- Google added two new systems to its eighth-generation TPU lineup, TPU 8tx and TPU 8i Max, bringing to four the number of TPU variants it has shipped within a single year — a first for the company.
- TPU 8tx adds hardware-enforced confidential computing, keeping data and model weights encrypted even while a processor is actively using them, and pairs with fifth-generation AMD EPYC host CPUs.
- TPU 8i Max packages two TPU 8i chips together, doubling compute, SRAM, and high-bandwidth memory within the same rack footprint as a single TPU 8i.
- The launch, announced on Google’s blog and reported by The Decoder on November 9, 2026, extends Google’s push to sell TPU capacity to outside customers and intensifies its infrastructure rivalry with Nvidia.
What Happened
Google added two new systems to its eighth-generation AI chip lineup on November 9, 2026: TPU 8tx, built for confidential computing, and TPU 8i Max, built for memory-intensive workloads. The additions bring to four the number of distinct TPU 8-series variants Google has shipped within a single year, which the company said is the first time it has done so, according to Google’s announcement and reporting from The Decoder (Nov. 9, 2026).
Why It Matters
Google has increasingly framed its in-house TPUs as a platform for external customers rather than purely internal infrastructure, pairing the new hardware with software tools like TorchTPU to lower the barrier for outside developers. The move follows an October report that OpenAI was in talks to buy Google TPU capacity to reduce its reliance on Nvidia chips, and comes as Nvidia’s own next-generation Kyber NVL144 rack system has reportedly slipped to 2028.
Technical Details
TPU 8tx uses the same silicon as the existing TPU 8t but adds hardware-level encryption so data and model weights stay protected even while a processor is actively working with them in memory, not just in storage or transit — a capability Nvidia already offers for its GPUs through its own confidential-computing architecture. Google says TPU 8tx delivers up to twice the latency reduction of TPU 8t with a near-zero performance penalty. TPU 8i Max combines two TPU 8i chips into one package, doubling compute, SRAM, and HBM bandwidth within the same rack space — a packaging and networking change rather than a new chip generation.
Who’s Affected
TPU 8tx targets enterprises and government agencies handling regulated or sensitive data that previously had to decrypt it in memory to run AI workloads. TPU 8i Max is aimed at developers running memory-bound, agentic workloads, such as agents working across large codebases or hours of video, where memory bandwidth rather than raw compute is the bottleneck. Cloud customers choosing between Google Cloud TPUs and Nvidia GPU instances now have two additional configurations to evaluate.
What’s Next
Google said its modular approach — new packaging, networking, and memory configurations built on existing processors — lets it ship new hardware variants in months rather than years, suggesting more TPU 8-series configurations could arrive before a next-generation chip. Pricing and availability details for TPU 8tx and TPU 8i Max had not been published as of November 9, 2026.
