AI in Drug Development 2026: The Dealmaker’s Guide

Executive Summary

For a decade, “AI in drug discovery” was a promise. In 2025–2026 it became a market with real money, real partnerships, and — for the first time — real patients responding to an AI-originated molecule. For business development teams, the question has shifted from whether AI matters to how to value, structure, and source AI-enabled assets.

This guide takes a deliberately commercial view. We are not here to explain transformer architectures; we are here to answer the questions a dealmaker actually asks: How large is the market and the capital behind it? Who are the platforms worth partnering with? How does Big Pharma structure AI deals — and what are they really paying for? Has any of it worked in the clinic yet? And where does cross-border sourcing, especially from China, fit?

The short version: the science has produced its first genuine proof point, the deal structures have standardized around contingent “biobucks,” and the smartest money is treating AI not as a separate asset class but as a probability-of-success multiplier on conventional pipelines.

How We Got Here

The current moment did not arrive overnight. It is the compounding of three waves. The first was structure prediction: DeepMind’s AlphaFold (2020) and AlphaFold3 (2024) effectively solved protein-structure prediction — work recognized with the 2024 Nobel Prize in Chemistry for Demis Hassabis and John Jumper. That collapsed one of biology’s hardest bottlenecks and made designing against a known structure computationally tractable at scale.

The second wave was generative design: models that do not merely predict but propose — novel small molecules, and de novo proteins from the David Baker lineage that seeded Xaira. The third wave, arriving in 2025, was clinical evidence: Insilico’s rentosertib became the first AI-discovered, AI-designed drug to post a positive, peer-reviewed Phase II signal. Capital chased each wave and then ran ahead of it — which is precisely why discernment now matters more than enthusiasm. The timeline compresses what used to take a generation of laboratory technique into roughly five years of computational progress.

The Market & the Money

Market-size figures for AI in drug discovery should be read with caution — they swing wildly with definition. Narrow estimates that count AI software and services land near $3 billion in 2026, growing at roughly 14% a year. Broader definitions that fold in AI-enabled R&D spending reach $8–9 billion. The methodology matters more than the headline.

The more revealing number is investment. More than $11 billion flowed into AI and machine-learning drug discovery and licensing across roughly 348 funding rounds in 2025. A handful of mega-rounds dominate: Xaira Therapeutics launched in 2024 with over $1 billion committed (backed by ARCH Venture Partners and Foresite Labs, led by former Genentech CSO Marc Tessier-Lavigne and co-founded by protein-design pioneer David Baker). Isomorphic Labs, the Alphabet/DeepMind spin-out, raised $600 million in early 2025.

How AI Changes Each R&D Stage

“AI in drug development” is not one thing — it touches each stage of R&D differently, and the value (and the hype) varies sharply by stage. A dealmaker should know exactly where a platform claims to add value before underwriting it.

• Target identification. Models mine multi-omic and clinical data to nominate novel disease targets — the step that produced the TNIK hypothesis behind Insilico’s rentosertib. High potential value, hard to validate until the clinic.
• Hit generation & lead optimization. The most mature use: generative chemistry proposes and ranks molecules, compressing the traditional 4–6 year discovery phase toward 18–36 months. Insilico reported roughly 18 months from target to clinical candidate for its IPF program.
• Biologics & protein design. The AlphaFold and Baker lineages generate functional antibodies and de novo proteins, widening the design space for large molecules.
• Translational & biomarker work. Predictive models improve patient stratification and biomarker selection — unglamorous, but directly tied to a trial’s probability of success.
• Clinical operations & trial design. AI assists protocol design, site selection, and enrollment forecasting — where China’s clinical-speed advantage compounds with computational tooling.
• Regulatory & medical writing. Large language models draft submission and trial documents — the fastest-adopted use, and the one squarely inside the FDA’s 2025 guidance.

The pattern: AI is most proven in discovery (hit generation, protein design) and most regulated in development (filings, evidence). The gap between those two — turning faster discovery into higher approval rates — is the value that remains to be proven, and the question every diligence should force a platform to answer.

The Platform Landscape

The field splits, roughly, into three camps. AI-native platforms (Isomorphic Labs, Generate:Biomedicines, Xaira, Iambic, Chai Discovery) build the engine first and monetize it through partnerships and an emerging in-house pipeline. Pipeline-first players (Recursion after its Exscientia merger, Insilico Medicine, Relay) use AI as the means to an end — clinical assets. And Big Pharma internal AI groups increasingly build or buy their own capability rather than rely entirely on partners.

Two structural moves defined 2024–2025. First, consolidation: Recursion and Exscientia merged in a roughly $688M all-stock deal (announced August 2024, closed November 2024), pairing Recursion’s scaled biology with Exscientia’s precision chemistry — then trimmed the combined pipeline in 2025 as reality met ambition. Second, the protein-design platforms (Generate, which has raised close to $750M and built a pipeline of ~17 protein drugs; Xaira; Isomorphic) moved from “filling pipelines” to producing actual clinical candidates.

For a fuller, regularly updated ranking of who is funded and who is advancing, see our companion AI Drug Discovery Companies power list, and for the broader capital backdrop, the biotech venture capital guide.

How Pharma Buys AI

The single most important thing for a dealmaker to understand is that Big Pharma is not mostly acquiring AI companies. It is partnering with them through multi-target research collaborations — and the economics are heavily back-loaded.

The template is Isomorphic Labs’ landmark 2024–2025 deals. With Eli Lilly, Isomorphic took roughly $45 million upfront against more than $1.7 billion in potential milestones to discover small-molecule therapeutics. With Novartis, the structure was about $37.5 million upfront and up to ~$1.2 billion in biobucks — a partnership the two expanded in February 2025. Together these deals approached $3 billion in headline value built almost entirely from contingent payments.

This is why deal structuring expertise matters more than ever. The mechanics — milestone design, royalty stacking, option-to-license, data rights — determine whether an AI partnership creates value or merely headlines. See our cross-border licensing term sheet guide and the live biotech licensing deal tracker for current comparables.

Does It Actually Work? The Proof Points

Skepticism has been warranted. For years, AI biotechs reported faster timelines and cheaper hit identification, but the clinic — the only court that matters — returned no verdict. In 2025, that changed.

In June 2025, Insilico Medicine published Phase IIa results for rentosertib (ISM001-055) in Nature Medicine. Rentosertib is a TNIK inhibitor for idiopathic pulmonary fibrosis whose target was nominated by AI and whose molecule was generated by AI — making it the cleanest test yet of end-to-end AI drug discovery. Patients on the 60 mg once-daily dose showed a mean improvement in forced vital capacity of +98.4 mL, versus a −20.3 mL decline on placebo, over 12 weeks. It is the first time an AI-discovered, AI-designed drug has shown a positive, peer-reviewed efficacy signal in a Phase II setting.

Meanwhile Recursion (post-Exscientia) guided toward a wave of clinical readouts — roughly ten across an 18-month window, including a Phase II ALDER readout expected around Q1 2026 — even as it cut several programs in 2025. The honest read on the field in 2026: the first green shoot is real, the next 24 months of readouts will decide whether it was a signal or an outlier.

The Open Questions

A disciplined view holds four unresolved questions in mind — each of which is also a diligence checklist for any AI deal:

• Does AI raise the probability of approval? So far AI has shown it can generate active molecules faster. Whether those molecules survive Phase III at higher rates than conventionally discovered drugs is unproven — the small, recent, survivorship- biased datasets cited as “80–90% Phase I success” should be treated with caution.
• Who owns an AI-designed molecule? Patent systems require a human inventor. Assets whose claims hinge on AI as inventor carry validity risk; clean human-in-the-loop documentation is essential and must be diligenced.
• Is the data clean? A generative model is only as defensible as its training data. Improperly licensed or contaminated data can taint downstream assets.
• Is the moat real? As foundation models commoditize, the durable advantage shifts to proprietary biological data and wet-lab validation loops — not the algorithm alone.

None of these is disqualifying; all are underwritable. But a platform that cannot answer them crisply is selling a narrative, not an asset.

The China Dimension

Western coverage of AI drug discovery often treats China as a footnote. That is a mistake. The company behind the field’s first clinical proof point — Insilico Medicine — has deep roots in Hong Kong and mainland China, and ran much of its rentosertib program through China’s clinical system. The combination that makes Chinese AI-biotech distinctive is structural: generative chemistry plus fast, lower-cost clinical execution.

For cross-border dealmakers, this creates a specific, fast-growing opportunity: AI-originated assets sourced from China. These programs increasingly reach human proof-of-concept faster and cheaper than Western equivalents, then become out-licensing candidates for global pharma. It is the AI-era extension of the broader China out-licensing wave we track in detail.

To go deeper on this angle, see Top Chinese Biotech Companies and the China outbound licensing tracker. The intersection of AI and China cross-border licensing is, in our view, one of the most underpriced deal categories in the market.

The Dealmaker’s Playbook

Pulling the threads together, here is how we advise clients to approach AI-enabled assets in 2026:

• Discount the biobucks, value the platform. Headline deal values are contingent. Underwrite the upfront and the probability-weighted milestones separately, and treat the platform’s renewal/expansion history as the real signal.
• Demand a probability-of-success thesis. Speed and cost savings are table stakes. Ask the platform to articulate why its approach raises the odds of a molecule surviving Phase II/III — and whether any evidence supports it yet.
• Separate validated platforms from funded narratives. Partnership renewals, peer-reviewed clinical data, and repeat pharma customers are the validators. A large Series A is not.
• Source globally — especially from China. AI-originated assets that reach human proof-of-concept in China are often the best risk-adjusted entries into the category.
• Structure for optionality. Option-to-license, staged milestones, and expansion rights let you pay for access now and conviction later — the same logic pharma uses on the platforms themselves.

Outlook 2026–2030

AI in drug development has crossed from narrative into evidence, but only just. The capital is enormous, the deal structures are mature, and the first molecule has worked in patients. What remains unproven is the only thing that ultimately matters: that AI raises the probability a drug reaches approval, not merely the speed at which candidates are generated.

The next 24 months of clinical readouts — from Recursion, Insilico, the protein-design platforms, and a wave of Chinese AI-biotechs — will settle it. For dealmakers, the window to build positions before that verdict becomes consensus is now. The firms that win will be those that can tell validated science from well-funded storytelling, value contingent structures with discipline, and source the best risk-adjusted assets wherever they originate.

Vision Lifesciences advises on exactly these decisions — sourcing, valuing, and structuring AI-enabled and cross-border assets. If you are evaluating an AI partnership or an AI-originated in-licensing opportunity, talk to our deal team.