Meta confirms TPU inference in production, per SemiAnalysis

SemiAnalysis says Meta has confirmed it is running TPU inference in production, a claim it presents as challenging CUDA's dominance.

The firm says it had reported Meta as a TPU customer on Nov 28, 2025, and that its accelerator supply chain model already included Meta's TPU procurement details.

The only source so far is a single X post from SemiAnalysis (@SemiAnalysis_). The post does not name the Meta statement or document that Meta itself made, so the confirmation rests on SemiAnalysis's account.