AI Drug Discovery Deal Tracker 2026

The Shift to Platform Deals

The defining change of 2025–2026 was not any single transaction — it was a change in what pharma buys. For years, the archetypal biotech deal was a bet on one asset. The new pattern is investment in AI platforms and infrastructure that can generate many shots on goal. A flurry of platform deals to open 2026 made the cultural shift explicit: from single-asset bets toward broad discovery capability.

That matters for how you value and structure these deals. You are no longer underwriting one molecule’s probability of success; you are underwriting a capability — and the deal terms reflect it.

Three forces drove the shift at once. Pharma faces a historic patent cliff — over $200B in revenue losing exclusivity this decade — creating urgent pipeline demand. AI matured from slideware to a first clinical proof point, lowering the perceived technology risk. And a cheaper-capital scramble for differentiation pushed the largest players to lock in access to the best platforms before competitors did. The result is a land-grab for AI capability that looks less like classic biotech licensing and more like strategic technology procurement — which is exactly why reading these deals with a traditional single-asset lens leads buyers astray, and why the benchmarks and valuation discipline below matter more than the headline figures.

This tracker is built for that reading: the table is the data, and the sections around it translate the terms into the questions a buyer, seller, or investor should actually ask before signing.

A Short History of AI Deals

The platform-deal boom did not appear from nowhere. It is the third act of a story that began nearly a decade ago, and the arc explains why today’s terms look the way they do.

• 2017–2021 — the first wave. Early AI platforms signed discovery-services deals with pharma: Exscientia with Sumitomo Dainippon, GSK, Sanofi, and Bristol Myers Squibb; Insilico and others with multiple partners. Terms were modest and the value proposition was speed and cost, not validated assets.
• 2021–2023 — the IPO wave and the reckoning. Recursion, Exscientia, Schrödinger, and Relay went public; capital flooded in. Then timelines slipped, some programs failed, and valuations compressed — a healthy correction that separated platforms with real chemistry from those with slide decks.
• 2024–2026 — the platform-and-proof wave. The Recursion–Exscientia merger consolidated the field; Isomorphic signed billion-dollar collaborations; and Insilico delivered the first clinical proof point. Deals shifted from one-off services to strategic, multi-target, and infrastructure-scale commitments.

The throughline: each wave moved the basis of value up the chain — from selling compute time, to selling discovery services, to selling validated capability. The deals below reflect where that arc stands today.

The Deal Tracker

Snapshot as of June 2026. Values reflect public reporting; verify against primary disclosures before transactional use. Most “total” figures are contingent milestones (biobucks), not upfront cash.

Two patterns stand out across the table. First, repeat customers concentrate: Eli Lilly appears as a partner to Isomorphic, Chai, NVIDIA, insitro, and XtalPi — pharma’s AI conviction is not evenly spread, it clusters in a few aggressive buyers. Second, the most-validated platforms attract multiple independent pharma partners (Isomorphic with three; Schrödinger and insitro with several), which is the single most reliable third-party signal that a platform’s science is real.

Three Deal Structures

The tracker resolves into three distinct shapes:

• 1. The classic platform collaboration. Modest upfront + large milestone biobucks + royalties, across multiple targets (the Isomorphic template). Pharma pays for repeated shots on goal; nearly all value is contingent.
• 2. The infrastructure deal. Lilly’s up-to-$1B tie-up with NVIDIA buys compute and a “continuous learning” capability rather than specific molecules — pharma building the factory, not buying the output.
• 3. The subscription / access model. Noetik–GSK’s $50M-upfront, subscription-based framework licenses ongoing access to AI models. This is the most novel structure — recurring, software-like economics layered onto drug discovery.
• 4. The consolidation (M&A).Rarer, but real: Recursion’s ~$688M acquisition of Exscientia combined two platforms rather than buying molecules. Expect more platform-on-platform M&A as capital concentrates around the validated few.

For comparables beyond AI — and the broader licensing and M&A backdrop — see our biotech licensing deal tracker and pharma M&A tracker.

Deal-Term Benchmarks

Patterns across the disclosed AI platform deals give a working set of benchmarks. Treat these as orientation, not appraisal — every deal is negotiated on its specifics:

• Upfront as a share of total. Strikingly small. Isomorphic’s ~$45M upfront against over $1.7B in milestones is ~2–3% of headline value. Across the platform deals, upfronts clustering at $30–50M against $1B+ in biobucks is typical — the headline is mostly contingency.
• Per-target economics. Multi-target collaborations spread milestones across programs, so the value attributable to any single target is a fraction of the headline. Always divide.
• Subscription/access deals. Noetik–GSK’s $50M upfront for multi-year model access points to recurring, software-like pricing — a different benchmark class entirely from milestone-driven research deals.
• Infrastructure deals. Lilly–NVIDIA’s up-to-$1B commitment is a capital-expenditure benchmark, not a per-asset one — it buys capability, and its “return” is measured in pipeline productivity over years.

How to Value an AI Deal

The discipline is the same risk-adjusted logic you would apply to any licensing deal — with two AI-specific adjustments. First, discount the biobucks hard: probability-weight each milestone by realistic stage-by-stage success rates, and resist the temptation to credit AI with higher odds until clinical evidence justifies it. Second, value the platform option separately from the lead program: a multi-target collaboration is, in effect, a portfolio of options, and its worth lies in the breadth and renewability of shots on goal, not in any single molecule.

One further adjustment is specific to platforms: the comparison set. A pure-asset licensing deal is benchmarked against other assets at the same stage; a platform collaboration should also be benchmarked against the platform’s other deals. If a pharma can sign the same platform for similar terms next quarter, your negotiated upfront has a ceiling. Conversely, exclusivity in a disease area or target class is where a platform deal earns a premium — and where the real negotiation should focus, rather than on the milestone headline that dominates the press release.

For the underlying valuation mechanics, see our rNPV valuation guide and the term sheet guide.

The Next Wave: China

The deals above are overwhelmingly US/EU platforms partnering with Western pharma. The next wave is already forming on a different axis: AI-originated assets out-licensed from China. Chinese AI-biotechs — led by Insilico Medicine, whose rentosertib is the furthest-advanced AI-designed molecule — pair generative design with fast, lower-cost clinical execution, reaching human proof-of-concept and then becoming out-licensing candidates.

The economics are compelling for both sides. A Chinese AI-biotech can advance a generatively designed molecule to human proof-of-concept for a fraction of Western cost and time, then out-license ex-China rights to a global pharma that supplies capital, regulatory reach, and commercial muscle. For the Western buyer, an asset with early human data and an AI-design pedigree is a de-risked entry into the category at a price still below the frontier US platforms. This is the AI-era extension of the broader China out-licensing surge that already accounts for a record share of global pharma dealmaking — now with a computational-design premium attached.

We expect AI-originated China assets to become one of the most active deal categories of the next 24 months. Track the broader China dynamic in our China outbound licensing tracker and the platforms in the AI Drug Discovery Companies power list.

What We’re Watching

The deal table is a snapshot; these are the developments most likely to move it over the next 12 months, and what each would signal:

• The first AI-platform acquisition at scale. So far pharma partners rather than buys. A large outright acquisition of an AI-native platform would mark a new phase — and reprice the field.
• More subscription/access deals. If others follow the Noetik–GSK model, recurring AI-licensing revenue becomes a category, changing how platforms are valued (software multiples vs biotech milestones).
• A major China-origin AI out-licensing deal. The first nine-figure out-licensing of an AI-designed Chinese asset to Western pharma would validate the thesis and open the floodgates.
• A partnership termination. The quiet end of a flagship collaboration would be the most informative — and most under-reported — signal that a platform’s science is not delivering.

Common Mistakes Buyers Make

Having advised on platform and asset deals, we see the same avoidable errors recur. Each is a place where disciplined diligence pays for itself many times over:

• Anchoring on the headline number. A “$1.7B deal” that is $45M upfront is a $45M decision with an option attached. Negotiating or valuing off the biobucks figure overpays for contingency.
• Crediting AI with success it hasn’t earned. Applying above-industry success probabilities to AI-originated programs, on the assumption that AI “de-risks,” is unsupported by the current evidence and inflates valuation.
• Ignoring data and IP provenance. Failing to confirm clean training-data rights and human inventorship can leave a buyer with an asset whose patents are contestable.
• Buying a platform with no proprietary edge. As models commoditize, a platform whose only asset is a re-trainable architecture has no moat. Pay for proprietary data and validation loops, not for access to the same models a competitor can license.
• Missing the renewal signal. Treating a flashy new partnership as validation while ignoring whether prior partners renewed (or quietly walked) gets the most important signal exactly backwards.

The connective tissue: AI deals fail the buyer when enthusiasm substitutes for the ordinary discipline of probability-weighted valuation and clean diligence. The novelty is in the science, not in the rules of a good deal.

Dealmaker Takeaways

• Discount biobucks hard. Underwrite the upfront and probability-weighted milestones separately; the headline is marketing.
• Read the structure for conviction. Subscriptions and infrastructure signal belief; one-off milestone deals signal optionality.
• Watch the renewals.Isomorphic’s expansion with Novartis and third deal with J&J say more than any single upfront.
• Position for the China wave. AI-originated assets from China are the under-priced entry into this category.
• Prefer optionality structures. Option-to-license and staged milestones let you pay for access now and conviction later — the same logic pharma uses on the platforms themselves.
• Underwrite the data moat, not the model. The durable value sits in proprietary data and validation loops a rival cannot simply re-train.

The meta-point for any buyer or asset-holder: the AI deal market in 2026 rewards discipline over excitement. The platforms are extraordinary and the science is real, but the deals that create value are the ones underwritten with the same rigor as any other — contingencies discounted, validation demanded, and the rare durable moat paid for while the rest is treated as the option it is.

For the full strategic context, read the AI in Drug Development: The Dealmaker’s Guide. To structure or source an AI deal, talk to our team.