier-Lavigne, Xaira is built on Baker's RFdiffusion and RFantibody generative protein design models.

Rather than license its AI to others, Xaira is pursuing a fully integrated model: discovering and developing drugs internally using its generative AI, thereby owning the IP outright. The company has expanded to over 100 employees, hired former Roche and regulatory executives, and is approaching first-in-human testing for an antibody-drug conjugate targeting GI cancers. Xaira's strategic choice to keep the AI and IP under one roof reflects investor preference for models where the AI advantage translates into proprietary assets rather than obligations to third parties. It also underscores a broader pattern: many AI-biotech investors see vertical integration as the most direct path to capturing the full value of AI-generated discoveries.

AbCellera

AbCellera (Nasdaq: ABCL) has evolved from a pure antibody discovery partner into a clinical-stage biotech with its own pipeline. The company's AI-enabled single-cell screening technology has supported 96 cumulative partner-initiated programs through 2024. Its earlier business model — discovering antibodies for partners like Eli Lilly (which yielded bamlanivimab) and retaining royalty rights on resulting products — provides one of the cleanest precedents for how an AI-enabled discovery platform can contractually secure downstream economics without being a direct co-inventor in the traditional sense.

AbCellera has since shifted toward building its own pipeline, with ABCL635 and ABCL575 now in Phase 1/2 trials. Revenue is expected to grow from $28.8 million in 2024 to an estimated $55.2 million in 2026, driven by new partnership milestones and pipeline advancement. The evolution from tool-provider-with-royalties to integrated drug developer mirrors the broader industry trajectory.

Other AI Platform Deals in Early 2026

The pattern extends beyond these established players. At JPM 2026, Chai Discovery — a biologics-focused AI company that reached a $1.3 billion valuation just 18 months after launch — announced a deployment deal with Eli Lilly to design novel biologics across multiple targets, including development of an exclusive AI model trained on Lilly's proprietary data. Noetik partnered with GSK, and Boltz with Pfizer. These deals represent a new wave of AI platform collaborations that are deeper than simple tool provision but structured differently from Isomorphic's full co-development model. Notably, none of these deals have been reported as involving outcome-based licensing in the OpenAI sense — they follow conventional upfront payment, milestone, and royalty structures anchored in the AI company's inventive contribution.

Anthropic and Other General AI Vendors

To date, OpenAI's closest general AI competitors have not publicly pursued direct royalties from client innovations. Anthropic has been active in offering its Claude AI model to pharma and life science companies for enterprise applications, including clinical trial data analysis and document automation. These are standard SaaS or API arrangements where the client pays for model usage and retains all IP. Google Cloud provides AI tools to biotech, sometimes in exchange for large cloud contracts, but not typically for direct royalties on products. This indicates that OpenAI's outcome-based licensing concept stands out from standard practice — most AI vendors have sought value by pricing their services, not by staking claims to inventions.

The Pattern

Across all of these examples, one theme is consistent: companies that capture royalty value from AI-driven drug discovery do so by being active participants in the inventive process. Their royalty rights flow from that participation — through co-inventorship, patent ownership, or contractual terms anchored in a genuine research contribution — not from merely providing a tool.

OpenAI's challenge is that it wants the scalability of AI-as-tool (horizontal deployment across many companies) with the upside capture of AI-as-inventor (royalties, milestones). Bridging these two models requires convincing partners that a general-purpose LLM deserves inventor-like economics — a significantly harder sell than the deals described above.

Preserving Royalty Rights Through Asset Transactions

Drug discovery is a long and winding road. A successful AI-discovered compound might pass through multiple corporate transactions — startups raising new funding rounds, spinning out assets into separate companies, licensing drugs to big pharma, or being acquired outright. For any outcome-based licensing arrangement to work, the AI vendor's economic rights must survive these transitions.

Consider a scenario: Biotech Startup X uses OpenAI's model (under an outcome-based license) to invent Drug Y. The startup later licenses Drug Y to Pharma Co. for $50 million upfront and milestones. If OpenAI's contract is only with Startup X, what happens when the IP for Drug Y is assigned or licensed to Pharma Co.? Without contractual protections, OpenAI could be left behind, since Pharma Co. never signed the original agreement and might not recognize any obligation.

Standard pharma licensing provisions address this through several mechanisms. Assignment and sublicense clauses ensure the royalty obligation is binding on successors and assigns. Anti-assignment provisions require the licensor's consent before IP can be transferred, giving OpenAI a say in any transaction. Sublicense passthrough provisions define the AI vendor's share of sublicense income — typically framed as "Company will pay X% on its (or its sublicensees') net sales of any Licensed Product worldwide, regardless of patent status," which covers both direct commercialization and out-licensing.

OpenAI might also structure its rights as a direct interest in the IP, rather than a personal contract with the startup. If OpenAI held a recorded license or partial ownership of a patent, any acquirer would need to obtain OpenAI's consent or buy out that interest. A conditional assignment — where the startup assigns an interest in the patent to OpenAI as security for royalty payment — mirrors structures used in royalty financing, where a fund receives a patent assignment that is released once payment obligations are satisfied.

Bankruptcy protection is another consideration. If a biotech licensee entered bankruptcy with an OpenAI royalty obligation in place, U.S. Section 365(n) provides some protection for patent licensees, but the AI vendor's position would depend on how the interest is structured. If OpenAI holds a security interest in the patent (akin to collateral), its position in bankruptcy is stronger. Well-drafted contracts should contemplate this scenario — the royalty financing market has developed extensive precedent here, including trust structures (as in the Nanobiotix/HCRx deal) that separate royalty payment streams from the originator's general obligations.

For acquirers conducting due diligence, an AI vendor royalty creates a new type of encumbrance to evaluate. If a pharma company considering a $500 million acquisition discovers a contingent royalty to OpenAI, the deal dynamics shift — the obligation must be priced in, negotiated around, or bought out. This is familiar territory (university licenses, platform royalties, stacked obligations are routine), but an AI vendor royalty is an unfamiliar category. A "low single-digit" royalty (1–5%) might be tolerable as another cost of the asset. A more aggressive rate would reduce acquirer appetite. OpenAI's CFO explicitly referenced taking a "small percentage," suggesting awareness that the rate must be proportional to the AI's perceived contribution.

The real-world precedents are encouraging: when a pharma company acquires a biotech, it routinely assumes existing royalty obligations to upstream licensors (universities, platform technology providers, co-development partners). An OpenAI royalty, if properly structured, would function identically. The question is whether acquirers will accept the premise that a general AI platform merits the same treatment as a research institution or co-inventor.

The Role of Royalty Financing — and Why It Matters Here

There is an established, rapidly growing market mechanism for exactly the type of value exchange OpenAI is attempting: pharmaceutical royalty financing. Rather than inventing new structures from scratch, OpenAI — and the biotech companies it hopes to partner with — should be looking at how the royalty financing market already operates, because it provides both the contractual templates and the commercial precedent for tethering economic rights to future drug revenues.

The Market Today

The pharmaceutical royalty financing market reached approximately $6.5 billion in aggregate transaction value in 2025, up from $5.7 billion in 2024 — continuing a growth trajectory from under $200 million per year in the early 2000s to a mainstream component of biopharma corporate finance. According to a Deloitte survey, more than 55% of biopharma leaders have shown increased interest in royalty funding due to its non-dilutive nature, nearly 90% plan to consider royalties for future capital needs in the next three years, and nearly 80% are interested in synthetic royalties specifically.

The core mechanism: a company monetizes a portion of its future product royalty stream in exchange for upfront, non-dilutive capital. The royalty investor acquires a contractual right to a percentage of future net sales, typically with caps, time limits, and structured payment mechanics. When the cap is reached or the time limit expires, the royalty reverts to the originator.

2025 Transaction Highlights

Several 2025 transactions illustrate the sophistication and scale of modern royalty structures:

BridgeBio's $300 million deal with HealthCare Royalty (HCRx) and Blue Owl Capital involved the sale of European royalties from Attruby (acoramidis), with proceeds used to fund U.S. self-commercialization — effectively non-dilutive launch capital structured post-approval.

Genfit's €185 million transaction with HCRx involved sale of a portion of royalties on Iqirvo (elafibranor) sales payable under its Ipsen license. Cumulative payments to HCRx are capped, after which all future royalties revert to Genfit. The deal simultaneously addressed Genfit's convertible debt overhang — demonstrating how royalty financing can serve dual strategic purposes.

Blackstone's $700 million synthetic royalty with Merck was tied to future net sales of sacituzumab tirumotecan (sac-TMT) via a development funding agreement — notably demonstrating that even large, well-capitalized pharma companies use synthetic royalty structures for portfolio de-risking and capital optimization.

BeOne Medicines' up to $950 million sale of a royalty interest in Amgen's Imdelltra to Royalty Pharma represents another precedent for royalty monetization at significant scale.

Nanobiotix's $71 million deal with HCRx was secured against royalties on JNJ-1900 (NBTXR3) payable under its Janssen license, with a trust structure distributing payments to HCRx and the European Investment Bank — demonstrating the structural sophistication available in modern royalty transactions.

The institutional maturation is also reflected in deal innovation: multiple underlying assets, staged funding tranches, step-down or step-up royalty rates, put/call rights, and creative arrangements like XOMA's royalty share agreement with Takeda, where Takeda's royalty obligations on one asset were reduced in exchange for XOMA receiving payments across a basket of nine development-stage assets. KKR's acquisition of a majority stake in HCRx in mid-2025 signals growing institutional conviction in the durability of royalty-based investing.

Direct Relevance to AI-Driven Discovery

The connection to OpenAI's proposal is direct. What OpenAI is attempting — providing upfront resources (compute) in exchange for a share of future product revenues — is structurally analogous to what royalty financing firms do every day. The differences are informative:

The royalty financing market has spent two decades developing the contractual architecture, valuation methodologies, and institutional infrastructure that OpenAI would need to build from scratch. Rather than reinventing these mechanisms, there is a strong argument that AI-driven drug discovery creates a natural new asset class for the existing royalty market.

The Emerging Asset Class

As more AI-discovered compounds enter clinical development, the royalty streams associated with these assets will become investable. Consider: if Recursion discovers a compound using its platform, licenses it to Sanofi with milestone and royalty terms, the resulting royalty stream is identical in structure to any other pharma royalty. It can be valued, traded, and financed using existing infrastructure. The fact that AI contributed to the discovery does not change the fundamental economics — a royalty is a royalty, regardless of whether the molecule was discovered by AI-guided screening, phenomics, or traditional medicinal chemistry.

Insilico's rentosertib, if it advances successfully through clinical development, will generate a royalty stream that looks like any other IPF drug royalty from the investor's perspective. The AI-specific question is whether the discovery method affects the probability of clinical success — and Insilico's Phase IIa data suggests it might, at least for target selection quality.

For royalty financing firms — HCRx, Royalty Pharma, XOMA, Blackstone Life Sciences, Blue Owl, and others — developing competence in evaluating AI-generated assets will become increasingly important. This means understanding which AI platforms produce higher-quality targets and leads, assessing the contractual terms governing AI vendor participation, and evaluating the IP positions of AI-biotech companies. The question is not whether AI-generated royalties will emerge as an investable asset class, but when — and whether the royalty financing market will be ready to participate when the first AI-designed drugs reach commercialization.

An Alternative Path for Biotech Companies

For biotech companies using AI in drug discovery, royalty financing offers a complementary — and potentially superior — path to the one OpenAI is proposing. Instead of granting the AI vendor an open-ended royalty claim, the biotech could use the royalty financing market to monetize a portion of its future product royalties to fund the compute and AI resources it needs. The economics might look similar — the biotech gives up some future revenue to fund its R&D — but the structure is cleaner: the AI vendor gets paid as a service provider, the biotech owns its IP free and clear, and the royalty investor holds a well-understood financial asset with established valuation methodologies.

This is precisely the model that has made royalty financing mainstream: companies like Genfit, BridgeBio, and Nanobiotix have used it to fund critical R&D and commercialization activities without diluting shareholders or encumbering their IP with novel, untested vendor claims. If a biotech needs $10 million in AI compute for a discovery program, it may be more efficient to raise that capital from a royalty investor than to accept outcome-based licensing terms from OpenAI — particularly given the attribution ambiguity and contractual complexity involved.

Feasibility and Gating Factors

Several practical factors will determine whether outcome-based licensing by AI vendors gains traction in the next 1–2 years.

Will biotech firms agree to share IP value? Early-stage companies are protective of their IP. The reaction from the open-source AI community has been pointed: the proposal "makes the case for local even stronger." Companies with resources would prefer to run AI tools locally rather than accept downstream obligations. The availability of capable open-weight models provides alternatives. However, not all firms can run competitive models in-house, and if OpenAI's technology is demonstrably superior for specific tasks, some companies may accept the terms — particularly cash-constrained startups that need compute subsidies.

How significant is the AI contribution? As Semafor noted, AI-generated drug candidates are not yet showing higher overall approval rates. The bottleneck remains clinical trials. If the AI only accelerates hit generation — the cheapest and fastest phase — the value proposition for a perpetual product royalty is weak. However, Insilico's rentosertib results suggest AI may identify better targets and design better molecules, not just faster ones. A comprehensive 2025 review noted that AI-designed molecules are achieving 80–90% Phase I success rates, though no AI-discovered drug has achieved FDA approval as of December 2025.

Attribution ambiguity. Defining when a drug was "enabled" by an AI model versus independently conceived remains inherently difficult. Contracts can attempt to define this, but it may become a perpetual source of dispute.

Competitive dynamics. Competitors can differentiate by explicitly not claiming downstream IP. Anthropic, Google Cloud, and open-source model providers currently charge standard fees with no IP encumbrances. However, the competitive landscape is fluid — Isomorphic Labs, several AI-biotech startups, and even general AI providers are exploring data licensing and partnerships with biotech companies.

Execution complexity. Structuring outcome-based licensing deals is substantially more complex than selling API access. Each deal may require custom negotiation, ongoing disclosure management, and audit infrastructure. The overhead could consume the upside from initial pilots.

Conclusion

OpenAI's vision of capturing royalties from AI-enabled drug discoveries represents an ambitious attempt to bridge the AI-as-tool and AI-as-inventor models. The contractual mechanisms to make it work exist — they are the same tools refined over decades in biotech licensing and, more recently, in the pharmaceutical royalty financing market: structured payment mechanics, assignment clauses, security interests, audit rights, cap structures, and milestone triggers.

But mechanism availability is not the same as market acceptance. The core question is whether the pharmaceutical industry will accept the premise that a general-purpose AI platform — as opposed to a domain-specific co-inventor — deserves a share of the economic value of drugs it helped discover. The existing case studies strongly suggest that royalty capture requires demonstrated inventive contribution: Isomorphic Labs, Recursion, Insilico, and Xaira all own or co-own the IP they helped create because they did the science.

For the pharmaceutical royalty market, the more immediate opportunity may lie not in outcome-based licensing by AI vendors, but in the royalty streams that AI-biotech companies are already generating through conventional deal structures. As AI-discovered compounds advance through clinical development, the royalty assets associated with these programs will become investable using existing frameworks. The question is one of timing and clinical validation: when the first AI-designed drug reaches commercialization, the royalty market will need to be ready.

In the meantime, OpenAI's proposal has usefully surfaced a question the industry needs to answer: as AI becomes a more material contributor to drug discovery, how should the economics of that contribution be captured? The answer, as always in pharma, will emerge one deal at a time.

Disclaimer: The author is not a lawyer or financial adviser. The content of this article is provided for informational purposes only and does not constitute investment, legal, or financial advice. Readers should consult qualified professionals before making any decisions based on the information presented here.