What is actually being reported
The Information reported on Tuesday, citing a person familiar with the matter, that OpenAI plans to buy Google's AI chips. Reuters, which could not independently confirm the report, notes that no financial terms or volumes were disclosed; Google did not respond to a request for comment and OpenAI declined to comment. Bloomberg subsequently reported the plan as part of OpenAI's build-out.
The distinction matters: OpenAI has rented Google TPUs since at least June 2025, when Reuters reported the arrangement. Buying is a different relationship — a capital commitment, a supply contract, and a statement that Nvidia is not the only option on the table.
Why it is a bigger deal than one purchase order
OpenAI is one of Nvidia's largest customers, and Nvidia is one of OpenAI's investors. A purchase from Google inserts a competitor's silicon into that relationship at the exact moment when every large buyer is trying to escape a single-supplier position.
It also completes a thaw that would have been unthinkable two years ago. Last week OpenAI said it would make its frontier models available through Google Cloud. Microsoft remains its primary cloud partner; the two facts now coexist, which tells you how much of the current AI economy runs on deals between rivals.
For Google, it is the vindication of a strategy that took years to become visible: TPUs have been sold directly to some customers since this year, Anthropic's demand has been a large part of the volume, and chief financial officer Anat Ashkenazi has said Alphabet expects to recognise a small percentage of TPU sales-agreement revenue by the end of this year, with the vast majority arriving in 2027. Oracle, according to TrendForce, intends to bring TPUs into its cloud in 2027 — meaning Google is becoming an arms dealer to the industry rather than merely its own supplier.
The reasons to be careful
One report, one anonymous source, no numbers. "Plans to buy" covers everything from a pilot rack to a multi-year programme, and there is no evidence this displaces Nvidia for training, where CUDA, tooling and institutional familiarity remain decisive. OpenAI also has its own silicon: in June it unveiled a first custom chip designed with Broadcom, slated to deploy by the end of this year as part of a multi-generation programme.
The likeliest reading is portfolio logic rather than a switch. Every large buyer wants two or three suppliers — for price leverage, for capacity, and because the memory shortage is already teaching them what single-source dependency costs. Nvidia's position is not that a competitor has arrived; it is that its largest customers now have somewhere else to go, and are willing to be seen going there.
Context for scale: Amazon, Microsoft and Alphabet are expected to spend as much as $600 billion combined on AI infrastructure this year, and investors are increasingly asking when that converts into returns. Alphabet reports quarterly results on Wednesday.
