Google Expands TPU Portfolio with 8tx and 8i Max for the AI Agent Era

Google has expanded its eighth-generation TPU portfolio with two new systems: the TPU 8tx, aimed at secure and confidential inference, and the TPU 8i Max, designed for massive context windows and maximum compute density per rack. It is the first time the company has launched four TPU systems in a single year.

The new accelerators respond to a structural shift in compute demand. With workloads migrating from human scale to AI agent scale, companies now require hardware capable of reasoning over extensive contexts while simultaneously offering security guarantees for sensitive data.

From human scale to agent scale

The TPU 8i Max was conceived for what Google calls "massive context windows" — a model's ability to process and reason over gigantic volumes of information in a single inference. This is a central requirement for autonomous agents, which need to maintain memory of long interactions and query extensive databases without losing coherence.

The TPU 8tx, meanwhile, addresses a specific demand from highly regulated clients that need specialized encryption capabilities for unique workloads. The system offers secure and confidential inference, a differentiator for sectors such as finance and healthcare, where data protection is mandatory.

Four systems in one year

In April, during Google Cloud Next, the company had already introduced the TPU 8t for large-scale training and the TPU 8i for real-time reasoning. At the time, the disclosed numbers were impressive: up to 2.7x better training performance for the 8t and 80% better inference cost-benefit for the 8i, both compared to the previous generation.

Now, with the addition of the 8tx and 8i Max, Google closes out the eighth generation with four specialized systems. The company emphasizes that the difference between a new chip and a new system is crucial: while a chip requires years of design and manufacturing of billions of transistors, a system involves integrating compute, networking, and software into a cohesive configuration.

Expanding customer base

TPUs are no longer an internal-use technology. Born more than a decade ago to meet Google's own demands — such as speech recognition and translation — tensor processing units now underpin a fast-growing business with nearly 100 customers, including frontier labs, trading firms, retailers, and universities.

Part of this growth comes from the ability for customers to deploy TPUs directly in their own data centers. Google has also been expanding access through software updates, such as recent announcements related to PyTorch and the XLA Serving Engine, which aim to make the architecture more accessible to developers.

The strategy combines modular custom silicon with an open, easy-to-use software stack, allowing the company to release new hardware and system configurations at the pace of the AI industry. With demand for agents and massive contexts on the rise, the eighth generation of TPUs signals that the race for inference infrastructure is only beginning.