Google expands its TPU range with a focus on data security and memory
On 9 November 2026, Google announced TPU 8tx and TPU 8i Max, two new configurations within its eighth-generation AI accelerator range; Reuters also reported the announcement.[3][4] The primary source makes an important distinction: these are new systems built around existing silicon, not an entirely new chip generation.[3]
## TPU 8tx: protecting data during processing
Google says TPU 8tx targets confidential inference for regulated businesses and governments, using the same silicon as TPU 8t alongside fifth-generation AMD EPYC host CPUs.[3] The company claims hardware-level protection and isolation for sensitive data and model weights even while they are in active use.[3] That is a different consideration from raw computing power: the question is how sensitive information is handled while a model processes it.
Google attributes up to two-times lower latency than TPU 8t to the Boardfly network, with a near-zero performance penalty. These are manufacturer claims, not findings from our own independent testing.[3]
## TPU 8i Max: two chips in one package
TPU 8i Max combines two TPU 8i chips in a single package; Google says it doubles compute, SRAM and HBM bandwidth relative to TPU 8i.[3] Reuters describes the system as aimed at workloads processing large amounts of information, including coding assistants and mathematical reasoning.[4]
The announcement therefore points to specialization: sensitive-data processing on one side and the memory constraints of larger workloads on the other.[3] It is not evidence that either system is the best choice for every business. Assessment requires workload-specific testing and contractual checks; this draft does not assign prices, general availability or independently measured performance that the consulted sources do not establish.
## Sources
[3] https://cloud.google.com/blog/topics/ai-infrastructure/tpu-8tx-and-tpu-8i-max
[4] https://www.sanluisobispo.com/news/business/article317992913.html