DeepMind

DeepMind is an AI research laboratory founded in London in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman. Google acquired it in January 2014 — reported at about GBP 400 million, though the price was never disclosed — and merged it with Google Brain in April 2023 to form Google DeepMind. Starting from reinforcement learning, it has worked outward into game-playing AI, protein-structure prediction, and general-purpose models (Gemini). On 5 August 2026 Hassabis stepped out of the CEO role to become Chair of Google DeepMind and Chief Scientist of Alphabet, with the lab's CTO, Koray Kavukcuoglu, taking over day-to-day leadership as SVP.

From an Atari paper to Gemini 3

Hassabis brought a PhD in neuroscience and a background in games; Legg, a theoretical account of what AGI would mean; Suleyman, social-entrepreneurship experience. The founding theme was general intelligence informed by neuroscience, and the early backers included Peter Thiel, Founders Fund, and Horizons Ventures. The December 2013 NIPS paper "Playing Atari with Deep Reinforcement Learning" introduced the Deep Q-Network (DQN), and Google announced the acquisition the following month. Press reporting at the time said the deal included a commitment to an independent ethics board.

Nature ran a February 2015 cover paper on a single DQN that reached human-level play across 49 Atari games. In March 2016 AlphaGo beat Lee Sedol 4-1. AlphaZero unified Go, shogi and chess through self-play (2017); AlphaFold (2018 and 2020) took protein-structure prediction to practical accuracy, work recognised in the Nobel Prize in Chemistry announced on 9 October 2024, one half to David Baker for computational protein design and the other half jointly to Hassabis and John Jumper for structure prediction. The Physics prize announced the previous day, to Hopfield and Hinton, was a separate award.

From 2022 the lab ran LLM development in parallel with Google Brain — Chinchilla (compute-optimal scaling), Sparrow, Gemini. The April 2023 merger created Google DeepMind, followed by Gemini 1.0 (December 2023), Gemini 1.5 (2024), Gemini 2.5 (March 2025), and Gemini 3 (November 2025). Speaking on the 20VC podcast in April 2026, Hassabis said the merged organisation had to "come back to almost our startup or entrepreneurial roots and be scrappier, be faster, ship things really quickly."

Games, science, general models, robots, molecules

The AlphaFold protein-structure database publishes over 200 million predicted structures — effectively every known protein — and Google DeepMind states that it is used by more than 3 million researchers in over 190 countries.

Key events

• 2010: Founded in London.
• January 2014: Google acquires DeepMind; reported at about GBP 400 m.
• February 2015: Nature paper — one DQN plays 49 Atari games at human level.
• March 2016: AlphaGo defeats Lee Sedol.
• October 2017: AlphaGo Zero learns without human games; AlphaZero follows in December.
• December 2019: Co-founder Suleyman leaves DeepMind for Google, becoming VP of AI product management and AI policy from January 2020. He co-founds Inflection AI in 2022 and becomes CEO of Microsoft AI in March 2024.
• November 2020: AlphaFold 2 wins CASP14 decisively.
• July 2021: AlphaFold 2 code and database released; Isomorphic Labs spun out.
• April 2023: Merged with Google Brain into Google DeepMind.
• May 2024: AlphaFold 3 released, extending predictions to protein-ligand and other complexes.
• 9 October 2024: Hassabis and Jumper share the Nobel Prize in Chemistry with David Baker.
• July 2025: An advanced version of Gemini with Deep Think is officially certified at gold-medal standard at the International Mathematical Olympiad (five of six problems, 35 points). The 2024 result, from AlphaProof and AlphaGeometry 2, was silver-medal standard (four problems, 28 points).
• November 2025: Gemini 3 released.
• February 2026: Isomorphic Labs unveils its IsoDDE drug-design engine; a US$2.1 bn Series B led by Thrive Capital follows in May.
• 5 August 2026: Hassabis becomes Chair of Google DeepMind and Chief Scientist of Alphabet; Koray Kavukcuoglu becomes SVP. Jeff Dean and Sanjay Ghemawat announce their departure from Google the same day.

Key metrics

• AlphaFold database: over 200 million predicted structures; more than 3 million researchers in over 190 countries (Google DeepMind's figure)
• Citations: the 2021 AlphaFold 2 paper in Nature has passed 34,000 citations in total, the most-cited Nature paper published between 2020 and 2024 (Nature, 2025)
• Gemini app monthly active users: over 750 million in February 2026; over 900 million announced at Google I/O in May 2026
• Parent infrastructure: Alphabet raised its full-year 2026 capital-expenditure guidance to US$195-205 bn in July 2026
• Isomorphic Labs: US$2.1 bn Series B, May 2026

An ethics board that never appeared

DeepMind established the three-layer template — games, science, general-purpose — that has come to define modern AI research. AlphaGo persuaded a wide public that machines could match human strategic intuition; AlphaFold demonstrated the "AI for Science" genre. Both successes occurred inside a commercial parent, and the timing and scope of open releases have repeatedly been contested. AlphaFold 3 was initially published without code, drawing academic criticism before a research-use licence was rolled out.

The same tension runs through Isomorphic Labs. Its February 2026 announcement claims that IsoDDE "more than doubles the accuracy of AlphaFold 3 on the most difficult systems" on the Runs N' Poses benchmark, which measures generalisation to novel pockets and ligands, and that on a novel antibody-antigen test set it "outperforms AlphaFold 3 by 2.3x" in the high-fidelity regime (DockQ > 0.8). These are the company's own claims on its own chosen benchmarks, and have not been independently reproduced.

Organisationally, the independent ethics-board commitment reported at the time of the acquisition has never had a visible role, and former staff have described a loss of research independence since the merger. The exit of co-founder Suleyman, Geoffrey Hinton's 2023 departure from Google to speak freely about AI risk (Hinton was at Google Brain rather than DeepMind), and the August 2026 departures of Jeff Dean and Sanjay Ghemawat alongside Hassabis's move to the chair all point to high turnover at the top.

On military use, the 2014 acquisition was widely reported to include a no-military-or-intelligence undertaking. In February 2025 Google revised its AI Principles, dropping the explicit pledge not to pursue weapons or surveillance applications, and drew renewed protest inside and outside the company. The biosecurity risk that structure-prediction tools might assist weapons design continues to be examined as part of frontier-model evaluations.