Google launches eighth-generation TPU 8t and 8i chips

Google has announced its eighth-generation Tensor Processor Units, TPU 8t and TPU 8i, at Google Cloud Next.

• The two chips are purpose-built for different workloads: TPU 8t for massive training and TPU 8i for low-latency inference and agentic workloads.
• Google says both chips are coming soon and will be generally available later this year; customers can request more information now.
• TPU 8t is described as a training powerhouse with up to 9,600 chips in a superpod, two petabytes of shared HBM, 121 ExaFlops of compute, and nearly 3x compute performance per pod versus the previous generation.
• TPU 8i is described as the reasoning engine, with 288 GB HBM, 384 MB on-chip SRAM, 19.2 Tb/s ICI bandwidth, and 80% better performance-per-dollar than the previous generation.
• Google also says both chips use Axion Arm-based CPU hosts, support common AI frameworks, and are part of Google’s AI Hypercomputer stack.