AI Materials & Chemistry Discovery

Billion-dollar bets on AI-native drug and molecule discovery

The theme's capital story is dominated by landmark institutional commitments to AI-native discovery platforms. Isomorphic Labs — the DeepMind AlphaFold spin-out led by Demis Hassabis — closed a $2.1B Series B backed by Alphabet, Thrive Capital, GV, Temasek, CapitalG, MGX, and the UK Sovereign AI Fund, the largest single round in the theme's history. Simultaneously, Chan Zuckerberg Biohub received $500M from CZI, pushing its total raised past $2B. These are not exploratory bets: Isomorphic has named pharma partnerships with Novartis, Lilly, and Johnson & Johnson and is advancing internal assets toward the clinic, while Biohub's ESMFold2 has demonstrated nanomolar antibody binders in laboratory validation. The concentration of top-tier crossover investors — sovereign funds, Big Tech, and flagship VCs — signals that AI-native discovery is no longer a frontier thesis but an infrastructure investment.

Biohub's ESMFold2 — a protein language model trained on over 6.8 billion proteins — achieved hit rates of 36–88% for compact minibinders and 15–29% for antibody-derived formats against cancer targets in confirmed laboratory binding experiments, representing a decisive crossing from computational benchmark to wet-lab proof. The model's hierarchical architecture (proteins → cells → whole biological systems) reflects a deliberate engineering strategy articulated by Mark Zuckerberg, and mechanistic interpretability techniques from LLM research are now being applied to protein models to extract biological knowledge. The hire of an Evolutionary Scale veteran as Head of Science at CZ Biohub further institutionalizes this protein-engineering pipeline.

Why it matters · Lab-validated computational protein design compresses the discovery-to-candidate timeline dramatically, making AI-bio companies credible partners for pharma and threatening traditional CRO workflows.

Seven seed deals totaling $234M in the last 90 days — anchored by Radical Numerics' $50M seed led by Emergence Capital with Patrick Collison participating, and Albert's $7.5M seed from Index Ventures — reveal that formation-stage capital is flowing into AI-bio at atypical scale. Radical Numerics' $50M seed is explicitly flagged as notable for both its size and its bio-AI thesis. This wave reflects LLMs reaching an inflection point where moving biology from a discovery-based to an engineering-based science is now seen as tractable by top-tier early-stage investors.

Why it matters · A seed wave of this density and average check size typically precedes a Series A/B surge 18–24 months out, giving investors a narrow window to build positions before valuations re-rate.

CuspAI, the U.K.-based materials-discovery AI startup, raised $200M at a valuation over $1B backed by Temasek, becoming the theme's first clear deep-tech unicorn outside of drug discovery. This signals that the AI-accelerated materials design thesis — extending beyond biotech into semiconductors, chemicals, and physical materials — is maturing into its own investment category.

Why it matters · A billion-dollar materials-AI company anchors a new sub-sector benchmark, attracting crossover capital from industrial and deep-tech funds that have historically avoided early biotech.

CZI's open data platforms — the Human Cell Atlas, Cell by Gene (a single-cell RNA corpus that spawned a self-organizing scientific community), and the ESM Atlas of 6.8 billion proteins — are emerging as the foundational training infrastructure for the next generation of transcriptomic and protein AI models. CZI's CRISPR Cures partnership with Jennifer Doudna at UCSF further extends this data-to-clinic pipeline. The openly shared nature of these assets creates a compounding moat: the more researchers contribute, the richer the corpus, and the more powerful the models trained on it.

Why it matters · Whoever controls the canonical open biological datasets controls the baseline for every future foundation model in bio-AI, making data stewardship as strategically important as model architecture.