• OpenAI begins using Google TPUs to cut inference costs.
• Reflects shift in OpenAI’s cloud and chip strategy.
• Google withholds latest TPU tech from direct AI rival OpenAI.

OpenAI has recently started renting Google's AI chips to enhance ChatGPT and other products, The Information reported on Friday, referencing a person involved in the arrangement.

The transition, marking OpenAI's initial significant use of non-Nvidia chips, indicates a change in the Sam Altman-led company's dependence on Microsoft's data centers, possibly enhancing the appeal of Google's tensor processing units (TPUs) as a more affordable substitute for Nvidia's graphics processing units (GPUs), the report noted.

OpenAI, being one of the biggest buyers of Nvidia's GPUs, utilizes AI chips for model training and for inference computing, where an AI model applies its learned knowledge to predict or decide based on fresh data.

According to the report, OpenAI expects that the TPUs it leases via Google Cloud will reduce the cost of inference.

According to a Google Cloud employee, The Information reported that Google, a competitor of OpenAI in the AI competition, is not leasing its most advanced TPUs to its rival.

Also Read: Samsung to Launch AI Coding Assistant Cline to Boost Developer Productivity

Neither OpenAI nor Google replied promptly to Reuters' requests for comments.

OpenAI intended to incorporate Google Cloud services to address its increasing demands for computing power, Reuters had reported exclusively earlier this month, signaling an unexpected partnership between two major rivals in the AI industry.

For Google, the agreement arrives as it increases the external accessibility of its internal TPUs, which have traditionally been allocated for internal purposes. This assistance allowed Google to attract clients such as major tech leader Apple and emerging companies such as Anthropic and Safe Superintelligence, both of which are OpenAI rivals founded by ex-OpenAI executives.