On October 23, 2025 Anthropic announced a major expansion of its use of Google Cloud’s Tensor Processing Units, planning to access up to one million TPUs and bring “well over a gigawatt of capacity online in 2026.” Anthropic described the deal as worth tens of billions of dollars. Google Cloud CEO Thomas Kurian said the expansion reflected “the strong price-performance and efficiency” Anthropic’s teams had seen from TPUs.
The agreement notably did not make Google Anthropic’s sole supplier. Anthropic said it runs across three chip platforms - Google’s TPUs, Amazon’s Trainium, and NVIDIA’s GPUs - choosing among them for cost, performance, and efficiency. The announcement came alongside continued growth in Anthropic’s enterprise business, which it said served more than 300,000 business customers.
This expansion sat next to Anthropic’s separate multibillion-dollar compute relationship with Amazon, including the Trainium2-based Project Rainier cluster, underscoring that a single frontier lab now signs overlapping mega-deals with multiple cloud and chip providers at once.
Why business readers should care: the deal shows the scale and multi-vendor nature of frontier compute procurement - a single lab committing to up to a million accelerators from one provider while still buying heavily from two others. It also validated Google’s custom TPUs as a credible alternative to NVIDIA for training and serving leading models.