Google's AI division has spent 2026 insisting it is still setting the pace in artificial intelligence. On Wednesday it handed the doubters their sharpest data point yet. Demis Hassabis is giving up the chief executive job at Google DeepMind, the London lab he co-founded in 2010 and sold to Google in 2014. Jeff Dean, the company's chief scientist and its 30th employee, is leaving Google altogether after 27 years. Alphabet shares fell 4%.
Axios reported the reshuffle and the market reaction as Google circulated internal memos, and put the stakes plainly: the company is rebuilding its AI organization while trying to close the gap with OpenAI and Anthropic. The slide interrupted a run, since the stock had rallied 13% from a post-earnings low.
Neither man is being fired. Hassabis becomes chairman of Google DeepMind and chief scientist of Alphabet, a newly created title, and keeps running Isomorphic Labs, the drug discovery company Alphabet spun out of DeepMind. Koray Kavukcuoglu, DeepMind's chief technology officer and a 13-year veteran of the lab, takes over daily operations as senior vice president, reporting to Sundar Pichai. That makes Kavukcuoglu, not Hassabis, the person now responsible for shipping Gemini.
Hassabis says the timeline has changed
The line in Hassabis's memo that explains the whole move is about timing. He told staff he has worked toward artificial general intelligence, meaning software that matches or exceeds humans across most cognitive work, for his entire career, and that he now believes "it is close at hand."
Until Wednesday, Hassabis was a product executive who happened to be a Nobel laureate, accountable for a consumer chatbot and a model release calendar. He is now arguing that the interesting problem is no longer shipping the next model but deciding what happens when the technology arrives, and he has restructured his job around that belief. For investors, the practical translation is that Google's most recognizable AI figure has moved away from the part of the business that generates revenue.
Semafor reported that the shift had been at least a year in the making, citing two people familiar with his thinking, and that Hassabis was not pushed out. He had been handing operational work to Kavukcuoglu for months, and struggled with the executive role rather than the science. Set against a flagship model that is late, that is not purely reassuring: the top seat was drifting while the calendar slipped.
What Dean is leaving to build
The Wall Street Journal's Meghan Bobrowsky first reported Dean's departure and the company he is founding. Dean will be chief executive of Discovery Loop, a public benefit corporation, which is a for-profit company whose directors are legally required to weigh a stated mission alongside returns. Three senior colleagues are going with him: Google senior fellow Sanjay Ghemawat, DeepMind vice president Oriol Vinyals, and Google Brain co-founder Quoc Le.
The pitch is to automate the cycle that scientific research runs on: propose an experiment, run it, read the result, propose the next one. Discovery Loop wants machines doing that loop thousands of times over. It will start with machine learning research itself, then move into hardware design, drug discovery, and clean energy, according to the Journal.
Google is a founding investor and will supply the computing power for at least the first year. Radical Ventures and Khosla Ventures are co-leading the seed round, with Lightspeed, Kleiner Perkins and Doerr Capital also participating, according to the law firm that advised on the launch. Dean told The New York Times that being outside a public company gives the group room to make choices that are not strictly optimal for shareholders.
The parts Google did not answer
The financing arrangement invites an obvious question. Alphabet is putting money into a startup that will spend a portion of it buying Alphabet's cloud services, the same circular structure that has drawn scrutiny across the AI industry this year. The seed round has not closed and Discovery Loop declined to give a valuation, so the size of Google's stake is not public.
The competitive picture is the harder problem. Gemini 3.5 Pro is months behind schedule, and some Google sources told Axios that low internal morale is part of the reason. Several senior researchers, including a Gemini co-lead, had already left for rival labs before Wednesday. Hassabis pointed staff to the unreleased Gemini 4 and said its pre-training is going well, according to Fortune's account of the memo, but Google has not given a launch date.
Alphabet reported $119.8 billion in revenue for the quarter ended June 30, up 24%, with Google Cloud growing 82%. None of that was the problem on Wednesday. Investors sold a company with excellent results because four of the people most associated with its future walked out on the same afternoon, and the replacement plan is a promise about a model nobody has seen.