/startuppedia/media/media_files/2025/11/25/ni-2025-11-25-17-58-34.png)
Jensen Huang and Mark Zuckerberg
Mark Zuckerberg-led Meta is reportedly discussing a major deal with Google to use Google’s custom AI chips, known as Tensor Processing Units (TPUs).
According to media reports, Meta may rent TPUs from Google Cloud as early as next year and deploy these chips in its own data centres by 2027.
If true, this would mark a strategic shift in Meta’s AI infrastructure and raise the stakes for Jensen Huang’s Nvidia, which currently leads the AI chip market.
Meta Google TPU Deal: Why Meta May Shift Away From Nvidia
The possible deal is getting strong attention because it comes at a time when AI companies face rising costs for Nvidia hardware. Google has been promoting its TPUs as a cheaper and steady option that can support large AI models without long wait times or supply limits.
For Meta, this is important. The company trains huge AI models for Facebook, Instagram, WhatsApp and its growing set of assistants. Using TPUs could lower its costs and spread out its hardware risk.
Reports say the plan starts with renting TPUs through Google Cloud next year, giving Meta time to test the chips before placing them in its own data centres in 2027.
Google insiders believe that deals like this could help TPUs gain real market share. Some estimates say Google could take a slice of spending that now goes almost fully to Nvidia.
Nvidia Stock Drop: How the Meta–Google Talks Affected the Market
Markets reacted within hours of the reports. Nvidia’s stock slipped about 2.7% as traders feared that Meta, one of the world’s top buyers of AI chips, could shift future orders away from the company.
Alphabet shares rose by a similar amount as investors saw the talks as a boost for Google’s cloud and AI plans.
So far, Meta, Google and Nvidia have not shared public comments. Still, the possible deal suggests the AI chip race could soon become far more competitive.
/filters:format(webp)/startuppedia/media/media_files/2025/04/16/vcxUBGNzvOnMECRNjn18.jpg)