What happened
Meta Platforms is reportedly evaluating the integration of Google's Tensor Processing Units (TPUs) into its data centres by 2027, with potential rental via Google Cloud commencing as early as 2025. This initiative introduces Google's decade-old TPU architecture as a viable alternative to Nvidia GPUs for training and running complex AI models. Google has affirmed its continued support for both its custom TPUs and Nvidia GPUs, while Broadcom, as Google's ASIC partner, is also implicated.
Why it matters
This development introduces a new dependency on Google's proprietary TPU hardware and associated cloud services for Meta's AI infrastructure, increasing the complexity of hardware procurement and platform integration. It raises due diligence requirements for evaluating multi-vendor AI accelerator strategies and managing potential vendor lock-in risks. Platform operators, IT security, and procurement teams will bear the oversight burden of integrating and maintaining a diversified hardware ecosystem and managing data flows across potentially new cloud boundaries.
Subscribe for Weekly Updates
Stay ahead with our weekly AI and tech briefings, delivered every Tuesday.
