OpenAI Google TPUs plan would turn an AI rival into a chip supplier

OpenAI, long one of Nvidia's biggest customers, is preparing to buy chips from a company it competes with every day. A plan to bring OpenAI Google TPUs, first reported by The Information on Tuesday, would see the ChatGPT maker buy Google's tensor processing units to add computing capacity.

What happened

Final terms are not yet known, Bloomberg reported, citing a person familiar with the private discussions, and OpenAI and Google declined to comment. Reuters said the original report did not disclose financial terms. Reuters had reported in June 2025 that OpenAI had begun renting Google's AI chips; buying them outright would be a step further.

The same day, Oracle said its cloud would offer Google's TPUs, OpenAI's in-house Jalapeño chip and Cerebras hardware, CNBC reported. Google's custom silicon has never before been offered on a rival hyperscaler's cloud. Oracle shares rose nearly 5% and Cerebras climbed about 8%. Oracle is also adding ChatGPT and Codex to its business applications, and OpenAI chief executive Sam Altman joined Oracle co-CEO Clay Magouyrk on stage at Oracle AI World in Las Vegas. "The next generation of AI will require a diverse range of compute," Google executive Amin Vahdat said of the Oracle arrangement.

Jalapeño, developed with Broadcom, was built with help from Google veterans, Bloomberg noted: Richard Ho, who runs OpenAI's chip team, previously worked on Google's TPUs. OpenAI unveiled its first custom chip in June and plans to deploy it by the end of this year, according to Reuters. For a lab that wants as much compute as possible, OpenAI Google TPUs would sit alongside Nvidia GPUs, Jalapeño and capacity at Oracle's Stargate campus in Abilene, Texas, where CNBC said the GPT-6 Astra model was trained.

Why it matters

The common thread is diversification. Google started direct TPU sales to selected customers earlier in 2026, and deals with Anthropic have fueled demand, Reuters said. Alphabet expects to recognize most of its TPU sales revenue in 2027, CFO Anat Ashkenazi has said. The two companies were already getting closer: last week OpenAI made its models available through Google Cloud, as we covered in our story on the OpenAI Google Cloud deal.

Nvidia is not being cut out. Oracle's cloud chief Mahesh Thiagarajan told CNBC that Nvidia chips remain "the best, most fungible" and that the AI market is very large, adding that chip scarcity is still an issue. Amazon, Microsoft and Alphabet together expect to pour up to $600 billion into AI infrastructure this year, Reuters said, a pace that keeps funding questions alive, including around OpenAI's own $30 billion raise.

What's next

Alphabet reports quarterly results on Wednesday, and investors will listen for cloud growth and TPU demand. Neither company has confirmed the size or timing of any OpenAI Google TPUs order, so the details remain unconfirmed. The scoop is at The Information.