Google has launched two new artificial intelligence processors, TPU 8tx and TPU 8i Max, as it expands its chip business to serve companies running highly demanding AI applications.

The two systems, which are part of Google’s eighth-generation Tensor Processing Unit (TPU) range, include the 8tx, which targets organisations handling sensitive information and the 8i Max, designed for AI applications that require large amounts of memory and computing power.

Google announced the products on Monday, November 9, 2026, bringing the number of eighth-generation TPU systems unveiled this year to four. The company previously introduced the TPU 8t and 8i in April.

With the innovation, Google is offering processors for different stages of AI development, from training large models to running real-time applications and processing complex tasks.

The TPU 8tx targets government agencies and businesses in regulated industries that need to process confidential information. Google has combined its existing TPU 8t silicon with fifth-generation AMD EPYC processors to provide hardware-level data protection.

Ordinarily, computers must decrypt information in memory before processing it, leaving sensitive data exposed during that stage. The 8tx protects data and AI model weights while they remain in use.

Google says the system also uses its Boardfly network technology to reduce latency. It can deliver up to twice the speed of the previous version, with almost no performance penalty.

The configuration also allows organisations to train and run AI models on the same system. Businesses can use private data to fine-tune models, develop smaller versions of larger models and train AI agents without moving sensitive information between separate computing clusters.

The second product, TPU 8i Max, addresses the memory demands of applications that process large amounts of information simultaneously.

Google built the system by combining two TPU 8i chips into a single package. This gives it twice the computing power, on-chip SRAM and high-bandwidth memory (HBM) bandwidth of the standard 8i.

The additional capacity is intended for applications such as coding assistants, mathematical reasoning and drug discovery. It also allows data centres to accommodate more computing power within the same rack footprint.

Google says the 8i Max delivers higher computing density without requiring additional rack space. This could help businesses run larger AI workloads within existing data centre facilities.

The launch also aligns with Google’s initiative to attract more businesses to its TPU infrastructure. The company says nearly 100 customers now use its processors, including AI research companies, financial trading firms, retailers and universities.

Although Google initially developed TPUs for its own services, including Search, YouTube and Gemini, it now makes them available through Google Cloud.

The cloud business has already started generating revenue from the chips. Alphabet reported that Google Cloud revenue more than doubled to $30.71 billion in the September quarter, with TPU sales making a meaningful contribution.

Google is also working to make its processors easier for developers to use. In October, it open-sourced TorchTPU, allowing developers to run PyTorch code and Hugging Face models on its hardware without rewriting their code.

Other additions include an open-source TPU agent suite, now in preview, which helps developers configure systems and identify performance problems. The company has also introduced the XLA Serving Engine for running trained models across TPUs, graphics processing units (GPUs) and central processing units (CPUs).

Meanwhile, its Agent Compute Engine can suspend inactive AI agent sessions and move their working memory off the chip. This is intended to reduce computing costs when agents pause while waiting for external tools, databases or human input.

Google has also made Cloud Pathways generally available on Google Kubernetes Engine. The software allows customers to coordinate large TPU workloads across multiple computing clusters through a single controller.

Its eighth-generation TPU range now comprises four systems: the 8t for large-scale model training, the 8i for real-time reasoning, the 8tx for confidential computing and the 8i Max for memory-intensive workloads.