Jeff Dean's Discovery Loop in Talks for $1B at $10B to Automate Science
Jeff Dean’s post-Google venture is moving fast. Discovery Loop — the public benefit corporation Dean described on X as focused on automating “machine learning, science, and technology” — is in talks to raise $1 billion at a $10 billion valuation, according to people familiar with the matter.
Dean announced his departure from Google in early August after 27 years, the last several as head of Google DeepMind’s research division. Discovery Loop has been described as one of the most talked-about new startups in Silicon Valley in the weeks since, though details have been sparse. Dean’s public statement positioned it as a mission-driven entity — a public benefit corporation structure that trades exit flexibility for mission alignment.
A $10B Bet on Research Automation
A $10 billion pre-money valuation at $1 billion raised is a notable anchor for a company that has not shipped a product or published a paper. It reflects two things: the market’s willingness to price in Dean’s credential directly, and the scarcity of credible deep research infrastructure bets.
Dean’s career at Google includes co-designing the Tensor Processing Unit, co-authoring the Transformer paper (indirectly — he supervised and enabled much of the work that produced it), and building the systems infrastructure that made large-scale ML training practical. If Discovery Loop’s pitch is that research automation requires someone who understands both the systems and the science, the team has a plausible case to make.
The public benefit corporation structure is notable. It limits how cleanly investors can extract returns in a typical exit scenario, which either signals that Dean is filtering for patient capital or is willing to accept a valuation premium from investors who want the reputational association.
The Details Not Yet Disclosed
No product, no team announcement beyond Dean, and no timeline for close have been made public. “Discovery Loop” as a name suggests iterative research cycles — the model of spinning through hypothesis, experiment, result, and synthesis that characterises scientific progress. Whether that means an AI agent system, an automated lab platform, or something closer to a research OS has not been specified.
The $10 billion valuation assumes the round closes at the currently discussed terms. Deal details could change.