Google DeepMind
UK-headquartered AI research laboratory owned by US Alphabet that builds the Gemini model family and created AlphaFold, earning its researchers the 2024 Nobel Prize in Chemistry.
Add to a list
No lists yet.
What it is
Google DeepMind is the UK-headquartered artificial intelligence research laboratory wholly owned by US Alphabet Inc. Formed in April 2023 by merging Google Brain (founded in the United States in 2011) and DeepMind (founded in London in November 2010), it brought approximately 6,000 researchers under a single mandate led by CEO Demis Hassabis. The lab develops the Gemini family of large language models, which power Google Search, Google Cloud's AI services, and the Gemini consumer assistant used by roughly 750 million people as of mid-2026. Alongside commercial AI, it conducts foundational research in reinforcement learning, protein biology, materials science, and weather forecasting, making it unusual among AI labs in spanning both frontier products and basic science.
History
DeepMind was founded in London in November 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman, combining expertise in neuroscience, computer science, and machine learning. Google acquired the company in January 2014 for approximately US$500 million, one of the largest acquisitions of a European AI startup to that point. The lab achieved global recognition in March 2016 when AlphaGo defeated world Go champion Lee Sedol 4-1, the first time a computer program had beaten a top professional without handicap. In July 2022, AlphaFold released predicted three-dimensional structures for over 200 million proteins, providing a freely accessible resource to the global biological research community. Hassabis and AlphaFold co-creator John Jumper received the 2024 Nobel Prize in Chemistry for that work, among the first Nobel awards given for an AI-produced scientific tool. Google consolidated Brain and DeepMind in April 2023, giving Hassabis effective control of all Alphabet AI research and product development.
Current state
As of July 2026, the central commercial product is the Gemini model line at version 2.5 Pro and 2.5 Flash. Gemini 2.5 Flash was optimised for agentic tasks, with improved tool-calling, extended context, and sub-second response latency at low cost. The lab reported revenues of £1.33 billion in 2024 with an operating profit of £217 million, reflecting its dual role as a research institution and a commercial AI supplier to Alphabet's cloud division. Isomorphic Labs, a sister company spun out of DeepMind in 2021, is advancing AlphaFold-derived drug-discovery programmes with the first clinical candidates expected in 2026. In December 2025, Google DeepMind signed a research collaboration with the UK government, committing to open an automated materials-science laboratory on UK soil in 2026 focused on superconductors. Gemma open-weight models, available from February 2024 onward, extend the Gemini architecture to researchers and edge devices.
Relationships
Google DeepMind is a wholly owned subsidiary of Alphabet, with Google Cloud as its primary route to market for AI services. It competes directly with OpenAI, which is developing custom AI chips via the Jalapeño programme with Broadcom, and with Anthropic, whose planned 10 GW compute buildout is backed by rival Amazon infrastructure. A sustained wave of researcher departures to Anthropic through the first half of 2026 included AlphaFold co-creator John Jumper and Gemini lead Jonas Adler, both drawn by pre-IPO equity at Anthropic. Hassabis told Semafor in June 2026 that Google was still winning the talent competition but acknowledged the pressure. On coding agent benchmarks, Gemini models trade leads with OpenAI and Anthropic through early 2026, with no sustained gap favoring any single lab.
What to watch
Whether AlphaFold-derived drug candidates advance through Phase I trials in 2026 will test if DeepMind's scientific outputs translate to clinical and commercial value beyond AI tooling. US AI export controls, whose scope expanded following the June 2026 export-control episode, may restrict international availability of Gemini APIs. Talent retention is a structural question: Alphabet's retention packages are large, but competitors can offer pre-IPO equity that public-company stock cannot match. The UK automated materials-science laboratory, due in 2026, will be a first test of whether fully autonomous AI-driven research can meaningfully accelerate discovery cycles in hard science, a proof point with implications well beyond DeepMind's own roadmap.