OpenAI Plans to Buy Google's AI Chips as It Looks Beyond Nvidia: Report
OpenAI plans to buy AI chips from Google, according to a person who knows about the plan and spoke to The Information. This move would make OpenAI less dependent on Nvidia, the company that supplies most of the computer hardware used in the AI boom.
Neither company has confirmed the plan. The price, the number of chips, and when they would be delivered have not been shared. OpenAI is one of Nvidia's biggest customers.
The chips OpenAI reportedly wants to buy are Google's tensor processing units, or TPUs. These processors are built for one main task: doing the math that AI models need. Google introduced the chip in 2016 and said it had already been using TPUs in its data centers for more than a year.
Reports from June 2025 said OpenAI had started renting Google's TPUs. At the time, the company described this as an early test and said it had no plans to use them on a large scale. Buying the chips would mean owning the hardware instead of paying Google to use it. Other details about the plan have not been made public.
A job listing from OpenAI in San Francisco asks for an engineer to lead work on making its most advanced AI models run better on TPUs. This gives some support to the reports. The job involves working on compilers, which are programs that turn an AI model into instructions a particular chip can understand. This helps explain why changing chips is harder than simply replacing one part with another.
Richard Ho, who leads OpenAI's hardware program, worked at Google for nearly nine years. Before joining OpenAI, he was one of the leaders of Google's Cloud TPU project.
In September 2025, Nvidia signed a letter of intent to invest up to $100 billion in OpenAI as the company deployed at least 10 gigawatts of Nvidia systems. A gigawatt equals one billion watts. Data center builders now use this unit to describe the size of their projects.
Two weeks later, AMD gave OpenAI the right to buy up to 160 million shares of AMD, or about 10% of the company. This right depends on OpenAI using 6 gigawatts of AMD chips.
Nvidia eventually invested $30 billion in an OpenAI funding round, far less than the $100 billion described in the letter of intent. In March, Nvidia CEO Jensen Huang said this might be the company's last investment of this kind before OpenAI becomes a public company.
In June, OpenAI introduced Jalapeño, its first custom-made chip, developed with Broadcom. The chip is designed for inference, which is when a trained AI model produces answers. OpenAI plans to begin using it by the end of 2026.
Google introduced two eighth-generation TPUs in April. The 8t is designed to train AI models, while the 8i is designed for inference, or running trained models. Google's performance figures compare these chips with its previous-generation chip, called Ironwood, rather than with Nvidia's hardware.
Anthropic, OpenAI's main competitor, signed an agreement with Google and Broadcom in April for several gigawatts of capacity from next-generation TPUs. Anthropic expects this capacity to start becoming available in 2027.
Alphabet began delivering TPU systems to customers for use in their own data centers in the second quarter, according to its chief financial officer. Last week, Google Cloud also began letting a select group of customers test OpenAI's GPT-6 models in a private preview.