Source : TechCrunch
MOUNTAIN VIEW, California, November 9, 2026 – Google has expanded its eighth-generation tensor processing unit (TPU) family with two new systems, TPU 8tx and TPU 8i Max, in a move it says will support more secure and longer-context artificial intelligence workloads. The announcement, posted on Monday on Google’s official blog, was also reported by TechCrunch and The Verge.
The new systems join TPU 8t and TPU 8i, which Google introduced at Google Cloud Next in April. The company said this is the first time it has launched four TPU systems in a single year. TPU 8t is aimed at large-scale training and, Google said, offers up to 2.7 times better training performance than the previous generation. TPU 8i is designed for real-time reasoning and is claimed to deliver 80 per cent better inference price-performance than the previous generation.
TPU 8tx, the third system in the eighth-generation lineup, is built for secure and confidential inference, and Google said it is also well suited to small-scale training. The company said it is intended for regulated enterprises and governments that want to run frontier or proprietary models, and autonomous agents, alongside sensitive data. The system uses the same silicon as TPU 8t and is paired with 5th Generation AMD EPYC processor hosts. Google said it “builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use”. By using the company’s low-latency Boardfly network, TPU 8tx is said to deliver up to twice as low latency as TPU 8t, with what the company called a near-zero performance penalty. Google said organisations will be able to fine-tune models on private data, distill larger models into smaller ones, and train agents through reinforcement learning without moving sensitive data between separate clusters.
TPU 8i Max, the fourth system, is designed to ease what Google calls the “memory wall”. The company said it combined two TPU 8i chips in a single package, doubling compute, on-chip SRAM and high-bandwidth memory compared with TPU 8i. The aim is to support very large context windows, including what Google described as million-token scale, such as when many AI agents work across a large software codebase or long stretches of video, while fitting more computing power into the same rack space.
Amin Vahdat, Google’s senior vice-president and chief technologist for AI and infrastructure, said the company first designed TPUs more than a decade ago for internal uses such as voice recognition and language translation. Google now says TPUs are used to train and serve its own Gemini models, run agentic workloads, serve Google Cloud customers, and, more recently, allow some customers to deploy TPUs in their own data centres. The company said it now has nearly 100 TPU customers, including frontier labs, trading firms, retailers and universities.
Google also said it is trying to make TPUs easier to use, including by supporting PyTorch and open-source software. TechCrunch reported that the new systems broaden Google’s in-house chip portfolio as large technology companies seek to reduce their reliance on processors designed by others.
About The Author
