Meta Explores Google TPU Chips in Potential Blow to Nvidia’s AI Dominance
Meta Platforms Inc. is in advanced discussions to deploy Google's tensor processing units (TPUs) across its data centers by 2027, signaling a strategic shift in AI infrastructure procurement. The social media giant may begin renting TPU capacity through Google Cloud as early as 2025, according to sources familiar with the matter.
Nvidia shares fell 3.2% in premarket trading following the report, while Alphabet gained 2.1% as markets reacted to the potential redistribution of AI chip market share. Meta's planned $70-72 billion AI infrastructure spend for 2024 positions it as a bellwether for industry procurement trends.
Google's TPUs represent the most credible challenge yet to Nvidia's GPU hegemony, with Anthropic already committing to purchase up to 1 million units. The custom chips, first developed for internal use in 2018, have evolved through multiple generations specifically optimized for machine learning workloads.
