Google Adds TPU 8i Max System With Double the Compute of TPU 8i

Google unveiled the TPU 8i Max on Nov. 9, adding a system with twice the compute and memory bandwidth of its TPU 8i for larger AI workloads.

The 8i Max combines two existing TPU 8i chips in one package. It is a new system configuration, not a new chip design. The TPU 8i is built for real-time AI reasoning, while the TPU 8t serves large-scale model training. Google introduced both in April.

Google also unveiled the TPU 8tx system, which protects sensitive data while AI models process it. The 8tx uses the same silicon as the TPU 8t and pairs with fifth-generation AMD EPYC processors. Together, the two configurations extend Google's existing TPU lineup to cover different processing and security needs.

The additions give Google more ways to offer AI computing capacity to customers using its processors as an alternative to Nvidia GPUs. Its TPU business serves nearly 100 customers. The 8i Max adds capacity within a single package, while the 8tx addresses workloads that require hardware-enforced protection for data in use.

Oracle separately plans to offer Google TPU capacity through Oracle Cloud Infrastructure starting in 2027, expanding access beyond Google Cloud. That plan does not specify that Oracle will offer the newly announced 8i Max or 8tx systems. The new systems themselves are expected to become available in 2027.