The Information: OpenAI plans to buy Google’s TPU AI chips

From renting to buying

The Information reported on Oct. 27, citing a person familiar with the matter, that OpenAI plans to buy TPUs (tensor processing units), the AI chips Google designs in-house. The paywalled exclusive by Ryan Barwick ran under the headline “OpenAI, a Longtime Nvidia Customer, Plans to Buy Google’s AI Chips.” CNBC confirmed the plan with its own sources. According to Reuters, no financial terms were disclosed; OpenAI declined to comment, and Google didn’t immediately respond. The reports don’t say how many chips are involved, which TPU generation or where the chips would run.

OpenAI has long been one of the biggest buyers of Nvidia GPUs. Reuters reported in June 2025 that it had begun renting TPUs from Google; the new report says it now plans to buy them instead of renting. In June of this year, OpenAI unveiled its first custom AI chip, designed with Broadcom, and said it planned to deploy the processor by the end of the year. The two companies compete in AI products and search but have increasingly struck commercial deals: a week earlier, OpenAI said its frontier models would also be offered through Google Cloud.

Google opens up its TPUs

According to CNBC, Google has been developing its own chips since 2013 but began selling TPUs to outside customers only in the second quarter of this year, when it also opened them up to data centers outside its own infrastructure. In a joint announcement on Oct. 27, Google and Oracle said TPUs will be offered through Oracle Cloud Infrastructure (OCI), the first time they will be available on a hyperscale cloud platform outside Google. TPUs on OCI will become available in 2027, ramping over several years. According to CNBC, Oracle said the same day that OCI will also add OpenAI’s Jalapeño chip and hardware from Cerebras.

The announcement says Google’s eighth-generation TPU comes in two architectures, the TPU 8t for training and the TPU 8i for inference. According to Google, the TPU 8t delivers nearly three times the compute performance per pod of the previous generation.

The biggest TPU customer is a rival

CNBC notes that Google’s biggest TPU customer is one of its top AI competitors: Anthropic, whose Claude models compete with Gemini, agreed in October 2025 to use up to 1 million TPUs in a deal worth tens of billions of dollars. In the same week, Broadcom detailed a $63 billion vehicle, backed by Blackstone and three big banks, that will buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years.

Estimates for the TPU business vary widely. According to CNBC, TD Cowen models $9.4 billion in TPU sales for the second half of this year and $62.5 billion for 2027, while Wells Fargo expects about $9 billion in the third quarter and $99 billion next year. Research firm SemiAnalysis estimates that Google’s latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server. On Google’s July earnings call, CFO Anat Ashkenazi said only a small portion of TPU revenue from existing agreements would be recognized this year, with the “vast majority” coming in 2027.