Nvidia Vera CPU Doubles Throughput in Modal's Tests

Modal has deployed Nvidia's Vera CPUs alongside its existing x86 fleet, and reports that internal testing showed roughly twice the throughput of comparable Intel and AMD chips in AI agent workloads.

Why it matters: AI agent workloads are increasingly CPU-bound in areas like orchestration, tool calls and data handling, not just GPU-heavy inference. If a CPU can meaningfully raise throughput per deployment, that changes how infrastructure teams think about capacity planning and cost.

The caveat matters too. These are Modal's internal results, not independent benchmarks, and mixed x86 and Vera fleets bring real operational questions around scheduling, software compatibility and tuning.

For engineers and architects, the takeaway is to track how workloads actually behave on new silicon, not just headline numbers. For leaders, it is a signal that CPU choices are back in the AI infrastructure conversation.

What workloads in your organization are most constrained by CPU throughput today?