Google Unveils Two Specialized TPU Chips for Agentic AI Workloads

Google on Thursday unveiled two new specialized tensor processing units designed to handle the computing demands of agentic AI, marking the company’s largest infrastructure bet on autonomous AI systems to date.

The eighth-generation TPU lineup, announced at Google Cloud Next, includes chips optimized for the distinct workloads that agentic AI systems generate — from sustained multi-step reasoning to high-throughput tool execution. The company said the new hardware reflects a fundamental shift in how AI infrastructure must be designed as models move beyond simple query-response patterns.

Agentic AI systems, which can independently plan, execute tasks and interact with external tools, place different demands on computing hardware than traditional large language model inference. These workloads require sustained processing over longer periods and the ability to handle multiple parallel operations as agents coordinate complex tasks.

Google has been among the most aggressive investors in custom AI silicon, having developed its TPU line since 2016 as an alternative to commercial GPUs for training and running AI models. The company’s cloud division offers TPU access to enterprise customers, positioning the chips as a competitive advantage against rivals Amazon Web Services and Microsoft Azure.

The announcement comes as the major cloud providers race to build infrastructure capable of supporting increasingly autonomous AI agents. Microsoft, Amazon and Oracle have each announced multibillion-dollar data center investments in recent months, while chipmaker Nvidia continues to dominate the broader AI accelerator market.

Google did not immediately disclose detailed performance benchmarks or pricing for the new TPUs. The chips are expected to be available through Google Cloud, though the company has not confirmed a specific timeline for general availability.

The TPU announcement is part of a broader set of infrastructure and AI product updates Google is rolling out at Cloud Next, as the company seeks to demonstrate its technical edge in the rapidly evolving AI market.