Google Cloud Says Its TPU Business Is More Than Twice the Size of Rivals

Google Cloud CEO Thomas Kurian says the company’s artificial-intelligence accelerator business is now more than twice the size of the next-largest AI accelerator operation among rival hyperscalers. Speaking at a Goldman Sachs conference, Kurian described Google’s Tensor Processing Unit, or TPU, business as a rapidly expanding part of the company’s cloud strategy. The comments underline Google’s growing ambition to compete in AI infrastructure alongside companies such as Amazon Web Services and Microsoft.

Google has developed TPUs for more than a decade, originally using them primarily to power its own products and AI systems. The company has now expanded the technology into a broader commercial business, allowing external customers to access TPU computing through Google Cloud while also preparing systems that can operate in customers’ own data centres. Google introduced its eighth-generation TPU family this year, with the TPU 8t designed for training and the TPU 8i focused on AI inference.

Kurian said Google is pursuing several ways to monetise the technology. Customers can rent TPU computing through Google Cloud, while selected large organisations can receive TPU systems for deployment in their own facilities. Google has also indicated that its external TPU strategy is intended to expand the addressable market beyond its hosted cloud infrastructure. The company has already identified major AI laboratories and other organisations as potential customers for the technology.

The expansion comes as demand for specialised AI computing continues to increase. Google says its TPUs are designed to offer strong performance and efficiency for demanding workloads, while its broader AI infrastructure also includes its Axion processors and NVIDIA GPUs. The company is therefore positioning itself as a full-stack AI infrastructure provider, combining chips, data centres, cloud services, models and software rather than depending entirely on third-party accelerators.

Google’s growing TPU business could become an important competitive advantage as AI companies spend heavily on computing capacity. Kurian has said Google does not view TPU usage by outside AI laboratories as competing with its own Gemini business because the company can generate revenue from multiple parts of its technology stack. With newer TPU generations being deployed and external sales expanding, Google is increasingly turning hardware originally built for its own AI systems into a major commercial business.