OpenAI has commenced renting artificial intelligence chips from Google to enhance the functionality of ChatGPT and its other applications, a source familiar with the situation disclosed to Reuters on Friday.
As one of the leading consumers of Nvidia’s graphics processing units (GPUs), OpenAI utilizes these AI chips for both training models and inference computing. Inference computing allows an AI model to apply its trained knowledge to generate predictions or decisions drawn from new data.
Earlier this month, Reuters reported that OpenAI intends to incorporate Google Cloud services to address its escalating demand for computational power. This development marks a notable partnership between two major competitors in the artificial intelligence arena.
For Google, this agreement represents an expansion of the availability of its proprietary tensor processing units (TPUs), which were previously used only for internal projects. This strategic move has allowed Google to attract clients such as Apple and emerging startups like Anthropic and Safe Superintelligence, both of which were founded by ex-OpenAI executives.
The decision to lease TPUs from Google indicates a significant shift for OpenAI, as it diversifies its reliance on Nvidia chips and reduces dependence on Microsoft’s data centers. This shift could position TPUs as a cost-effective alternative to Nvidia’s GPUs, according to a report from The Information.
OpenAI is optimistic that the TPUs, accessed via Google Cloud, will help reduce the costs associated with inference, as outlined in the report.
Nevertheless, it is important to note that Google, while partnering with OpenAI, is not providing its most advanced TPUs to the company, as reported by The Information, referencing a Google Cloud employee.
While Google has refrained from commenting on the deal, OpenAI did not respond promptly to Reuters’ inquiries.
The addition of OpenAI to Google’s clientele underscores the tech giant’s ability to leverage its in-house AI capabilities, spanning from hardware to software, to fuel the growth of its cloud computing services.
© Thomson Reuters 2025
