Google announces two TPUs specialized for confidential data and high memory
🎧 Listen to this story
Google announced two new configurations for its eighth generation of AI accelerators: the TPU 8tx, designed for confidential model training and inference, and the TPU 8i Max, aimed at workloads requiring more memory. The family now has four versions, alongside the TPU 8t for training and the TPU 8i for inference and low-latency execution.
Google says the 8tx builds protection into the hardware, keeping sensitive data and model weights isolated during processing. The 8i Max combines two TPU 8i chips in a single package, doubling computing capacity, SRAM, and memory bandwidth. Bloomberg, cited by Xataka, reported that the new variants will be available next year; prices and exact dates were not disclosed.
Why it matters · editorial interpretation
The new configurations broaden the TPU line’s segmentation across training, inference, data protection, and memory, while future availability leaves costs and timing open. Hardware-based protection may also matter to organizations processing sensitive data, although this is Google’s claim.
Sources
Hardware