Overview
- Google has moved from internal use to selling and leasing TPUs externally, with Anthropic reported to be planning deployments exceeding one million TPUv7 chips via a mix of direct purchase and Google Cloud commitments.
- Investor sentiment shifted after reports that Meta is considering TPU orders and after strong reception for Gemini 3, driving an intraday drop in Nvidia shares and gains for Alphabet.
- Analyst models cited in industry coverage estimate TPUv7-based systems deliver roughly 30–44% lower total cost of ownership than Nvidia’s GB200-class deployments, positioning cost as a key purchasing driver.
- Google is reducing software friction by backing native PyTorch execution on TPUs via XLA with eager mode and by investing in open-source inference tooling to broaden adoption.
- Nvidia has responded with public claims of platform breadth and released Alpamayo-R1, an open-source vision-language-action model for autonomous driving research unveiled at NeurIPS.