ChipsThe story, in brief
Two new TPUs to power the next wave of AI training and inference at Google
Google just split its TPU strategy. 8t for training. 8i for inference. Here's what that means for your compute roadmap.
The KeyNews take
Why it matters
Google's eighth-generation TPU split signals a strategic shift toward specialized silicon for distinct workloads—training vs. inference—potentially reshaping how enterprises architect their AI infrastructure and where they'll place compute bets.
The key facts
5 to knowTwo new TPU architectures announced: TPU 8t (training) and TPU 8i (inference)
Eighth-generation TPU line released
Announced at Google Cloud Next 2026 on April 22, 2026
Custom silicon designed to handle next-generation AI workloads
Specialized architecture split across training and inference use cases
Go to the source
SiliconAnglesiliconangle.com
Publisher excerpt: Google LLC introduced two new custom silicon chips for artificial intelligence today at Google Cloud Next 2026, unveiling two distinct Tensor Processor Unit architectures built for training and inference: the eighth-generation TPU 8t and TPU 8i. The company said it designed the pair of chips to…