Google adds TPU 8tx and 8i Max systems

Google expanded its eighth-generation Tensor Processing Unit portfolio with two new systems, TPU 8tx and TPU 8i Max, the company said on November 9, 2026. Google called this the first time it has introduced four TPU systems in a single year. Codema’s TPU vs GPU explainer covers how TPUs differ from GPUs; the new systems sit inside that eighth-generation family. The broader snapshot of GPUs, TPUs and HBM is in our AI chips and memory guide.

TPU 8tx is built for secure and confidential inference, and Google also says it works for small-scale training. It uses the same silicon as TPU 8t and pairs with 5th Generation AMD EPYC CPU hosts, Google wrote, so sensitive data and model weights stay protected and isolated while in use. Google said TPU 8tx adopts the low-latency Boardfly network used for real-time reasoning and delivers up to 2x lower latency than TPU 8t with a near-zero performance penalty. That latency claim is Google’s comparison of 8tx to 8t, not a third-party benchmark.

TPU 8i Max packages two TPU 8i chips on a single package. Google said the design delivers double the compute, double the SRAM and double the HBM bandwidth of TPU 8i, aimed at large context windows and higher compute density per rack. Those “double” figures are Google’s claims versus 8i; they are not independent lab measurements. Google said fitting twice the silicon in the existing rack footprint lets data centers deploy more compute per square foot and per watt.

The four eighth-generation systems Google lists are TPU 8t for scale-up training, TPU 8i for low-latency reasoning, TPU 8tx for confidential computing, and TPU 8i Max for massive context windows and rack density. Google first introduced 8t and 8i at Google Cloud Next in April; today’s pair brings the count to four for the year. The chips will need somewhere to run, and Google has been adding capacity, including Crusoe’s Wyoming data-center site.