OpenAI reportedly plans to buy Google's AI chips

OpenAI plans to buy Google's artificial intelligence chips outright, not just rent them, according to a report published Tuesday night. The Information first reported the plan, citing one person familiar with the matter. Reuters and other outlets picked it up Wednesday. Finally, neither OpenAI nor Google has confirmed it, and no price, volume or timeline has been disclosed. Read more: OpenAI reportedly plans to buy Google's AI chips

Google's own AI chips are called Tensor Processing Units, or TPUs. They are built to train and run AI models, an alternative to the graphics processors Nvidia sells.

Reports surfaced in mid-2025 that OpenAI had begun renting Google's TPUs. OpenAI said at the time, however, that it was only testing them, with no plans to deploy them at scale. So, owning the chips instead of leasing server time would mark a deeper commitment to a second chipmaker.

Those reports landed hours before Alphabet, Google's parent company, reported its own third-quarter results on Wednesday. Google Cloud revenue reached $30.71 billion, up 103% from $15.16 billion a year earlier. Meanwhile, total company revenue rose about 25% to $127.6 billion.

Chief executive Sundar Pichai told investors Alphabet is "seeing strong demand for our TPU systems across a wide range of customers."

Related: OpenAI epilepsy lawsuit says ChatGPT coached patient off medication before death

Nvidia's grip on OpenAI

OpenAI remains Nvidia's largest customer by a wide margin. Nvidia has said OpenAI's current and planned commitments add up to about 12 gigawatts of computing power. That figure could grow toward 16 gigawatts. In fact, that relationship, Nvidia says, is worth roughly $600 billion in business through 2030.

Nvidia agreed in August to guarantee up to $105 billion for an OpenAI data center campus in Ohio. The guarantee covers lease and power payments. In exchange, that campus will run only Nvidia hardware, with limited exceptions.

A purchase of Google's chips, even a modest one, would shift some spending toward a second supplier. Still, it would be the first time at real scale. Google has also started selling TPUs directly to other customers this year, including through deals with Anthropic, an OpenAI rival.

Alphabet's finance chief has said the company expects only a small share of TPU sales revenue this year. Most of it will arrive in 2027.

Related: OpenAI cuts live internet access from AI model training

A widening list of suppliers

OpenAI has been diversifying its hardware for more than a year. It started renting Google's TPUs in mid-2025. Then, last week, it said it would make its AI models available through Google Cloud, a business it competes against.

OpenAI is also developing its own chip with Broadcom, unveiled earlier this year. It plans to begin deploying the chip by the end of 2026.

Alphabet shares rose about 1% on Tuesday, when the chip report surfaced, and gained further Wednesday after the earnings landed. Nvidia shares ticked up slightly Tuesday, then fell about 1% Wednesday to $215.38, roughly 9% below their October 6 record.

Still, no deal has been signed. How much OpenAI might buy, when it would arrive and what it would cost all remain unknown.

Oracle, for its part, has said it plans to start offering Google's chips on its own cloud platform in 2027.