Google Unveils Two New Systems, TPU 8tx and TPU 8i Max
Posted As of
Google announced TPU 8tx, focused on security, and TPU 8i Max, built for large-context workloads. Reports differ on pricing and launch timing, so neither can be confirmed.
What happened
Based on Google's blog and the Google Cloud blog (official) announcements.
- Google said these two systems bring its new TPU lineup to four in one year. At Cloud Next in April, the training-focused 8t and inference-focused 8i were unveiled.
- TPU 8tx combines the same silicon as TPU 8t with a custom AMD CPU host, and reportedly adds hardware security features that protect data and model weights during use. Its intended uses are private fine-tuning and running secure agents.
- TPU 8i Max is described as a product targeting large context windows and the highest compute density per rack. According to a secondary report, it packages two 8i chips into one package, doubling compute, SRAM and HBM bandwidth in the same rack footprint, but this figure is not something directly confirmed in the blog text.
- On the software side, Agent Compute Engine was introduced, which pauses idle agent sessions and moves working memory off the chip, and Cloud Pathways reached general availability (GA) on GKE.
- Reports on launch timing conflict: one citing Bloomberg says next year, while others say no pricing or schedule has been given.
Our take
This is not a product that Mac, PC, or glasses users running local AI can use directly. However, it signals that cloud providers are increasingly building hardware aimed at long-context and private fine-tuning demand. Absolute performance figures, pricing, and release dates have not been confirmed yet.
Sources and related posts
- Google Blog Primary source
- Google Cloud Blog Primary source