OpenAI reportedly plans to buy Google TPUs as Oracle adds them to its cloud

A reported purchase from Google

According to a person familiar with the matter, OpenAI plans to buy AI chips from Google, a report The Information published on Tuesday, 27 October. Financial terms and other details of the planned purchase have not been disclosed. OpenAI declined to comment to Reuters, and Google had not responded when Reuters followed up. The Next Web said it had not independently verified the report. Bloomberg also reported that the final terms of any deal are not yet known. A purchase would move OpenAI, one of the largest buyers of Nvidia's graphics processors, toward a rival supplier of AI processors.

Oracle opens its cloud to Google's TPUs

Oracle said it will make Google Cloud's tensor processing units available through Oracle Cloud Infrastructure, so customers can choose the hardware they use to train and run AI models without going through Google Cloud. The offer covers the eighth-generation TPU 8t, built for training, and TPU 8i, built for inference, both launched by Google Cloud earlier this year. Those chips are set to reach OCI in 2027. The partnership was announced at Oracle's annual AI World conference, and Oracle is the first cloud vendor outside Google to sell the chips. OCI President Mahesh Thiagarajan has argued that AI workloads differ in their compute needs, which is why customers gain an option beside Nvidia's GPUs. Nvidia remains the dominant supplier of AI accelerators, with a market capitalisation of $5.265 trillion.

Google AI and Infrastructure senior vice president and chief technologist Amin Vahdat put the rationale in one line.

“The next generation of AI will require a diverse range of compute.”

Google's wider TPU sales

In April, Google said it would sell TPU hardware directly to a small group of customers who run the chips in their own data centres. Demand has risen with deals involving Anthropic, and Google has sold so much TPU capacity that its own researchers have had to wait for access. Alphabet finance chief Anat Ashkenazi expects a small share of TPU sales revenue to be booked by the end of 2026, with most of it arriving in 2027.

OpenAI's own chips and its history with TPUs

OpenAI is also building its own silicon, a custom AI chip designed with Broadcom that it unveiled in June and plans to put into use by the end of 2026. Pb.pl reports that OpenAI's semiconductor team is led by Richard Ho, who previously worked on Google's chips. The relationship has shifted before. In June 2025, Reuters reported that OpenAI had begun renting Google's chips, but soon afterwards OpenAI said it had no active plans to use them in its products. The two companies compete in AI products and search, yet their business links are expanding, and OpenAI said its frontier models would become available on Google Cloud after a similar step with Amazon Web Services.

Spending backdrop

Reuters estimates that Amazon, Microsoft and Alphabet could together spend as much as $600bn on AI infrastructure in 2026. Alphabet was due to report its quarterly results on 28 October, the same day the OpenAI report and the Oracle announcement drew coverage.