Google expanded its eighth-generation TPU portfolio on 9 November 2026 with two new systems: TPU 8tx and TPU 8i Max. They are built for very different jobs, so which one wins depends on what you need.
First, these are new systems, not new chips. Both reuse silicon Google had already announced: TPU 8tx uses the same silicon as TPU 8t, while TPU 8i Max puts two TPU 8i chips in one package. Google says this is the first time it has introduced four TPU systems in a single year.
Where these two fit
At Google Cloud Next in April, Google introduced TPU 8t for large-scale training and TPU 8i for real-time reasoning. The new systems build on those in opposite directions: 8tx is optimised for secure and confidential inference, and also good for small-scale training; 8i Max is built for the "memory wall" — massive context windows and the highest compute density per rack.
TPU 8tx: built for confidential inference
TPU 8tx is built on TPU 8t silicon, paired with 5th Generation AMD EPYC host CPUs, and it uses the Boardfly network from TPU 8i. Google says it keeps sensitive data and model weights protected and isolated even while in active use — protection during use, not just when data is stored, is the hard part of confidential computing.
Google claims up to 2 times lower latency than TPU 8t, with a near-zero performance penalty. That is Google's own claim, not independently tested. The target users are banks, governments and regulated enterprises. Google gave no price, no general-availability date and no named customer; Amin Vahdat told Bloomberg the systems arrive next year, in 2027.
TPU 8i Max: built for massive context
TPU 8i Max puts two TPU 8i chips in a single package, doubling the compute, on-chip SRAM and HBM bandwidth of the 8i. HBM is high-bandwidth memory, the fast memory close to the chip; SRAM is even faster on-chip memory. Doubling both is aimed at workloads that need to hold a lot of information at once.
Google describes the target workloads as dozens of AI agents working across an entire codebase or hours of video at once. A context window is how much text, code or video a model can consider in one go, and for those jobs it has to be very large. Like 8tx, 8i Max has no price, no general-availability date, no named customer and no independent benchmark.
Head-to-head comparison
| Feature | TPU 8tx | TPU 8i Max |
|---|---|---|
| Purpose | Secure and confidential inference | Massive context windows and compute density |
| Silicon | TPU 8t silicon, with 5th Gen AMD EPYC host CPUs | Two TPU 8i chips in one package |
| Headline claim | Up to 2x lower latency than TPU 8t | 2x compute, SRAM and HBM bandwidth of 8i |
| Best for | Banks, governments, regulated enterprises | Long-context agent work |
| Also good for | Small-scale training | Hours of video, whole codebases |
| Price | Not announced | Not announced |
| Availability | Announced 9 Nov 2026, expected next year | Announced 9 Nov 2026, expected next year |
The split is clear. TPU 8tx is about trust; TPU 8i Max is about capacity.
Which one wins?
If you run AI on regulated or sensitive data — banking, government, or any work where isolation matters — TPU 8tx is the relevant system. If your bottleneck is context — agents reading a whole codebase or working through hours of video — TPU 8i Max is the relevant system.
Neither wins in general, because they are not competing with each other. They answer two different limits: security on one side, and memory on the other.
Key takeaways
- Google announced TPU 8tx and TPU 8i Max on 9 Nov 2026, its third and fourth TPU systems of the year.
- New systems, not new chips: 8tx reuses TPU 8t silicon; 8i Max puts two TPU 8i chips in one package.
- TPU 8tx is for secure, confidential inference, with up to 2x lower latency than TPU 8t claimed by Google.
- TPU 8i Max doubles the compute, SRAM and HBM bandwidth of 8i for massive context windows.
- Neither has a price, a general-availability date or a named customer; both are expected in 2027.
- All performance figures are Google's own and have not been independently tested.
Conclusion
TPU 8tx and TPU 8i Max show Google splitting its AI hardware in two directions at once: one to keep sensitive work safe while it runs, the other to give AI agents enough memory for very large jobs. Both are announcements for now, with no price or shipping date. The question is simple: is your limit trust, or context?
FAQ
Are TPU 8tx and TPU 8i Max new chips?
No. They are new systems that reuse silicon Google had already announced. TPU 8tx uses the same silicon as TPU 8t, and TPU 8i Max puts two TPU 8i chips in one package.
When will TPU 8tx and TPU 8i Max be available?
Google has given no date or price for either system. Amin Vahdat told Bloomberg they arrive next year, in 2027.
Which TPU should a bank or government agency choose?
TPU 8tx is the one Google aimed at banks, governments and regulated enterprises, for secure and confidential inference.
Sources
- https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/?ref=explainx
- https://madrobot.blog/2026/11/09/google-tpu-8tx-8i-max-four-tpus-nvidia/
- https://www.rackline.news/story/1gcy9ig?from=front
Related reading on Daily Tech & AI:
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for the latest tech and AI updates: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r
Comments
Post a Comment