OpenAI Reportedly Plans to Buy Google's AI Chips, the TPUs

What happened

The account of a person familiar with the matter is the entire basis of this report: The Information cited one such person, with Reuters and Bloomberg relaying the same account the same day, and the source asked not to be named because the talks are private; final terms are not settled, neither company has confirmed a purchase, and at the reporting level this remains a plan rather than a signed deal.

The two companies' relationship swings between competition and cooperation: OpenAI and Google compete directly in models, products and search, but their compute cooperation deepened over the past year — Reuters reported in June 2025 that OpenAI began renting TPUs on Google Cloud to cut ChatGPT inference costs, and on Oct 21, 2026 the two announced OpenAI's frontier models would be offered to enterprise customers through Google Cloud; on Oct 27, Oracle announced the same day that Google TPUs will be offered through Oracle Cloud from 2027.

The supply-side backdrop appears in the same reports: Google's Tensor Processing Units are in-house AI accelerator chips, long used inside Google's own services, which began selling directly to some outside customers this year, and Reuters describes them as gaining traction as a cheaper alternative to NVIDIA's market-leading chips, suited to running models; OpenAI remains one of NVIDIA's largest GPU buyers while also developing its own chip with Broadcom, which Reuters says it plans to start deploying before the end of 2026; Alphabet was due to report quarterly results after the US market close on Oct 28.

Key facts

1. Source: The Information first reported the plan on Oct 27, citing a person familiar with the matter, with Reuters and Bloomberg relaying the same account the same day.
2. Plan: OpenAI plans to buy Google's AI chips, the TPUs, with no amount, quantity or delivery schedule disclosed.
3. Responses: OpenAI declined to comment, and Google did not immediately respond to Reuters' request for comment.
4. Existing rental: Reuters reported in June 2025 that OpenAI had begun renting TPUs on Google Cloud, its first large-scale use of non-NVIDIA chips.
5. TPU positioning: Google's in-house Tensor Processing Units began selling directly to some outside customers this year, and Reuters says they are gaining traction as an alternative to NVIDIA's chips.
6. Same-day link: Oracle announced on Oct 27 that Google TPUs will be offered through Oracle Cloud from 2027, the first hyperscale cloud channel outside Google.
7. In-house chip track: OpenAI is also developing its own chip with Broadcom, which Reuters says it plans to start deploying before the end of 2026.