Google Adds TPU 8tx and TPU 8i Max to Chip Arsenal in Push Past Nvidia

By Nathan Brooks

Google unveiled two new tensor processing units on Monday, TPU 8tx and TPU 8i Max, expanding its eighth-generation chip family to cover confidential AI workloads and larger memory demands. The move deepens Google's effort to cut its dependence on third-party silicon — read Nvidia — for its cloud computing business.

Both chips build on the eighth-generation TPUs Google introduced in April, which brought more inference capacity and support for large training requirements. These two variants push the portfolio into narrower, higher-stakes use cases.

What does TPU 8tx change for cloud buyers?

TPU 8tx, designed with AMD, is based on the TPU 8t chip and is built for confidentially training and running AI models. In a blog post, Google said the chip "builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use."

For practitioners, that's a meaningful shift. Confidential computing has mostly been a software-layer exercise; moving protection into silicon changes what regulated industries can safely run on shared cloud infrastructure.

The chip also carries a custom inferencing network that lets it run reasoning alongside training. Google says that means customers can distill and train small models on the fly — useful, according to the company, when a business wants to train a model on proprietary data without moving that data between devices. That's a direct answer to the data-sovereignty question enterprise buyers keep asking.

What does TPU 8i Max actually deliver?

TPU 8i Max is simpler to grasp: it combines two of the inference-focused TPU 8i chips launched in April into a single package. That doubles the memory capacity, processing power and bandwidth.

In practical terms, a doubled memory envelope means AI models can handle a larger range of information during queries — think longer context windows and heavier retrieval workloads without splitting the job across multiple chips.

Why the timing matters

The launch lands amid a broader silicon race. Hyperscalers Microsoft and Amazon, and AI labs OpenAI and Anthropic, are all developing their own AI-specific semiconductors and accompanying software to reduce their reliance on Nvidia's dominance.

Google's two-chip salvo shows how that strategy is maturing: the portfolio is no longer one general-purpose accelerator but a lineup split by workload — training, inference, and now confidential compute.

The news also arrives at a vulnerable moment for Nvidia. Over the weekend, CEO Jensen Huang admitted to having an affair with someone alleged to be a Chinese spy. Rivals rarely get a cleaner opening to press their silicon alternatives to cloud customers weighing vendor risk.

For cloud buyers, the so-what is straightforward: more TPU variants mean more pricing pressure on accelerated computing, and confidential silicon opens doors for workloads that previously stayed on-premise.