AI Drug Discovery: Relay's $4.1B Is the Only Phase 3 Name

AI biotech is 2026’s loudest story, and it is also the easiest place in the market to lose money on a story that is not yet a science.

Capital is flooding in, pharma is paying up for platforms, and the first AI-designed molecules have reached the clinic. The contrarian read is the opposite of the hype. Most of these names are pre-revenue platform bets where the stock catalyst is a partnership or a clinical milestone, not an FDA date. The money behind the narrative is real, the August 2026 IPO reopening and pharma’s patent-cliff M&A spending are both pointed at platform companies, but the financing window that funded the run can shut just as fast as it opened. The only sane way to rank these names is by who is closest to a real readout, not by who has the best AI pitch.

The field splits across three camps: physics and dynamics simulates molecules at the atomic level, generative AI and machine learning designs antibodies and small molecules from data, and the clinical-stage camp has the few companies with an AI-designed molecule actually in late-stage trials. The trade is not about the technology. It is about who owns a proprietary data plus wet-lab loop that turns software into a drug, which is the same filter the pre-revenue framework applies. Here is the field, ranked by nearest catalyst, with prices from Polygon’s August 28 close.

Relay Therapeutics (RLAY): $18.77, $4.1 billion. The only name with a real Phase 3.

Relay is the exception that proves the rule. Zovegalisib (RLY-2608) is the first allosteric, pan-mutant, isoform-selective PI3Kα inhibitor, designed from a full-length cryo-EM structure of the enzyme. About 40% of HR+/HER2- breast cancer tumors carry a PIK3CA mutation, and existing PI3Kα inhibitors are dose-limited by the off-target toxicity that comes from hitting the wild-type enzyme. Zovegalisib’s mutant selectivity is the whole point. In heavily pretreated PI3Kα-mutant HR+/HER2- metastatic breast cancer, zovegalisib plus fulvestrant delivered a 9.2-month median progression-free survival in the first-in-human ReDiscover trial. The FDA granted Breakthrough Therapy designation in February 2026, and the Phase 3 trial against capivasertib plus fulvestrant is now enrolling.

This is what AI-native drug discovery looks like when it works: a physics-derived molecule in a registrational trial with a Breakthrough Therapy designation. At $4.1 billion it is not cheap, but it is the only name in the group with a catalyst that can move the stock on a binary outcome. Compare that to Schrödinger at $1.5 billion for software or Recursion at $1.8 billion for breadth, and Relay’s premium is the price you pay for actually being in a registrational trial. Buy on weakness ahead of the Phase 3 enrollment milestones.

Absci (ABSI): $8.71, $1.5 billion. Generative AI with a real clinical readout coming.

Absci is the purest generative AI play: it designs antibodies from scratch using its own data and wet-lab loop. ABS-101, its anti-TL1A antibody for inflammatory bowel disease, dosed its first patients in a Phase 1 trial in 2025, making Absci clinical-stage for the first time, with interim data now expected. TL1A is one of the hottest targets in immunology, and this readout is the first real test of whether Absci’s generative design produces a differentiated antibody rather than a copy of someone else’s molecule.

The target alone tells you the prize. Merck paid $10.8 billion for Prometheus Biosciences in 2023 almost entirely for its anti-TL1A antibody, and Roivant and Pfizer are racing a competing TL1A program toward approval. A differentiated TL1A antibody is a multi-billion-dollar asset, which is why a $1.5 billion market cap for Absci is both cheap on the upside and expensive on the downside. The TL1A interim data is a genuine binary. Speculative buy at a 1% weight, and only if you are willing to hold through a readout that could cut the stock in half.

Recursion (RXRX): $3.34, $1.8 billion. Breadth, but no near-term binary.

Recursion merged with Exscientia in an all-stock deal worth about $688 million to build a pipeline of more than 10 clinical and preclinical programs, then cut three of them (REC-2282, REC-994, REC-3964) in early 2025 to conserve cash. The lead is now REC-394, a selective C. difficile toxin B inhibitor in Phase 2, alongside the Exscientia oncology assets. The platform and the data flywheel are real, and the near-term stock catalyst is not. Recursion has banked over $450 million in upfront and milestone payments from partners against more than $20 billion in potential milestones, which is why it can afford to be patient.

Recursion is a Hold. It is a legitimate long-term platform bet, but at $1.8 billion with no binary readout on the calendar, there is nothing to trade today.

Schrödinger (SDGR): $19.70, $1.5 billion. A software company, not an AI-native drug company.

Schrödinger is the case study in why the segmentation matters. It is not “AI-native” in the data-plus-wet-lab sense. It is a physics-based software business, and its Q2 2026 numbers prove it: software annual contract value of $29.6 million, up 27% year over year, and drug discovery revenue of $23 million driven by partnership milestones, including a new collaboration with Simcere. The internal pipeline is thin after the CDC7 inhibitor SGR-2921 was discontinued following two patient deaths in Phase 1, leaving the MALT1 inhibitor SGR-1505 as the sole internal clinical asset.

Schrödinger is a software company with a drug-discovery milestone kicker, not a binary biotech. The software is the moat; the pipeline is a lottery ticket. At $1.5 billion, value it on software multiples and treat the pipeline as a free option. Hold.

The distinction that matters

The line between “AI as a tool” and “AI-native” is the line between Schrödinger and Relay. Schrödinger sells the software that other companies use to design drugs, and its revenue is the proof. Relay uses the same physics to own a molecule outright, and its Phase 3 is the proof. Both are “AI drug discovery” in the marketing, but only one gives you a binary catalyst on a real asset. When you see an “AI biotech” deck, ask which side of that line the company sits on, because the answer determines whether you are buying a software multiple or a drug binary.

Risks

The bear case is simple. Not a single AI-designed molecule has proven itself in a Phase 3, and the first one that fails will reset the entire theme, the same way a single failed IPO can chill a financing window. These are pre-revenue companies living and dying on cash runway, and they burn hundreds of millions a year. The ones without a near-term readout to justify the next raise will dilute first. Every name here except Relay is a platform story, and platform stories get repriced toward zero when the narrative cools and the seven-filter screener that should have caught the hype kicks back in.

Verdict

Relay is the Buy: it is the only name with an FDA Breakthrough Therapy designation in a registrational trial. Absci is the Speculative Buy on the TL1A readout. Recursion is a Hold. Schrödinger is a Hold, but value it as software, not as a drug company. The discipline is the same as every other binary biotech trade: 1 to 2 percent positions, and never pay a hype multiple for a platform that has not yet produced a late-stage readout.