Alphabet is now funding the startup built by the researchers it lost
Alphabet reorganised its entire artificial intelligence command structure on Wednesday and lost four of its most senior technical figures on the same day, a sequence that erased about 4% of the company’s market value and left investors reading two memos for evidence of how much trouble Google DeepMind is in.
Demis Hassabis, the Nobel laureate who co-founded DeepMind in 2010 and continued running it from London after selling the lab to Google for around $650 million in 2014, becomes Alphabet chief scientist and chairman of Google DeepMind.
Koray Kavukcuoglu, the unit’s Chief Technology Officer, takes over daily operations with the title of Senior Vice President, and the CEO title attached to those duties has been retired. Sundar Pichai, Chief Executive Officer of Alphabet and Google, told staff in a companywide memo reported by Reuters that he and Hassabis had been discussing a role allowing him to “put his full attention on actively shaping the future of AGI”.
The memos describe a plan the timeline does not support
Two months earlier, at a University of Washington commencement, Jeff Dean had described to computing graduates how he “got the itch to join a startup in 1999”, which was the year he became employee number 30 at a company then occupying rooms above a Palo Alto storefront. He was speaking as Google Chief Scientist, a 27-year veteran who had co-founded Google Brain in 2011 and whose name sits on the papers that introduced MapReduce, Bigtable, Spanner and the Google File System. On Wednesday he called his time at the company “an amazing journey” and left with three colleagues to build something else.
The something else is Discovery Loop, a public benefit corporation co-founded with Google Senior Fellow Sanjay Ghemawat, Google DeepMind Vice President Oriol Vinyals and Google Brain co-founder Quoc Le, and its purpose is to automate the process of scientific research itself. Dean described the mission on X as an effort to “automate machine learning, science, and engineering to accelerate discoveries and progress”, and explained the naming logic by observing that in many domains “you can fully computerise that whole loop”.
Radical Ventures and Khosla Ventures are co-leading a seed round that remains open, and the Wall Street Journal reported that the company intends to begin with machine learning research before extending into hardware design, drug discovery and clean energy.
Alphabet is funding research it could no longer keep in house
Google is a founding investor in Discovery Loop and will supply its computing power for at least the first year. The arrangement gives Alphabet a stake in work it once owned outright, and it keeps four researchers within reach who might otherwise have taken their next decade to a competitor. Ghemawat said the founders wanted infrastructure built to requirements different from those Google maintains for its consumer and advertising products, which describes the moment a large platform stops being the natural home for the people who built its foundations.
The sector has seen this pattern repeatedly through 2026. Gemini co-lead Noam Shazeer left for OpenAI in June, less than two years after Google paid billions to hire him and his team away from the startup he had been running, and senior research scientist John Jumper, who shared the 2024 Nobel Prize with Hassabis, said in the same week that he would join Anthropic. Capital is now available to any credible research team willing to leave a large employer, and Discovery Loop, whose founders describe themselves as three of the most-cited researchers in artificial intelligence, shows that the people clearing that bar include some who had looked permanent.
Kavukcuoglu reports directly to Pichai and oversees Gemini model development, research, the Gemini app and the developer teams, having spent 13 years inside DeepMind on work including WaveNet and Deep Q-Networks. Withholding the CEO title from him consolidates the division into a business line answerable to the parent company’s chief executive, and it ends the distance DeepMind had maintained from Google’s product cadence for more than a decade.
