Google Unveils TPU 8tx and 8i Max to Take On Nvidia

Google has unveiled two new artificial intelligence chip systems, the TPU 8tx and the TPU 8i Max, taking its eighth-generation Tensor Processing Unit lineup to four systems in a single year as its challenge to Nvidia gathers pace.

The announcement was made on Monday by Amin Vahdat, Google’s senior vice-president and chief technologist for AI and infrastructure. The two new systems target two of the fastest-growing demands in AI: keeping sensitive data locked down while it is being processed, and giving autonomous agents enough memory to work across huge amounts of context at once.

Two systems, two different jobs

The TPU 8tx is built for secure and confidential inference, and also suits small-scale training. It targets regulated enterprises and governments running frontier or proprietary models alongside their most sensitive data.

Normally, data is encrypted when it is stored or sent over a network, but it has to be unlocked in memory while a computer processes it. The 8tx is built on the same silicon as the TPU 8t training chip and paired with fifth-generation AMD EPYC processor hosts, with security built into the hardware layer so data and model weights stay protected and isolated even while in active use.

Google claims the system delivers up to twice the lower latency of the TPU 8t, with a near-zero performance penalty. Because training-grade compute sits on the same system as a fast inference network, teams can fine-tune models on private data and train agents through reinforcement learning without moving sensitive data between separate clusters.

The TPU 8i Max tackles a different problem: the memory wall. When dozens of AI agents work across an entire codebase or hours of video at once, memory bandwidth becomes the bottleneck. Google’s answer is to put two low-latency TPU 8i chips onto a single package, doubling the compute, the on-chip SRAM and the high-bandwidth memory of the 8i, while fitting twice the silicon into the existing rack footprint.

Four systems in a single year

The eighth generation began at Google Cloud Next in April, when the company introduced the TPU 8t for large-scale training and the TPU 8i for real-time reasoning. Google says the pair offer up to 2.7 times better training performance and 80 per cent better inference price-performance than the previous generation. The business now serves nearly 100 customers, from frontier labs to retailers, and customers can install TPUs in their own data centres.

“This is the first time we have introduced four TPU systems in a single year,” Vahdat wrote in Google’s announcement. The speed is possible, he said, because Google repackages the compute, networking and security around a chip in months, while designing new silicon takes years.

The launch is the latest step in Google’s wider push into AI, which has also seen the company press schools and developers to adopt its tools.

A direct challenge to Nvidia

The announcement lands squarely on Nvidia’s territory. Nvidia dominates the market for AI processors, and demand for its graphics processing units remains greater than available supply. Google’s pitch is a purpose-built accelerator for every stage of the AI lifecycle: the 8t for training, the 8i for reasoning, the 8tx for confidential computing and the 8i Max for massive context windows.

The commercial stakes are large. Bloomberg Intelligence estimates that sales of Google’s TPUs, including related software and hardware, could exceed $135 billion by 2028, compared with an estimated $12 billion this year.

What Google did not say

Google gave no prices, release dates or regions for either new system, and secondary coverage suggests the specialised variants will become available next year. The performance figures are Google’s own; no independent benchmarks were published.

That leaves the most important question unanswered for now: whether customers outside Google’s existing base will buy TPUs in the volumes the company is projecting. The market will judge the hardware on price, availability and real-world performance — none of which Google has yet disclosed.

Frequently asked questions

What is a TPU?

A Tensor Processing Unit is a processor Google designs in-house for AI workloads, as an alternative to Nvidia’s graphics processing units. Google uses TPUs to train and run its Gemini models and rents them through Google Cloud.

When will the TPU 8tx and TPU 8i Max be available?

Google has not given a price, a general-availability date or a list of regions for either system. Secondary coverage suggests the specialised variants will become available next year.