
OpenAI has just lately begun renting Google’s synthetic intelligence chips to energy ChatGPT and its different merchandise, a supply near the matter instructed Reuters on Friday.
The ChatGPT maker is likely one of the largest purchasers of Nvidia’s graphics processing models (GPUs), utilizing the AI chips to coach fashions and likewise for inference computing, a course of wherein an AI mannequin makes use of its skilled information to make predictions or selections primarily based on new info.
OpenAI deliberate so as to add Google Cloud service to fulfill its rising wants for computing capability, Reuters had solely reported earlier this month, marking a shocking collaboration between two distinguished rivals within the AI sector.
For Google, the deal comes as it’s increasing exterior availability of its in-house tensor processing models (TPUs), which have been traditionally reserved for inside use. That helped Google win clients together with Huge Tech participant Apple in addition to startups like Anthropic and Protected Superintelligence, two ChatGPT-maker rivals launched by former OpenAI leaders.
The transfer to hire Google’s TPUs alerts the primary time OpenAI has used non-Nvidia chips meaningfully and reveals the Sam Altman-led firm’s shift away from counting on backer Microsoft’s information facilities. It may probably increase TPUs as a less expensive different to Nvidia’s GPUs, in keeping with the Info, which reported the event earlier.
OpenAI hopes the TPUs, which it rents via Google Cloud, will assist decrease the price of inference, in keeping with the report.
Nonetheless, Google, an OpenAI competitor within the AI race, just isn’t renting its strongest TPUs to its rival, The Info mentioned, citing a Google Cloud worker.
Google declined to remark whereas OpenAI didn’t instantly reply to Reuters when contacted.
Google’s addition of OpenAI to its buyer listing reveals how the tech big has capitalized on its in-house AI know-how from {hardware} to software program to speed up the expansion of its cloud enterprise.
© Thomson Reuters 2025