Photo Credit: Getty Images
Google has announced the broad rollout of Ironwood, its seventh-generation Tensor Processing Unit (TPU), marking the company's strongest challenge yet to Nvidia's dominance in artificial intelligence hardware. The chip, first introduced for testing in April, will become widely available in the coming weeks.
Built in-house, Ironwood is designed to handle both training and inferencing for large-scale AI models. Google says the chip is more than four times faster than its predecessor, Trillium, and capable of connecting up to 9,216 chips in a single pod. This setup, the company adds, removes data bottlenecks and allows customers to scale the most demanding AI workloads efficiently.
The release underscores Google's ambitions to expand its footprint in AI infrastructure, an area increasingly contested by Microsoft, Amazon, and Meta. While Nvidia's GPUs remain the backbone of most large AI systems, Google is betting that its custom silicon can offer advantages in price, performance, and energy efficiency.
Major clients are already lining up. AI startup Anthropic will reportedly use up to one million Ironwood TPUs to power its Claude models. Google said early responses from developers have been "overwhelmingly enthusiastic," citing strong price-performance improvements.
Alongside Ironwood, Google previewed new Arm-based Axion CPUs and virtual machine instances aimed at improving cost efficiency for general workloads. The company said the mix of specialized AI accelerators and flexible CPUs is essential to support rapidly evolving model architectures.
Google Cloud posted third-quarter revenue of $15.15 billion, a 34% increase from the previous year, and raised its capital spending forecast to $93 billion amid soaring AI demand. CEO Sundar Pichai said TPU-based and GPU-based products have become major growth drivers.
Analysts say Google's TPUs are now the most credible alternative to Nvidia's GPUs. D.A. Davidson's Gil Luria estimates the TPU business, if combined with Google DeepMind, could be worth up to $900 billion. Melius Research's Ben Reitzes called Ironwood "the most proven ASIC yet," signaling Google's growing leverage in the AI chip race.
