Google adds TPU systems for confidential inference and large AI contexts

Google has announced two additions to its eighth-generation TPU portfolio, targeting confidential AI processing and workloads with large context windows. TPU 8tx and TPU 8i Max are new system configurations around existing chip designs, rather than a new processor generation, the company said on November 9.

TPU 8tx uses the same silicon as TPU 8t and pairs it with AMD server processors. Google says hardware-level protections keep sensitive data and model weights isolated during processing. The system is intended for regulated organisations handling private inference, fine-tuning and related AI work.

TPU 8i Max combines two TPU 8i chips in one package. According to Google, this doubles compute, on-chip memory and high-bandwidth-memory bandwidth compared with TPU 8i, while increasing computing density within the existing rack footprint.

These descriptions and performance comparisons come from the vendor, not independent testing. The announcement focuses on how compute, memory, networking and security are assembled into systems. Buyers comparing the options should distinguish those system changes from a new underlying chip generation and test their own workloads before drawing performance conclusions.

Dates: Google published and announced the configurations on 9 November 2026. The supplied briefing is dated 10 November.