Particle.news
Download on the App Store

Google’s TPUs Enter the Spotlight as Meta Explores Access, Drawing a Public Response From Nvidia

Google’s in-house chips now power Gemini, signaling a push to offer an alternative to GPU-centric AI infrastructure.

Overview

  • Meta is reported to be in talks to rent Google Cloud TPU capacity as soon as next year, with potential purchases for its own data centers starting in 2027.
  • Nvidia publicly downplayed competitive threats, saying it remains a generation ahead and continues supplying Google, as its shares fell more than 3% after the reports.
  • Independent comparisons indicate Google’s TPU v5p pods can outperform high-end Nvidia systems on certain workloads tuned to Google’s software stack.
  • Google trained Gemini on TPUs and gains greater control over cost and availability, with wider cloud access to TPUs positioned as a possible shift in AI training economics.
  • Investor reassessments have extended beyond Nvidia, with AMD shares pulling back on reduced GPU demand concerns and Broadcom seen benefiting from its role codesigning TPUs.