Artificial intelligence startup Anthropic has deepened its partnership with Google, securing access to up to one million AI chips — worth tens of billions of dollars — to train future versions of its Claude chatbot, the companies announced Thursday. The expanded deal highlights the explosive demand for computing power that underpins the global AI arms race.
Beginning in 2026, Anthropic will gain access to over one gigawatt of computing capacity through Google’s Tensor Processing Units (TPUs) — specialized chips originally reserved for Google’s internal AI research. One gigawatt of compute is estimated to cost around $50 billion, illustrating the sheer scale of Anthropic’s infrastructure bet as it vies to compete with OpenAI and Microsoft in developing the world’s most advanced AI systems.
Anthropic said it selected Google’s TPUs for their superior price-to-performance ratio and energy efficiency, as well as for their proven reliability in powering existing Claude models. In addition to the TPU capacity, Google Cloud will provide expanded cloud infrastructure and AI development tools.
The deal places Anthropic among the largest buyers of high-end AI chips globally, alongside rivals like OpenAI, which has reportedly signed multitrillion-dollar compute agreements involving Nvidia (NVDA.O) and AMD (AMD.O) processors. Analysts say these investments reveal the massive cost of developing frontier AI systems capable of reasoning, problem-solving, and creative generation at near-human levels.
Anthropic, backed by Alphabet (GOOGL.O) and Amazon, is one of the world’s fastest-growing AI firms, with an expected revenue run rate approaching $7 billion next year. The company says its Claude models emphasize AI safety, transparency, and enterprise reliability, distinguishing them from consumer-focused rivals.
