OpenAI in Talks to Buy Google TPU Chips in Push to Break Free From Nvidia Monopoly

In a development that underscores the severe shortage of artificial intelligence processing capacity, OpenAI is in discussions to purchase custom AI semiconductors directly from Alphabet Inc.’s Google, according to sources familiar with the negotiations.

While commercial terms and transaction volumes remain private, the procurement talks mark an unprecedented hardware pact between the tech sector's fiercest generative AI rivals.

Sourcing Compute Across Battle Lines The push to acquire Google's custom silicon—known as Tensor Processing Units (TPUs)—comes as OpenAI scrambles to secure every available gigawatt of compute to train and serve next-generation models:

The TPU Alternative: Google began engineering its proprietary TPUs more than a decade ago to optimize internal search indexing. Over successive generations, the application-specific integrated circuits (ASICs) have evolved into one of the only viable high-scale enterprise alternatives to Nvidia’s flagship graphics processing units (GPUs).

• Broadening Availability: Google has increasingly pushed its TPUs into the broader market through external partnerships, including a major deployment announced with Oracle Corp. to offer TPU clusters via Oracle Cloud Infrastructure.

"The next generation of AI will require a diverse range of compute." — Amin Vahdat, Vice President and Head of Systems and Cloud Infrastructure, Google

The Multipronged Silicon Strategy OpenAI's interest in Google silicon is part of a deliberate, multi-track effort to loosen Nvidia's pricing grip on the frontier AI sector:

• First-Party 'Jalapeno' Silicon: Alongside third-party merchant chips, OpenAI has designed its own in-house inference processors, dubbed Jalapeno, developed in partnership with semiconductor designer Broadcom Inc.

• The Google Lineage: The engineering talent steering OpenAI’s internal hardware division shares deep ties with Alphabet. Richard Ho, who leads OpenAI’s in-house silicon group, previously served as a senior technical lead on Google’s TPU engineering teams, bringing institutional knowledge of custom ASIC fabrication and interconnect fabric directly to OpenAI's roadmap.

With hyperscalers and labs facing relentless compute demand from agentic AI workflows, OpenAI’s willingness to buy silicon from Google confirms that in the race for infrastructure supremacy, pragmatism beats corporate rivalry.