DeepMind CEO: Google will spend $100+ billion on AI

DeepMind CEO Demis Hassabis says Google's processing power exceeds rivals'.

Speaking at a TED conference in Vancouver, DeepMind CEO Demis Hassabis said Google will spend $100 billion or more on AI over time.

According to Bloomberg, Hassabis - who leads DeepMind, the AI lab Google acquired in 2014 - was responding to a question about the $100 billion supercomputer Microsoft is allegedly planning to build with OpenAI, known as Stargate.

Hassabis said, "We don't talk about our specific numbers, but I think we're investing more than that over time."

Although he didn't say exactly how that money would be spent, or how quickly, he stressed that Google's processing power is higher than its rivals like Microsoft, Meta or AWS.

Presumably some of this money will go to DeepMind, which is not only Google's premier AI workplace, but now has to compete directly with Microsoft for talent in its London base.

For years, DeepMind has been practically uncontested in the London talent pool. Not only does Google generally have a reputation as an innovative workplace, but it could offer salaries higher than nearly anyone else could afford.

With Microsoft announcing its own London AI office last week, DeepMind - via Google - will have to get used to digging more deeply to attract and retain top talent.

Not that all of the money will go to talent. DeepMind has pushed the boundaries of AI in recent years, using the tech for everything from eye scans to controlling nuclear fusion.

One of its more recent developments was the release of Gemma, a family of small AI models that can run on laptops and PCs without sacrificing performance.

Not that all of DeepMind's developments have been a hit. Its material discovery work, Graph Networks for Materials Exploration (GNoME), was found to promise more than it delivers, and DeepMind itself said that its weather prediction model can't replace traditional forecasts yet.

Regardless, all of this work needs a ton of compute, which was one of the original reasons DeepMind teamed up with Google 10 years ago. Even if the company doesn't spend $100 billion on a single supercomputer, we're almost certain see the company adding more processing power to its portfolio.