Google has launched a new artificial intelligence chip designed to support AI applications. The chip, a tensor processing unit (TPU) named Ironwood, is the first of its kind.
Ironwood was unveiled at the Google Cloud Next conference in Las Vegas this week. Per Google Cloud Vice President Amin Vahdat, it is the most powerful and energy-efficient TPU to date and enables larger scale development and deployment of inference-based AI models.
The chip is designed to work in configurations of 256 or 9,216 chips, with a maximum computing power exceeding the world’s largest supercomputer, El Capitan, by more than 24 times. This enables it to support the high-performance computing requirements of Google’s latest AI model, Gemini 2.5.
Ironwood is now one of the few alternatives to NVIDIA’s AI processors, allowing Google to strengthen its AI cloud infrastructure and reduce its dependence on NVIDIA hardware. In the future, it will additionally support NVIDIA’s AI superchip Vera Rubin.