OpenAI Clarifies Plans Regarding Google’s TPU Chips

OpenAI has made it clear that it currently has no intention of utilizing Google’s in-house artificial intelligence chips, referred to as Tensor Processing Units (TPUs), in its products, despite recent speculation.

The clarification comes in response to a report from The Information which stated that OpenAI was investigating the possibility of using Google’s TPUs to accommodate the increasing demand for AI computing. A representative from OpenAI confirmed to Reuters on Sunday that while the company is engaged in “early testing with some of Google’s TPUs,” it has “no plans to deploy them at scale.”

While it is common for artificial intelligence companies to experiment with various chips, rolling out new hardware extensively usually requires considerable modifications to system architecture and software integration. OpenAI remains predominantly reliant on Nvidia’s graphics processing units (GPUs) and has also incorporated chips from AMD to cater to its expanding computational needs.

Moreover, the company has collaborated with Google Cloud to bolster its computing capacity, as noted by Reuters earlier in June. This partnership marks an intriguing alliance between two direct competitors in the AI sector. Nonetheless, it is anticipated that the majority of OpenAI’s computational power will continue to be sourced from GPU servers provided by CoreWeave, a specialised cloud services provider.

As the AI competition intensifies, OpenAI is making strides to lessen its dependency on third-party chips. The company is currently in the process of creating its own custom AI processor and is on track to achieve a pivotal “tape-out” milestone later this year. The tape-out milestone indicates that a chip’s design is finalized and ready to move to manufacturing. Recent reports from Taiwanese media outlet Commercial Times suggest that OpenAI’s proprietary chip, being developed in conjunction with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC), may debut as soon as the fourth quarter of 2025.

In the meantime, Google has been extending access to its previously internal TPU chips to external customers, which has already drawn significant clients, including Apple, along with competitors of OpenAI such as Anthropic and Safe Superintelligence, both of which were established by former OpenAI executives.

The surging demand for high-performance AI chips, especially those produced by Nvidia, has transformed the global technology landscape and propelled Nvidia to become the world’s most valuable publicly traded company earlier this year. As competition escalates and costs soar, OpenAI and other entities are actively seeking to diversify their hardware approaches.