Jeff Dean and Google veterans launch Discovery Loop to automate scientific experiments

Discovery Loop is an AI research company building systems that automate the iterative process of scientific discovery — generating hypotheses, designing and running experiments at massive compute scale, and refining them — starting with machine learning before expanding to other fields.

That AI can automate the experimental loop — hypothesize, design, run experiments in parallel, analyze, iterate — and accelerate science as compute did ML.

On August 5, 2026, Jeff Dean — Google's chief scientist, its 30th employee, and a 27-year veteran — announced on X that he was founding Discovery Loop with Sanjay Ghemawat, Quoc Le, and Oriol Vinyals. The company is a Delaware public benefit corporation whose stated mission is to automate machine learning, science, and engineering to accelerate discoveries and progress.

The initial funding round was co-led by Radical Ventures and Khosla Ventures, with Lightspeed, Kleiner Perkins, Doerr Capital, and Alphabet participating, and the company signed a long-term partnership with Alphabet for cloud and compute resources. Dean is expected to serve as CEO; at Google, Demis Hassabis took the new title of Alphabet chief scientist and moved from DeepMind CEO to chairman in the same reshuffle.

Brand new: as of September 2026 Discovery Loop has disclosed neither the amount nor the valuation of its initial round, and its public agenda is to first automate machine-learning research before expanding into other scientific and engineering disciplines.

The experimental loop is the universal bottleneck: manual, iterative lab work scales poorly while compute and models have scaled exponentially.

The founding team's 14–30 years of shared history at Google gives it rare ability to build and operate massive-scale infrastructure.

Automating machine learning research first is a wedge into every other scientific field that depends on ML.

An Alphabet cloud-and-compute partnership gives the company the scale needed to run thousands of parallel experiments.

Extreme pedigree can skip the proving ground: the field's most-cited researchers and an Alphabet compute partnership let Discovery Loop start from a research agenda, not a product.

As of September 2026, Discovery Loop has only announced its founding and initial round; no funding amount or valuation has been disclosed, and no products or research results are public. Reports of talks to raise about $1 billion at a $10 billion valuation circulated but were not confirmed by the company. The bet is entirely ahead of it: whether automating experimental loops can actually accelerate discovery will be judged by the research it eventually produces.