OpenAI Moves to Buy Google TPUs in Bid to Reduce Nvidia Dependence

OpenAI is moving to purchase Google's Tensor Processing Units, according to reporting by The Information on October 27 and confirmed by Reuters the following day. The deal's terms, including volume and price, haven't been disclosed, and talks are ongoing.

The procurement matters because of what it says about the architecture of OpenAI's compute stack. The company is already among the largest buyers of Nvidia GPUs in the world, and it has a custom inference chip, codenamed Jalapeno, built with Broadcom, that it's targeting for deployment by end of year. Buying TPUs isn't a replacement for either of those bets; it's a third leg on a stool that didn't used to have three legs.

The diversification logic is straightforward. As reported by Reuters citing people familiar with the matter, OpenAI has been renting Google's AI chips since June 2025, so this would convert an existing capacity relationship into an ownership one. Owning the hardware gives OpenAI more scheduling control and, presumably, better unit economics at scale than renting. The question of which workloads land on TPUs vs. Nvidia GPUs vs. the Broadcom custom chip is the operational detail that matters, and nothing disclosed so far answers it.

Google's side of the trade is equally interesting. As noted in The Information's reporting, Google only began selling TPUs directly to external customers this year. The customer list already includes neocloud providers and, according to reporting reviewed by the Seoul Economic Daily, companies such as Anthropic, Apple, and Citadel Securities have been leasing them for several years. Oracle separately announced this week it will become the first major cloud provider outside Google to offer TPUs to customers, buying them directly and making them available on its own infrastructure. OpenAI is a major Oracle cloud customer, so the two announcements sit close together even if the supply chains are distinct.

For Google, landing OpenAI as a buyer is commercial validation of a different kind than landing Anthropic. OpenAI is Nvidia's most prominent customer. If it's shopping for alternatives, that's a signal to every other large buyer that TPUs are a credible option at production scale, not just a cost-reduction experiment.

The person running OpenAI's internal chip team is Richard Ho, who previously worked on TPU development at Google. That lineage matters for the engineering feasibility of the move: optimizing frontier model workloads for TPU architecture is non-trivial, and according to The Information, that work is still at an early stage. OpenAI has posted job listings specifically seeking engineers with hands-on experience optimizing real training and inference workloads on TPUs.

The broader context is the one everyone in this industry is watching. As reported by Reuters, Amazon, Microsoft, and Alphabet combined plan to spend up to $600 billion on AI infrastructure this year, while investors increasingly ask when those outlays produce returns. Gartner, separately, has estimated that more than $3.5 trillion in AI infrastructure investment could be impaired by 2030 if AI revenue doesn't keep pace with spending. OpenAI buying its compute from multiple vendors instead of one is partly a supply chain hedge and partly a negotiating posture. Nvidia knows it, Google knows it, and the terms of the eventual deal will reflect that.

Sources cited:
- The Information (https://www.theinformation.com)
- Reuters / Detroit News (https://www.detroitnews.com/story/tech/2026/10/27/openai-plans-to-buy-googles-ai-chips-the-information-reports/92470838007/)
- Business Standard / Bloomberg (https://www.business-standard.com/technology/artificial-intelligence/openai-plans-to-buy-chips-from-google-as-part-of-its-ai-build-out-126102800194_1.html)
- Seoul Economic Daily (https://en.sedaily.com/international/2026/10/28/openai-weighs-buying-google-tpus-easing-nvidia-reliance)
- The News (Pakistan) / Reuters wire (https://www.thenews.com.pk/latest/1421576-openai-plans-to-buy-google-ai-chips-to-reduce-nvidia-reliance-report)