By Dan Gallagher

This is an online excerpt of the weekly AI & Business newsletter. Get more insights in your inbox each week by signing up here.

"Frenemies" are not a new concept in tech, where major companies will often work together in some areas while competing fiercely in others. Even in that light, the relationship between Google and Nvidia stands out.

Google has long been one of the biggest customers for Nvidia's artificial-intelligence chips. Sundar Pichai, CEO of the internet search giant, has even specifically called out the company's use of Nvidia's products in 10 of his last 12 quarterly earnings calls, according to data from S&P Global Market Intelligence. There is good reason: Nvidia's chips are in great demand and short supply. Even the deep-pocketed Google has to compete with other cash-rich tech giants who want access to the same components.

All of this makes it seem odder that Google has become one of the biggest thorns in Nvidia's side. Google's Gemini 3 large-language model, released in mid-November, vaulted the company to the top of the AI heap due to its strong performance on AI benchmarks.

More impressive: Google's claim that Gemini 3 was fully trained on its own in-house chips called TPUs. That apparently gave Google some expansionist ideas. Just a week after Gemini 3's launch, the Wall Street Journal reported that the company was shopping its TPU chips around to other companies, including Meta Platforms.

That would effectively make Google a direct competitor to Nvidia. But selling chips is harder than it looks.

Meta ended up signing a new deal with Nvidia. And it is an ambitious one that includes deployment of Nvidia's CPU chips in more traditional server equipment.

CPUs are the main processors of a computing system, and Nvidia's are typically tied into systems with the company's graphics processing chips - or GPUs - that accelerate AI workloads. The Meta deal is the first time that Nvidia's CPU processors will be deployed at such a large scale in systems without the company's GPUs attached.

Google is hardly sitting still. The Wall Street Journal reported that the company is increasing its financial support of data-center partners, including talks to invest around $100 million in a cloud-computing startup called Fluidstack. That also takes a page from Nvidia's playbook, as the chip maker has made its own investments into neocloud operators like CoreWeave, Nebius and Lambda.

But Google also has to get its chips actually made. That is no small matter when Nvidia has tied up much of the manufacturing capacity and necessary components. The top chip-production lines at TSMC are now so dominated by Nvidia that even Apple is having a hard time getting enough capacity for its mobile-device processors. In a recent report, Mark Lipacis of Evercore ISI said Nvidia has locked down production capacity "aggressively" while also being "first in line" for the type of specialized memory that AI systems require.

Google is most definitely an AI powerhouse. Becoming an AI-chip powerhouse is still no simple matter.

This item is part of a Wall Street Journal live coverage event. The full stream can be found by searching P/WSJL (WSJ Live Coverage).

(END) Dow Jones Newswires

February 24, 2026 16:00 ET (21:00 GMT)

Copyright (c) 2026 Dow Jones & Company, Inc.