Google unveils TPU 8t and 8i for AI supercomputing
Google has announced its eighth-generation Tensor Processing Units (TPUs), the TPU 8t and TPU 8i. The custom-engineered chips are designed for AI supercomputing, with the TPU 8t focused on training and the TPU 8i on inference. Google positions the new hardware as built for efficiency and scale to support the next generation of AI, particularly for the 'agentic era'.
Key Takeaways
- TPU 8t is Google’s new training chip for AI supercomputing.
- TPU 8i is the companion inference chip in the eighth-generation lineup.
- Google describes both chips as custom-engineered for efficiency and scale.
- The launch follows a decade of TPU development, according to Google.
Why It Matters
Google is splitting its eighth-generation TPU line between training and inference, which gives it a clearer hardware story for AI workloads at different stages of the stack. The company is framing TPU 8t and TPU 8i around efficiency and scale, with “the agentic era” as the stated target. For streaming teams building or buying AI infrastructure, the relevant signal is how Google positions these chips in Google Cloud and whether the company provides more detail on performance and deployment. Watch for any technical specs or customer availability tied to TPU 8t and TPU 8i.
Read full article at blog.google