Google develops AI chips 24 times faster than supercomputers

Google introduces its most efficient Tensor Processing Unit (TPU) called Ironwood, designed specifically for AI models. This TPU is built to help AI models work faster and smarter, especially for work that requires reasoning or prediction. Unlike its predecessors, which were designed to train Ai models from scratch, Ironwood was born to serve inference activities, Interesting Engineering reported on April 10.

Launched at the Google Cloud Next ’25 event, the 7th generation TPU model provides computing power 24 times faster than any supercomputer in the world when deployed at scale. "Ironwood will support the next phase of generative AI and its massive compute and communication requirements," said Amin Vahdat, Google vice president and general manager of ML, systems and cloud AI. "This is what we call the 'age of inference' when AI agents proactively access and produce data to collaboratively provide insights and answers, rather than just the data."

Ironwood comes with an impressive range of specifications. This processor can provide 42.5 exaflops of computing power when scaling up to 9,216 chips/bay, far exceeding the world's fastest supercomputer El Capitan (1.7 exaflops). Ironwood also has 192GB High Bandwidth Memory (HBM), 6 times more than Trillium, Google's previous generation TPU, announced last year. Ironwood can achieve 7.2 terabits/second per chip in terms of memory bandwidth, a 4.5 times increase compared to Trillium.

Google also describes Ironwood as the foundation for all of its high-end AI models, including Gemini 2.5. At the conference, Google introduced Gemini 2.5 Flash, a lower-cost version of its flagship model, which adjusts the depth of reasoning based on the complexity of the suggestion. While Gemini 2.5 Pro is aimed at high-end work like drug discovery and financial modeling, Flash is optimized for fast everyday use, when speed is of the utmost importance. Ironwood is part of Google's larger AI infrastructure strategy. The company also introduced Cloud WAN, a management system that allows businesses to directly access Google's network infrastructure.

An Khang (According to Interesting Engineering)