Verda is deploying the Arm AGI CPU across our next-generation AI infrastructure, paired with NVIDIA GB300 systems today and the upcoming NVIDIA Vera Rubin (VR200) fleet. The result is a fully Arm-native stack from orchestration to inference, purpose-built for agentic AI at scale.
Why the CPU became the critical layer in agentic AI
As AI shifts from isolated inference tasks to agentic workflows, the role of the CPU is changing. Agentic AI broke the old CPU-GPU contract. In a traditional inference workflow, the CPU tokenizes a request and hands it to the GPU. In an agentic workflow, the CPU orchestrates hundreds of parallel sub-agents.
That changes what infrastructure has to deliver. Agents run thousands of coordinated interactions across models, tools, memory, and external services, and long-tail latency anywhere in the chain breaks the SLA everywhere. The CPU is where that coordination lives, which makes it the control layer that determines whether an agentic system performs at scale. As agents scale, the CPU-to-GPU ratio has to scale with them.
The Arm AGI CPU is built for this role, with high core scalability, strong memory bandwidth, and system-level efficiency designed for tight integration with accelerators. Pairing it with our GB300 and upcoming VR200 systems gives ML teams on Verda a platform optimized for the orchestration demands of agentic AI.
What it means for ML teams building on Verda
Tighter CPU-GPU coupling. A consistent Arm architecture across the rack. The Grace CPU inside GB300 is Arm-based, so pairing it with the Arm AGI CPU keeps the entire stack on a single instruction set, simplifying everything from container builds to performance tuning.
Better rack density provides better economics. The Arm AGI CPU’s efficiency profile allows us to increase the compute density in the rack. We pass this through directly, meaning more performance per dollar for teams running at scale.
A consistent, standards-based platform. From GB300 today to Vera Rubin when it ships, this architecture gives you a predictable foundation.
The same stack we trust for our own work
This is the infrastructure powering Verda AI Lab and the automated orchestration systems we run the platform on every day. The same stack our customers rely on is the one we use ourselves.
This deployment is part of how we are building Verda: a renewable-powered AI cloud designed for ML teams, with the kind of infrastructure agentic AI actually needs underneath it.
Get early access
We are opening capacity on the Arm AGI CPU + GB300 configuration soon. If you want to get on the infrastructure ahead of the curve, contact us.
More technical details and availability timelines coming soon.