Scientist in a lean AI-native biotech lab with a protein model and liquid-handling robot | Healthcare Venture Capital Fund

AI Could Reopen Discovery’s Front Door. Who Will Write the First Check?

For a decade, discovery got too slow and expensive for most venture funds to wait on. AI is starting to change that math. Investors who understand what is proven, and what is not, will be first in line for the companies it makes possible.

The case for funding late-stage biotech rests on cost and time. Discovery used to take years of work, a full laboratory and a team large enough to run it before there was anything to show an investor. So capital moved downstream to companies with human data, and the earliest rounds thinned out. In the first half of 2026, seed and Series A biotech rounds fell to their lowest first-half count in a decade even as total venture dollars rose, according to BioCentury. Our companion analysis, Who Will Fund Discovery?, lays out how that happened.

Artificial intelligence goes after the economics that pushed capital downstream in the first place. If it becomes much cheaper and faster to go from a biological question to a candidate worth testing, the front of the pipeline becomes financeable again, and with smaller checks than before. That is a large if. Here is what the evidence supports so far.

What AI Has Actually Proven in Discovery

The scientific foundation is no longer in question. The 2024 Nobel Prize in Chemistry went to Demis Hassabis and John Jumper of Google DeepMind for AlphaFold, which predicted structures for virtually all of the roughly 200 million known proteins, and to David Baker for computational protein design. Work that once took a structural biology lab years can now start from a structure computed in minutes.

The clinical evidence is earlier but real. Insilico Medicine’s rentosertib is the first drug candidate in which both the disease target, a kinase called TNIK, and the molecule itself were identified with generative AI. The company reports about 18 months from project start to preclinical candidate. In a randomized, placebo-controlled Phase 2a trial of 71 patients with idiopathic pulmonary fibrosis, published in Nature Medicine in June 2025, patients on the highest dose gained an average of 98.4 mL in forced vital capacity over 12 weeks, compared with a 20.3 mL decline on placebo. Side effects were comparable across groups overall, though some discontinuations were linked to liver toxicity or diarrhea (Xu et al., 2025). Insilico started a Phase 3 trial in July 2026.

The broader pipeline shows both promise and limits. An analysis presented at the 2026 ASCO annual meeting counted 117 AI-enabled therapeutic assets from 63 companies that had entered human trials; 60 had completed Phase 1 and only 8 had completed Phase 2. Industry analyses have found AI-discovered molecules passing Phase 1 at 80 to 90 percent, well above historical averages, but succeeding in Phase 2 at about 40 percent, roughly the same as conventionally discovered drugs. As of October 2026, no AI-discovered drug had received full FDA approval.

The fair reading: AI is already making the first steps of discovery faster and cheaper. It has not yet shown that it makes drugs more likely to work in patients. For an investor in discovery, the first point matters more than it might seem.

The Cost of Asking a Biological Question Is Falling

The clearest sign is organizational. In October 2026, Anthropic disclosed that it had acquired Mithril Medicine, a Cambridge, Massachusetts biotech whose only full-time employee was its founder. Using Claude to coordinate contracts, logistics and experiments across 11 outside laboratories, with help from a small group of consultants, Mithril took an antibody for a rare muscular dystrophy from concept to early mouse data in under 18 months. The data are preclinical and come from small groups of animals, but the operating model is the point. Healthcare Discovery covered the science in The One-Person Biotech, and we examined why AI companies are buying drug developers in Frontier AI Labs Are the New Strategic Acquirers in Healthcare.

Regulation is moving in the same direction. In April 2025 the FDA announced a plan to reduce, refine or replace animal testing requirements, starting with monoclonal antibodies, and to accept new approach methods including AI-based computational toxicity models, organoids and organ-on-a-chip systems. In December 2025 it issued draft guidance on reducing studies in nonhuman primates for antibody toxicity testing. Primate studies are among the slowest and most expensive steps between a candidate and the clinic.

Put those together and the math of a seed round changes. When a founder can rent lab capacity by the experiment, use AI to design molecules and manage the work, and lean on computational evidence for parts of safety testing, the capital needed to reach a decisive early result falls. A few million dollars can again buy a real answer to a real question.

Where the Capital Is Going Now

So far the money has split in two. At one end are enormous rounds for AI platforms backed by large technology companies. Isomorphic Labs, the Alphabet company founded by Demis Hassabis to apply DeepMind’s AI to drug discovery, raised $2.1 billion in a Series B in May 2026. At the other end are small, specialized checks, such as the $10 million seed round London-based Helical raised in April 2026 to build what it calls a virtual AI lab for drug discovery teams. Between those extremes, the conventional Series A for a brand-new discovery company remains scarce.

That middle is the opportunity. Megarounds are closed to almost everyone. Small, early checks into lean, AI-native teams are open to angels, family offices, micro funds and physician investors, provided they can judge the science. Physicians in particular bring something capital alone cannot: they see unmet clinical problems first and can pressure-test whether a target matters to patients.

How to Underwrite an AI-Native Discovery Company

The phrase “AI-discovered” now appears in pitch decks for companies that use machine learning anywhere in their workflow, so investors need sharper questions:

  • Is the target new or known? Applying AI to a well-validated target lowers risk but also limits upside. A novel target, like TNIK in lung fibrosis, is riskier and potentially far more valuable.
  • Is there a real wet-lab loop? Models are only as good as the experiments that test them. Ask how many molecules were made and tested, how quickly results flow back into the model, and who runs the lab work.
  • What does a candidate cost? A lean team should be able to state the cost and time from biological question to development candidate. That number is the heart of the investment case.
  • Who owns the data? Proprietary experimental data compounds in value over time. Models trained only on public data are available to every competitor.
  • What survives a failed program? Platforms that learn from every experiment are worth more than single-asset bets. That is the original argument for funding discovery organizations rather than individual molecules.
  • How does it reach patients? Plan the route to a partner, licensee or acquirer early. Large drugmakers remain the main buyers, and frontier AI labs have joined them.

What Could Go Wrong

Three risks deserve real weight. First, faster discovery does not guarantee better drugs. Phase 2 is where most programs fail, and so far AI has not changed that rate. Second, AI could flood the clinic with more candidates of ordinary quality, pushing cost and failure downstream instead of eliminating it. Third, lean companies depend on outside laboratories and contract manufacturers, so their execution risk sits with vendors they do not control. None of these risks argues against early investment. They argue for small positions, many shots on goal and genuine scientific diligence.

The Front Door Is Opening

For most of the past decade, the honest answer to “who funds discovery?” was fewer investors every year. AI does not make biology easy. It may make it far cheaper to learn whether an idea is good, and that cost is the variable that pushed capital away from the front of the pipeline.

Investors who want exposure to that shift do not need billion-dollar checks. They need deal flow, scientific judgment and patience. Healthcare Venture Capital Fund tracks financings and acquisitions in its healthcare M&A deals database, and Healthcare Discovery provides the research intelligence on the science behind them. For investors balancing long-dated discovery exposure with steady income, Healthcare Real Estate Fund covers the real estate end of that barbell.

Sources

This article is for informational purposes only and is not an offer to sell or a solicitation of an offer to buy any security or interest in any fund.

Similar Posts