The Information Reports OpenAI Plans to Buy Google TPUs; Neither Company Confirms

The Information reported on October 27 that OpenAI plans to purchase Google's Tensor Processing Units (TPUs), citing a person familiar with the matter. Reuters relayed the story the same day. As of this writing, Google did not immediately respond to a request for comment and OpenAI declined to comment [1]. No financial terms, volume, delivery schedule, or deal structure were disclosed [1][2].

What the report says—and what it doesn't

The Information's headline frames the move as a diversification by a longtime Nvidia customer. The only supporting detail publicly circulated is a job listing for an engineer to "lead the optimization" of OpenAI models on TPUs [2]. The report uses the verb "buy," a shift from June 2025 coverage—also citing The Information—which said OpenAI had begun renting Google TPUs through Google Cloud [1][2]. At that time, an OpenAI spokesperson said the company was in early testing with no plans to deploy at scale, and a Google Cloud employee said the most powerful TPUs were not being rented to OpenAI [2].

Neither company has confirmed a purchase agreement, letter of intent, or active negotiations. The existence of a job listing is consistent with buying chips, renting cloud capacity, or simply building expertise before a decision [2].

Google's TPU strategy is shifting

Google has historically consumed most of its TPU output internally. Its Cloud TPU page describes the chips as a service customers use on Google Cloud [2]. Selling physical chips to an external buyer—especially a direct competitor in chatbots, coding agents, and foundation models—would mark a strategic shift toward merchant silicon supply. Alphabet CFO Anat Ashkenazi previously said the company expects to recognize a small percentage of revenue from TPU sales agreements by end of 2026, with the vast majority in 2027 [1]. Anthropic is a widely reported TPU customer [1][2].

The same day as the OpenAI report, Oracle announced that Google TPUs will arrive on Oracle Cloud Infrastructure in 2027, the first hyperscaler outside Google to offer them [2]. Whether the two announcements are connected is not known.

OpenAI's compute diversification is already underway

OpenAI has been expanding its hardware portfolio for months. In June 2026 it unveiled a custom AI chip designed with Broadcom, targeting deployment by year-end as part of a multi-generation program [1]. It also runs a Cerebras-backed "fast tier" for inference and has an Oracle Cloud Infrastructure plan adding Cerebras capacity [2]. Compute chief Sachin Katti announced his departure, and outlets have reported a $750 billion compute plan through 2030—a figure not published by OpenAI [2].

Diversification makes operational sense: GPU supply has been constrained, and a credible second source changes negotiating leverage with any supplier. But the report does not say OpenAI is replacing Nvidia; it frames the move as adding an alternative [1][2].

What remains unknown

Whether a signed deal, LOI, or formal negotiation exists
The TPU generation, chip count, delivery timeline, or whether chips would sit in Google data centers, OpenAI facilities, or partner sites such as Oracle's
Whether the plan covers training, inference, or both
Pricing or commercial terms
Any connection to the Oracle announcement the same day

Silence from both companies on a single-source report is common and does not confirm or refute the claim [2].

Why it matters if true

If Google sells TPUs directly to OpenAI, it signals confidence in its silicon roadmap and a willingness to arm a rival. For OpenAI, it would be the highest-profile example yet of a frontier lab building a multi-vendor hardware stack—alongside Amazon's Trainium, Microsoft's Maia, Cerebras, and its own custom silicon. For customers, the chip underneath the API remains invisible; any impact on capacity, pricing, or speed tiers is speculative until deployment details emerge [2].