AI Scientific Discovery by the Numbers (2026)

The key verified statistics behind AI's takeover of scientific research in 2026: AlphaFold's reach, AI-discovered drug success rates, and what AlphaEvolve saves Google.

AI's impact on science is now measurable in hard numbers: 200 million+ protein structures predicted by AlphaFold and used by more than 2 million researchers, AI-discovered drug candidates passing Phase I trials at roughly 80-90% versus a ~52% historical industry average, and DeepMind's AlphaEvolve continuously recovering about 0.7% of Google's worldwide compute. This page collects the verified statistics - each with its source - that define the state of AI-driven discovery in 2026.

200M+
Protein structures predicted by AlphaFold
Google DeepMind
2M+
Researchers using the AlphaFold database
Google DeepMind
80-90%
Phase I success rate of AI-discovered drugs (vs ~52% industry average)
BCG / Drug Discovery Today
0.7%
Of Google's worldwide compute continuously recovered by AlphaEvolve
Google DeepMind
23%
Speed-up of a key Gemini training kernel found by AlphaEvolve
Google DeepMind

Structural biology: the AlphaFold baseline

AlphaFold has predicted the structures of over 200 million proteins - essentially every catalogued protein - and DeepMind reports its database has been used by more than 2 million researchers in over 190 countries. The 2024 Nobel Prize in Chemistry recognised the breakthrough. Five years after AlphaFold 2, the frontier has moved to predicting full conformational landscapes and routine de novo design of protein binders (Communications Biology, 2026).

Drug discovery: success rates flipped

Analyses of AI-discovered drug candidates report Phase I clinical success rates of roughly 80-90%, against a historical industry average near 52% (BCG analysis published in Drug Discovery Today). The chart below uses the conservative 80% lower bound. AI-originated molecules now form a substantial and growing share of early-stage clinical pipelines.

AI optimising science's own infrastructure

DeepMind's AlphaEvolve - a Gemini-powered coding agent combined with evolutionary search - has discovered new mathematical constructions and, deployed inside Google's infrastructure for over a year, continuously recovers ~0.7% of worldwide compute and accelerated a key Gemini training kernel by 23%. AI is now optimising the very systems that train AI.

Methodology

Every figure on this page traces to a named public source: Google DeepMind publications and blog posts, the Stanford HAI AI Index, peer-reviewed journals (Nature, Science, Communications Biology) and the BCG clinical-success analysis. Figures are reported as published; where sources give ranges we state the range and chart the conservative bound.

Tools mentioned

Related guides

FAQ

Are AI-discovered drugs actually more successful in trials?

In Phase I, yes, so far: BCG's analysis in Drug Discovery Today found AI-discovered candidates succeeding at roughly 80-90% versus a ~52% historical average. Phase II evidence is still maturing, so the long-run advantage is not yet settled.

What is AlphaEvolve?

A Google DeepMind coding agent that pairs Gemini models with evolutionary search. It has found new mathematical constructions and practical optimisations - including recovering ~0.7% of Google's global compute and speeding a key Gemini training kernel by 23%.

Where do these numbers come from?

Named public sources only: Google DeepMind, the Stanford HAI AI Index, Nature, Science, Communications Biology and BCG's published clinical analysis. Each stat on this page links to its source.

More reports

Compiled by ToolGlance from publicly reported data; figures link to their sources. Updated 2026-06-11.

How we rate: ToolGlance scores combine pricing, core features, user-review signals and update frequency, compiled from public sources and vendor documentation — see our methodology. Figures are indicative and change often; always verify pricing and features on the vendor site before buying. Last updated 2026-06-12. Compiled by the ToolGlance editorial team.