Google adds TPU variants for confidential computing and long-context AI
Bloomberg reports that the TPU 8tx targets sensitive workloads, while the TPU 8i Max handles applications requiring large amounts of input data.
Google is expanding its AI processor lineup with two specialized chips, Bloomberg News reported. The TPU 8tx is intended for confidential computing, while the TPU 8i Max supports AI applications that process extensive text or data within one request.
Amin Vahdat, Google’s chief technologist for AI infrastructure, told Bloomberg that the designs adapt existing processors to meet customer requests. The 8tx targets government agencies and businesses in regulated sectors. Under a Google partnership with AMD, it operates alongside AMD chips, according to Bloomberg.
The additions bring Google’s eighth-generation TPU family to four processors. They derive from the TPU 8t and TPU 8i, unveiled in April at Cloud Next. Those two chips have separate roles: the 8t trains AI models, and the 8i operates models after training.
Google provides access to TPUs through cloud rentals rather than standalone chip sales. Its supply chain includes four design partners, among them Broadcom and MediaTek. Bloomberg reported that the new variants address the industry’s shift toward AI agents—tools that perform tasks for users—as businesses encounter increasing security risks when deploying them.
