Accelerating the agentic era: How Arm and ​​Google​ Cloud co-design​ infrastructure for TPU 8i Max

Google Cloud is pairing​​ ​one of ​its newest TPU system​s​ ​with​ ​Google Axion processors that are custom Arm Neoverse®-based CPUs​. With TPU 8i Max, Google is extending its use of Axion CPUs as the host compute for its AI accelerators, building on Axion’s role in TPU 8t and TPU 8i. 

​​Google Axion provides the CPU foundation to support Google’s AI infrastructure across large-scale training, real-time reasoning, and increasingly demanding agentic workloads. Google’s expansion to TPU 8i Max marks another significant milestone for Arm at the heart of the systems powering next-generation AI infrastructure. 

Advancing performance across the AI system

AI performance depends on more than the accelerator. CPUs prepare data, orchestrate workloads and execute the infrastructure software that keeps the entire system productive. As agents reason, use tools and coordinate complex tasks, the demands on that CPU foundation continue to grow. 

With Axion, Google brings together Neoverse performance and efficiency with custom silicon designed for its infrastructure. Extending that approach to TPU 8i Max reinforces the importance of the CPU and accelerator co-design in scaling AI. 

“Delivering AI performance at scale requires​ co-designing​ CPUs and accelerators ​to work​ seamlessly​ together,” said Mark Lohmeyer, VP & GM, AI and Computing Infrastructure​​, Google Cloud. “​Our ​Arm-based Google Axion ​processors deliver ​​​the performance, efficiency and flexibility​ required​ to ​​power​ the demanding host workloads behind our TPU systems. With TPU 8i Max, we’re extending that ​co-designed ​foundation to​ help customers tackle their most complex AI applications.” 

Unlocking the age of inference

Google Cloud’s continued investment reflects a broader industry shift toward pairing CPUs with accelerators to unlock the next era of AI inference.​ Beyond serving as accelerator host compute, Google Axion also powers general-purpose cloud instances that run agent runtimes and orchestration frameworks.​ 

Earlier this year, Google Cloud announced Google Axion N4A, which delivers up to 2x better price-performance than comparable current-generation x86-based virtual machines. With GKE Agent Sandbox running on Axion N4A, customers can also securely isolate agentic AI workloads while achieving 30% better price-performance.​ Building on this momentum, Agent Substrate on GKE also natively supports Google Axion processors, providing an open-source, secure-by-default agent execution runtime engineered to run millions of sandboxes with 10x higher density than standard container runtimes.​ 

From powering Google’s latest TPU systems to enabling millions of AI agents to operate at scale, Arm is helping build the compute foundation for the next era of AI.