According to a new report from Reuters, ChatGPT maker OpenAI is turning away from Nvidia’s AI chips for the first time. Instead, the AI giant has started renting Google’s AI chips to fuel its products, including ChatGPT.
This comes as a notable change for OpenAI, which has been one of the biggest purchasers of Nvidia GPUs for AI training and inference computing, the process of applying AI models to real-world tasks. The company is now also turning to Google’s infrastructure to meet its rapidly growing computing demands.
Earlier this month, Reuters reported that OpenAI planned to integrate Google Cloud services into its operations, signaling an unexpected partnership between two fierce rivals in the AI industry.
For Google, the move is part of a broader effort to make its custom-designed tensor processing units (TPUs) available to outside clients. Once used primarily for internal projects, TPUs are now helping Google attract a wide range of customers, including tech giant Apple and AI startups like Anthropic and Safe Superintelligence—both of which were founded by former OpenAI executives.
OpenAI’s decision to begin renting Google’s TPUs is also a shift away from using data centers operated by its primary backer, Microsoft. According to The Information, which first reported the development, this could position TPUs as a more cost-effective alternative to Nvidia’s GPUs. The report also claims that using these Google chips, OpenAI aims to reduce the high costs associated with inference.
However, despite this new collaboration, Google is reportedly not offering its most advanced TPUs to OpenAI. Citing a Google Cloud employee, The Information reported that the tech giant is holding back its top-tier chips from a direct competitor in the intensifying AI race.