Google to pump over $100B into AI, says DeepMind boss

When rumors broke that Microsoft and OpenAI were drafting plans for a staggering $100 billion supercomputer dubbed Stargate, everyone wondered how Google would respond.

Now we have an answer, straight from the top of Google's AI empire.

Speaking at a TED conference in Vancouver, Google DeepMind CEO Demis Hassabis was asked point-blank about the reported $100 billion rival project. His reaction was characteristically measured, yet remarkably aggressive: while Google doesn't share specific figures, he believes the company will eventually pump more than $100 billion into AI over time.

Even bolder, Hassabis insisted that Google's actual computing power currently outclasses Microsoft's or anyone else's in the game.

Where all that cash goes

Hassabis didn't map out a exact timeline or line-item budget, but Google's spending blueprint is already taking shape across hardware and infrastructure.

On the hardware front, custom silicon is taking center stage. Google recently introduced its Arm-based Axion CPUs, claiming they run 30 percent faster than the fastest general-purpose Arm instances and outperform standard x86 processors by 50 percent in speed and 60 percent in efficiency. Independent benchmarks are still pending, but building custom silicon is clearly Google's bid to escape third-party chip premiums.

Then comes the physical brick-and-mortar footprint. Google isn't necessarily pouring all $100 billion into a single monolithic machine, but it is aggressively expanding capacity, including a planned $1 billion data center in Kansas City.

DeepMind itself will soak up a healthy chunk of that budget on the software side. Hassabis candidly noted that his team's breakthroughs require immense compute, calling it one of the core reasons DeepMind sold to Google back in 2014.

While some recent DeepMind initiatives have faced scrutiny—such as AI-discovered materials yielding limited practical utility and weather models falling short of replacing conventional forecasts—the lab continues making strides in complex domains like teaching AI models social skills.

Why it matters

This isn't just executive posturing; it signals that the threshold for frontier AI development has officially entered the 12-figure territory.

For enterprise customers and developers, Google's willingness to match or exceed rival spending means infrastructure capacity will keep expanding, while in-house chips like Axion could eventually keep cloud operational costs in check. The flip side is clear: frontier foundation research is now exclusively a playground for balance sheets capable of absorbing historic capital expenditures.

We think Hassabis played this question masterfully. By avoiding an exact spreadsheet figure while claiming Google will spend more over time, he blunted the hype around Stargate without committing to a rigid engineering delivery date. Claiming current compute dominance also reassures enterprise clients who feared Google had fallen behind in the infrastructure race.

However, the real risk lies in monetization. Pumping over $100 billion into silicon and data centers is an unprecedented gamble, and if the practical enterprise utility of these models fails to scale at the same exponential rate, the pressure on Google's margins will become impossible to ignore.