Demis Hassabis Says Google Leads AI Talent Race as Researchers Defect

Three days after reports emerged that four Google DeepMind researchers had moved to Anthropic, Demis Hassabis gave Semafor an exclusive on June 23: Google, he said, was still winning the AI talent contest. The claim deserves scrutiny, because the organisational structure Hassabis runs is unlike anything else in the field.

What Hassabis said, and the context around it

In a Semafor exclusive published June 23, Hassabis countered the narrative of a talent drain at Google DeepMind, saying the lab continues to recruit competitively. The interview followed an Eastern Herald report on June 26 detailing that four DeepMind researchers had joined Anthropic within a single week.

The departures are real but need scaling. Google DeepMind employs approximately 3,000 researchers following the 2023 merger of Google Brain and the original London-based DeepMind. Four departures in a week is a fraction of one percent of headcount. What matters more is the seniority and specialisation of those who leave, which neither Hassabis nor the subsequent reports have specified by name.

The broader structural context: OpenAI itself was seeded in large part by Google Brain alumni, and Anthropic was founded by former OpenAI researchers. The wave of researcher exits from big-compute labs to well-funded startups is not new; what is new is that Anthropic's $47B ARR valuation and $965B implied enterprise value now makes it a credible destination for researchers who want both mission and financial upside.

Isomorphic Labs: the retention argument Hassabis can make that others cannot

The most concrete card Hassabis holds is Isomorphic Labs. In May 2026, the company closed a $2.1 billion Series B led by Thrive Capital, with participation from Alphabet, GV, MGX, Temasek, CapitalG, and the UK Sovereign AI Fund, bringing its total raised to $2.7 billion. The round followed a $600 million Series A in 2025, also steered by Thrive.

Isomorphic's pharma partnerships give the funding concrete commercial footing. According to FierceBiotech, the Eli Lilly collaboration is structured as $45 million upfront with up to $1.7 billion in milestone payments; the Novartis deal carries $37.5 million upfront and up to $1.2 billion in milestones. Both partnerships task Isomorphic with using AlphaFold 3 to discover small molecules for targets the companies have long considered undruggable.

Hassabis posted on X after the Series B closed: "I've always believed the No.1 application of AI should be to improve human health. That work started with AlphaFold, and now at @IsomorphicLabs with the mission to reimagine drug discovery and one day solve all disease! We are turbocharging that goal with $2.1B in new funding." The target for filing its first Investigational New Drug application is end of 2026. In a separate interview he said: "What I can say is that we're in pre-clinical with a bunch of our programs, and it's going very well."

For a researcher attracted by scientific mission rather than equity alone, the promise of putting AI-designed molecules into human trials for the first time is a recruitment story no pure-software lab can match. The 2024 Nobel Prize in Chemistry, shared by Hassabis and John Jumper for AlphaFold, signals to the scientific community that this work is measured against the highest academic standards, not just product metrics.

The Gemini business: what DeepMind's compute scale actually buys

The second part of Hassabis's argument is more prosaic but financially substantial. Alphabet's Q1 2026 earnings filing with the SEC reported Google Cloud revenue of $20.0 billion, up 63% year-on-year. The acceleration was attributed to enterprise AI solutions and infrastructure, the exact stack that Google DeepMind's Gemini models power. Alphabet's total consolidated revenues rose 22% to $109.9 billion in the same quarter.

The Gemini App now counts over 750 million monthly active users, and DeepMind's models process more than 10 billion tokens per minute through direct API consumption, according to Alphabet's Q1 2026 disclosure. These numbers represent the commercial output of the 3,000-person organisation Hassabis oversees, and they give Alphabet reason to keep capital investment moving. The company's 2026 capital expenditure guidance is $180 to $190 billion, much of it directed at AI infrastructure.

For DeepMind researchers, that compute envelope is a practical advantage. Training frontier models at the scale needed to compete with OpenAI's GPT-5 class and Anthropic's Claude 4 class requires infrastructure that most well-funded startups cannot replicate. Hassabis made exactly this argument when DeepMind was originally sold to Google in 2014 for $650 million: that Google's infrastructure was necessary to pursue the research at the required scale. In 2026, the argument is structurally the same, the numbers are just larger.

What it means

Hassabis is managing two distinct value propositions simultaneously: Gemini, a consumer and enterprise AI product with 750 million users and a $20 billion-per-quarter cloud revenue tailwind, and Isomorphic Labs, a scientific computing operation with Nobel-credentialed leadership and its first clinical drug candidates due this year. Neither proposition alone is unique; the combination is. The talent question is not whether Google DeepMind loses researchers, because it does and will. The question is whether the dual mandate, compute scale, and scientific mission make the loss rate lower than it would otherwise be. Hassabis's answer, delivered to Semafor six days ago, is that it does.

Sources

• Semafor: DeepMind Chief Demis Hassabis says Google's still winning AI talent (June 23, 2026)
• SEC: Alphabet Q1 2026 Earnings Release (Form 8-K)
• FierceBiotech: Alphabet's AI biotech Isomorphic Labs bags $2.1B Series B (May 2026)
• FierceBiotech: Isomorphic stacks deals with Lilly and Novartis worth nearly $3B
• PR Newswire: Isomorphic Labs secures $2.1 billion funding (May 2026)
• Eastern Herald: Google Loses Four Researchers to Anthropic in a Week (June 26, 2026)
• Demis Hassabis on X: Isomorphic Labs $2.1B announcement (May 2026)

Editorial standards: every claim is sourced. Tips: [email protected]