Recently, OpenAI started renting Google’s AI chips to power ChatGPT and other applications, according to a person familiar with the situation.

The company that creates ChatGPT is one of the biggest buyers of Nvidia’s GPUs. It employs the AI chips for both model training and inference computing, which is the process by which an AI model applies its prior knowledge to forecast or decide on the basis of fresh data.

In an unexpected partnership between two of the biggest rivals in the AI space, OpenAI intended to integrate Google Cloud service to accommodate its expanding demands for processing power, as Reuters had exclusively revealed earlier this month.

The agreement for Google coincides with the company’s efforts to make its internal tensor processing units (TPUs), which were previously only available for internal usage, more widely available to the public. As a result, Google was able to attract clients from Big Tech giant Apple as well as startups like Anthropic and Safe Superintelligence, two rivals of ChatGPT that were founded by former OpenAI executives.

The decision by OpenAI to rent Google’s TPUs marks the company’s departure from dependence on Microsoft’s data centers, its investor, and the first significant use of non-Nvidia chips. OpenAI is run by Sam Altman. According to the Information, which first reported the development, it might make TPUs a more affordable option than Nvidia’s GPUs.

According to the report, OpenAI thinks the TPUs, which it rents through Google Cloud, would significantly reduce the cost of inference.

However, according to a Google Cloud employee quoted by The Information, Google, an OpenAI contender in the AI battle, is not leasing its most potent TPUs to its competition.

The fact that Google has added OpenAI to its list of clients demonstrates how the internet behemoth has leveraged its proprietary AI hardware and software to boost the expansion of its cloud business.