Google DeepMind: What the Lab Builds and Who Runs It

Google DeepMind merged DeepMind and Google Brain in 2023. A dated look at its Gemini models, AlphaFold, weather work, safety framework, and limits.

Guide: AI Tools and IndustryLast verified: October 20266 min read

Google DeepMind is the unit of Google that builds the company's Gemini models and carries out its fundamental AI research. It was formed in April 2023 by combining the London-based lab DeepMind, founded in 2010, with the Google Brain team, and it is led by Demis Hassabis. Its best-known scientific result, the AlphaFold protein structure model, earned Hassabis and John Jumper half of the 2024 Nobel Prize in Chemistry.

Key takeaways

  • DeepMind started in 2010, Google confirmed its purchase on January 26, 2014, and the lab merged with Google Brain on April 20, 2023, when the combined unit was announced.
  • Gemini, announced on December 6, 2023, was built by the Google DeepMind team with colleagues across Google, including Google Research.
  • AlphaFold predictions for more than 200 million proteins are served through a database that Google DeepMind runs with EMBL's European Bioinformatics Institute.
  • Isomorphic Labs, the drug discovery company, is a separate Alphabet company that grew out of DeepMind's protein work.
  • Budget, headcount, and compute spending for the unit are not published in the primary sources checked for this page.

From a London startup to a division of Google

DeepMind was founded in London by Demis Hassabis, Shane Legg, and Mustafa Suleyman, and its stated aim was general-purpose learning algorithms that combine machine learning with ideas from neuroscience. Google confirmed on January 26, 2014 that it had bought the company. Google did not state a price. Press reports at the time ranged from USD 400 million to more than USD 500 million, so no single figure can be treated as confirmed.

Google Brain had a separate origin. Per the Google DeepMind website, it began in 2011 inside X, the company's research group for long-range projects. For about twelve years the two teams worked as separate groups. On April 20, 2023, Sundar Pichai announced that they would become a single unit, with Hassabis as CEO, and described the goal as the bold and responsible development of general AI. The same website now lists Hassabis as chief executive and co-founder.

Main lines of work

The unit does not have one product. Its output falls into a few families, and each has a different kind of evidence behind it.

Line of workWhat is documentedWhere to check it
Language and multimodal modelsGemini 1.0 was announced on December 6, 2023 in three sizes: Ultra, Pro, and Nano. Hassabis signed the announcement on behalf of the Gemini team.Google's announcement post and the technical report
Protein structureAlphaFold 2 was presented in 2020. The AlphaFold Protein Structure Database, run with EMBL-EBI, holds more than 200 million predictions.The Nobel Committee's press release and the database's own about page
WeatherGraphCast, published in Science in 2023, forecasts hundreds of variables up to 10 days ahead at 0.25-degree resolution and beat a leading operational system on 90 percent of 1,380 verification targets.The Science paper, volume 382
Safety and governanceThe Frontier Safety Framework was published on May 17, 2024. It defines critical capability levels in autonomy, biosecurity, cybersecurity, and machine learning research and development.The framework document

Games were the early showcase, with AlphaGo as the best-known example. Since the merger the emphasis has moved toward general models and toward science problems where a prediction can be checked against a measurement.

The Nobel Prize and what it covered

On October 9, 2024, the Royal Swedish Academy of Sciences gave one half of the Nobel Prize in Chemistry to David Baker (University of Washington) for computational protein design and split the other half between Hassabis and Jumper, both listed as Google DeepMind, London, for protein structure prediction. The Nobel Committee's press release says AlphaFold2 was presented in 2020, that its predictions now cover nearly all of the 200 million proteins researchers have listed, and that more than two million people from 190 countries have used it.

The prize recognized a specific, checkable result: a model that predicts protein structures, presented in 2020 and used by researchers worldwide since. It does not certify the lab's other work, and a Nobel citation is not an endorsement of any product built on a later model.

Safety and oversight inside the organization

The organization describes an internal Responsibility and Safety Council, co-chaired by its chief operating officer, that reviews research and projects against Google's AI Principles. A separate AGI Safety Council, led by co-founder Shane Legg, works on risks from future highly capable systems. The Frontier Safety Framework adds a procedure: run early warning evaluations on frontier models, and apply a mitigation plan when a model approaches a critical capability level. The framework's authors call it exploratory and expect it to change.

These are self-described internal processes. Publishing a framework shows what a lab commits to do, but outside audit of how it is applied is a separate question, and this page did not find an independent review of it.

What is separate from Google DeepMind

Isomorphic Labs, announced on November 5, 2021, is a London-based Alphabet company that applies AI to drug discovery and builds on the AlphaFold work. A company spokesperson said at launch that it is separate from DeepMind and has its own resources, so its drug programs and funding rounds should not be read as Google DeepMind results. Character AI, which has a licensing and cloud arrangement with Google, is also a different company.

Checking a claim about the lab

  • Is the result in a peer-reviewed paper, a preprint, or only a company blog post?
  • Does the headline number compare against a named baseline, and is that baseline current?
  • Was the work done by Google DeepMind, by Google Research, or by a partner such as a university or a public institute?
  • Is the product named (Gemini, AlphaFold Server) the same thing as the research model in the paper?
  • Is a funding, user, or revenue figure attributed to the right entity, given that Isomorphic Labs and Google are different organizations?

FAQ

Is Google DeepMind a separate company from Google?

No. Since April 2023 it has been one unit of Google, formed from DeepMind and Google Brain. Isomorphic Labs is the separate company.

Who runs Google DeepMind?

As of the last check on October 27, 2026, the organization's own website lists Demis Hassabis as CEO and co-founder. Leadership titles can change, so the website is the place to confirm them.

Did Google DeepMind make ChatGPT?

No. ChatGPT is made by OpenAI. Google DeepMind builds the Gemini models, which compete with it.

Where does Google DeepMind work?

It was founded in London. In 2017 the company announced offices in Edmonton and Montreal in cooperation with Canadian universities. The current list of sites could not be confirmed from the primary pages checked, so it is not given here.

Last verified: October 2026

Sources

  • A message from Google and Alphabet CEO Sundar Pichai on the Google Brain and DeepMind merger, Google, April 20, 2023
  • About Google DeepMind and responsibility and safety pages, Google DeepMind, accessed October 2026
  • The Nobel Prize in Chemistry 2024, press release, Royal Swedish Academy of Sciences, October 9, 2024
  • Introducing Gemini, Google, December 6, 2023
  • Introducing the Frontier Safety Framework, Google DeepMind, May 17, 2024
  • About the AlphaFold Protein Structure Database, EMBL's European Bioinformatics Institute, 2026
  • Lam R et al., Learning skillful medium-range global weather forecasting, Science 382, 2023
  • Google acquires artificial intelligence company DeepMind, press reports, January 26 to 27, 2014
  • DeepMind announcement on Canadian research and a Montreal office, Google DeepMind, October 6, 2017
  • Alphabet launches Isomorphic Labs, CNBC, November 2021

New to artificial intelligence?

Start with the basics, then follow the guides to tools, rules, and effects on society.

Start with AI Basics