Google's eighth-generation TPUs split training and inference into two chips
Source
x.com
Author
Google AI
Date
Context
Google AI describes its eighth-generation AI chips as two designs aimed at different stages of a model's life. TPU 8t focuses on training, when a model learns from data; TPU 8i focuses on serving the trained model as it answers requests and takes steps in an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → workflow. The April 29 article emphasizes feeding data to training chips and recovering from hardware failures for 8t, while emphasizing memory and response latency for 8i. The split lets the designs target different sources of delay in building a model and putting it to work.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
postGoogle opens SynthID watermark checks to everyone, with partner coverage
postGemma 4 orchestrates a plant DNA model to find crop mutations in minutes
articleGoogle's Project CC gives a family agent its own identity and sandbox
