Demis Hassabis states that Google DeepMind needed to revert to its startup origins following the merger with Brain.

      In summary: Demis Hassabis, during an early April 2026 appearance on the 20VC podcast with Harry Stebbings, discussed how Google DeepMind has increased its momentum in the last two to three years by integrating Google Brain's computational resources with DeepMind's research approach and returning to what he referred to as a “startup or entrepreneurial” style of operation. He also mentioned that he manages Isomorphic Labs, the group’s pharmaceutical AI spinoff, during a “second workday” starting around 10pm, in anticipation of upcoming human trials in oncology later this year.

      Gathering the elements

      The formal integration of DeepMind and Google Brain was completed in 2023. Hassabis described this period as one of intentional acceleration, aligning talent from across the company to move in a unified direction, gaining access to the scalable computing infrastructure that DeepMind previously lacked, and driving what he termed a “relentless sort of focus and pace.” He emphasized that this transformation demanded a cultural shift as much as a structural one: the organization needed to return to its startup or entrepreneurial roots, becoming scrappier, quicker, and more efficient in its output. He characterized the current competitive landscape as “ferocious,” stating that long-time employees with careers spanning 20 to 30 years remarked that it’s “the most intense environment they’ve ever seen, perhaps in all of technology.”

      Hassabis noted that he talks to Sundar Pichai, CEO of Alphabet, “every day,” highlighting how Google DeepMind now resides at the core of Alphabet's operational and research strategies. This proximity is complemented by a capital investment of significant scale. Google’s expansion of its computing capabilities, partly through custom chip partnerships with companies like Broadcom, is crucial to this strategy: Alphabet allocated $91.4 billion to capital expenditures in 2025 and anticipates $175 billion to $185 billion for 2026, nearly doubling, with supply limitations cited as the main constraint rather than a lack of available capital.

      The 90% assertion

      One of Hassabis’s more bold statements during the podcast involved DeepMind’s role in the AI landscape. He claimed that about 90% of the breakthroughs that form the basis of the modern AI industry originated from either Google Brain, Google Research, or DeepMind. This assertion aligns broadly with academic records regarding foundational developments, including the transformer architecture created by Google Brain in 2017, early advancements in reinforcement learning from human feedback, and the deep reinforcement learning techniques developed by DeepMind. The 2024 Nobel Prize in Chemistry, awarded to Hassabis and John Jumper, shared with David Baker for the AlphaFold protein-folding system, stands as the most officially recognized of these accomplishments. Whether the 90% figure is accurate is subject to interpretation, especially since the industry has expanded significantly since those early publications. This framing serves as both a positioning statement and a historical assertion.

      The operational impact of this legacy is a markedly increased frequency of product releases. Google’s open-weight model initiative, most recently with Gemma 4, now provides models developed through the same research and training infrastructure as Gemini 3, bridging the gap between cutting-edge research and open-source contributions that previously existed. By the end of the fourth quarter of 2025, Gemini had approximately 750 million monthly active users, with reports indicating that Gemini 3's release in November of that year triggered an urgent internal response at OpenAI.

      The second workday

      In addition to his leadership at Google DeepMind, Hassabis oversees Isomorphic Labs, the pharmaceutical AI spinoff launched by DeepMind in 2021. He detailed his work setup during the 20VC discussion: a primary workday at DeepMind followed by a “second workday” starting around 10pm focused on Isomorphic’s drug discovery initiatives. This dual role reflects his belief that applying AI in drug discovery is both his most significant long-term goal and an endeavor that necessitates continuous personal involvement rather than delegation.

      In April 2025, Isomorphic secured $600 million in funding and has established partnership agreements with Eli Lilly and Novartis, with cumulative milestone values reaching up to $3 billion. In February 2026, the firm launched IsoDDE, a drug design tool that Isomorphic claims enhances the accuracy of AlphaFold 3 for generating drug candidates. Human clinical trials in oncology are anticipated later in 2026. The competitive landscape in AI-driven drug discovery is becoming more challenging across the sector; for example, Anthropic’s acquisition of Coefficient Bio for about $400 million in April 2026, a startup formed by former Genentech computational biology researchers, indicates that general-purpose AI companies are beginning to view pharmaceutical discovery as a legitimate product category rather than merely a showcase of model capabilities.

      The competitive context

      The conversation on the 20VC podcast, alongside Sebastian Mallaby’s biography of Hassabis titled “The Infinity Machine,” published on 31 March 202