Snapshot

AI-powered drug discovery uses machine-learning models to design new drug molecules and predict their biological activity, replacing years of lab trial-and-error with computational search. The unit of output is an "AI-discovered candidate" — a molecule whose target or structure was identified by an AI system and that has entered human clinical trials. As of early 2026, roughly 173 such programs are in clinical development globally (94 in Phase I, 56 in Phase II, ~15 in Phase III). est. No AI-discovered drug has yet received FDA approval. The four US-listed pure-play companies in this space (RXRX, SDGR, ABSI, plus EXAI which merged into RXRX in November 2024) generate combined annual revenue under $340M — against a global pharma R&D spend of ~$288B that is the addressable market for their platforms and molecules.

Pharma spends $288B/year on R&D and gets roughly 50–60 new drug approvals annually — about $2–5B per approved drug when failures are included. est. AI platforms claim 80–90% Phase I success rates vs. 52% historically. All 10 of the largest global pharma companies have signed AI drug discovery partnerships since 2023. Actual revenue recognized by the three remaining public pure-plays is still under $340M combined.

The product & how money is made

AI drug discovery companies sell three things, usually in combination:

• Software licenses — Computational platforms that pharma companies pay to use for their own drug design. Schrödinger's software business ($199.5M in FY2025 revenue, 74% gross margin) is the clearest example. Annual contract value, recognized ratably.
• Collaborative partnerships — A pharma company pays an upfront fee plus milestone payments tied to research, development, regulatory, and commercial progress. Recursion has $500M+ in cumulative partner payments received and up to $20B in remaining contractual milestones across Roche ($150M upfront, up to $300M per program across 40 programs), Sanofi ($100M upfront, up to $5.2B total), Bayer ($80M upfront, up to $1.5B), and Merck KGaA ($20M upfront, up to $674M). Revenue is lumpy — recognized when milestones are delivered.
• Internal pipeline — The company develops its own drug candidates and captures the full economics (or partners/out-licenses at clinical stage). Schrödinger's co-discovery of zasocitinib (now Takeda's) generated $147M in cash from a 3.8% equity stake in Nimbus when Takeda paid $4B for the asset. Schrödinger also holds ~6% of Ajax Therapeutics, being acquired by Lilly for up to $2.3B.

The cost structure is almost entirely R&D personnel and compute. Wet-lab validation, clinical trials, and data acquisition (Recursion paid $49.9M to Tempus for clinical data in 2025) are the major cash outflows. None of these companies are profitable.

Demand

Contracted

Sources: Company press releases, SEC filings (10-K, 8-K), Recursion partner page.

Forecast

• The AI drug discovery market was $3.1B in 2025, projected to reach $43.9B by 2035 (30.5% CAGR). est.
• All 10 of the largest global pharma companies signed AI drug discovery partnerships since 2023 (CB Insights). Biologics-focused AI companies alone raised $1.6B in 2024 venture funding, 2× the 2023 level.
• 8 of the 10 largest M&A deals in the sector occurred since 2023.
• Global pharma R&D spending grew to $288B in 2024, projected to reach $340B by 2030. est. AI platforms are tools applied against this spend.
• A first FDA approval of an AI-discovered drug has roughly 60% probability of occurring in 2026 or 2027 per analyst estimates. est.

Supply

Capacity

• ~173 AI-discovered programs in clinical development globally as of early 2026 — ~94 Phase I, ~56 Phase II, ~15 Phase III. est.
• 500+ companies apply AI to drug discovery (CB Insights), but only a handful have molecules in human trials. Most are pre-clinical or software-only.
• The three remaining US-listed pure-plays (RXRX, SDGR, ABSI) collectively have ~20+ clinical and advanced preclinical programs. Recursion alone has 10+ clinical/preclinical programs plus ~10 advanced discovery programs and 10+ partnered programs.
• Schrödinger's partnered pipeline includes 7 programs in Phase I or later, with zasocitinib (Phase III, psoriasis) the most advanced AI-discovered molecule in the US pipeline.

Bottlenecks

• Clinical trial execution — Human clinical trials still take 6–10 years regardless of how fast AI finds candidates. The FDA regulatory pathway is not compressed by compute.
• Proprietary training data — High-quality biological datasets (clinical outcomes, protein structures, binding affinities) are expensive and scarce. Recursion spent $49.9M on Tempus oncology data in 2025.
• Hybrid bio+AI talent — People who understand both molecular biology and ML are rare. Recursion has ~800 employees; Absci and Schrödinger are smaller.
• Compute — Recursion operates BioHive-2, a 504-GPU NVIDIA H100 supercomputer (ranked 69th on Top500). Absci partnered with AMD for GPU access.
• Wet-lab validation — AI-generated candidate molecules must be physically synthesized and tested in biological assays before entering trials. This step is serial, not parallelizable beyond a point.

The gap

Upfront payments for platform partnerships have grown from $20–80M (2020–2023) to $100–115M (2025–2026). Milestone totals per deal have escalated from hundreds of millions to multi-billions. Takeda paid $4B for a single AI-co-discovered molecule (zasocitinib).

The players

Note: EXAI (Exscientia) merged into RXRX (Recursion) on November 20, 2024 and no longer trades separately. Recursion's financials below include the combined entity.

Sources: Q1 2026 and FY2025 earnings reports, SEC filings, StockAnalysis.com (Jun 2, 2026 prices).

Business model differences

• RXRX — Revenue comes from pharma partnerships (milestone payments) and eventually its own drug candidates. The merged Recursion+Exscientia entity is the largest pure-play by headcount and pipeline breadth. Cash burn is the highest at ~$372M/year. Revenue is lumpy and milestone-dependent — Q1 2026 revenue was $6.5M (vs. $35.5M in Q4 2025).
• SDGR — 78% of FY2025 revenue came from software licenses — recurring, 74% gross margin. Drug discovery upside comes from equity stakes in portfolio companies built on the platform. The Nimbus/Takeda deal ($147M cash from a 3.8% stake) and the pending Ajax/Lilly deal (~6% stake, up to $2.3B total) demonstrate this model. Lowest cash burn and longest runway of the three.
• ABSI — Earliest-stage: $2.8M in FY2025 revenue, one clinical program (ABS-101, Phase I — seeking a partner rather than advancing internally). ABS-201 (hair loss) is the lead internal asset, Phase I/2a initiated December 2025. The generative-AI antibody design platform is the differentiator. Cash runway to H1 2028 after a $64M raise in mid-2025.

The price of exposure

At $3.61, RXRX trades at $1.92B market cap with $665M in cash — the market values the combined Recursion+Exscientia platform, 800 employees, 10+ clinical programs, and $20B in contracted milestones at $1.26B above cash. Down from ~$10B market cap when the Exscientia merger was announced in mid-2024.

SDGR at $14.94 has ~$200M/year recurring software revenue and the lowest burn. The pending Ajax/Lilly deal could yield up to ~$138M (6% of up to $2.3B). At 2.8× EV/revenue, it trades at the lowest multiple of the three.

ABSI at $6.35: $0.99B market cap on $2.8M revenue. The Almirall partnership ($650M in potential milestones) is the largest contracted revenue source, but that cash arrives only if programs succeed.

For reference, Takeda paid $4B for a single AI-co-discovered molecule (zasocitinib). None of these companies are profitable, and all will need their cash reserves to fund operations through at least 2028.

What to deep-dive next

• Zasocitinib Phase III data quality — Takeda reported 71% sPGA 0/1 response at Week 16 vs. 11% placebo and 32% apremilast, with 90%+ durability at Week 60. An FDA filing is the next catalyst for SDGR re-rating.
• Recursion's milestone delivery cadence — 6 Roche phenomaps delivered, 5 Sanofi programs advanced, 25 Bayer data packages delivered. Track how many programs move from preclinical to IND to Phase I per quarter.
• Absci ABS-201 interim efficacy (H2 2026) — 13-week hair regrowth proof-of-concept data will determine whether Absci's generative-antibody platform translates to clinical results.
• Cash runway arithmetic — RXRX burns ~$93M/quarter against $665M cash; ABSI burns ~$28M/quarter against $144M. Both guide to early/H1 2028. If no major milestone payments arrive, dilution or partnership dependency follows.
• Phase III failure rates for AI-discovered drugs — The 80–90% Phase I success claim lacks sufficient Phase III data. Only ~15 programs are in Phase III globally. The traditional Phase III success rate is ~50–60%. Whether AI-discovered drugs maintain their advantage in late-stage trials is the single most important unknown for the sector.

Sources & confidence