Google unveiled its eighth-generation TPUs as two separate designs: the TPU 8t for pre-training and the TPU 8i for inference. It is the first time the company has split a TPU generation. Google expects general availability later in 2026.
Google has unveiled its eighth-generation Tensor Processing Units as two chips: the TPU 8t for large-scale pre-training and the TPU 8i for inference and high-concurrency reasoning. The announcement came at Google Cloud Next on April 22, 2026. It marks the first time Google has split a TPU generation into distinct designs.
Google framed the launch around the so-called agentic era, in which AI systems juggle many tasks at once. That puts different demands on hardware than training a model does.
TPU 8t targets pre-training at scale
Google says the TPU 8t can scale to superpods of up to 9,600 chips, delivering 121 ExaFlops using FP4. Per chip, it delivers 12.6 FP4 PFLOPs and pairs 216 GB of High Bandwidth Memory (HBM) with 128 MB of Static Random-Access Memory (SRAM).
TPU 8i is built for memory-heavy inference
The TPU 8i carries 288 GB of HBM and 384 MB of SRAM, which Google says is three times the SRAM of the previous generation. Its compute comes in at 10.1 FP4 PFLOPs per chip.
Google's headline efficiency claim for the TPU 8i is up to 80% better performance-per-dollar than the Ironwood generation. The chip also uses a new Boardfly topology, which governs how chips are wired to talk to one another, along with a CAE engine that Google says boosts performance.
Shared manufacturing and a later rollout
Both chips are manufactured on TSMC's N3 process technology. They also integrate Google's own Arm-based Axion CPUs and fit into the company's AI Hypercomputer architecture.
The eighth generation follows Ironwood, the seventh-generation TPU that launched in 2025. Google expects general availability later in 2026. The company plans to share more technical details at Hot Chips 2026 in August.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
