Nvidia's vera CPU demonstrates significant efficiency gains for AI agents in data centers

Nvidia released a blog post on November 5th detailing performance metrics for its Vera CPU, which was launched on March 16, 2026. The data showcases the processor's ability to deliver industry-leading agent execution speed alongside high per-watt throughput, enabling data centers to run more agents under strict power constraints. Nvidia noted that agents require more than just generating answers; they involve tasks like code testing, database querying, and searching, which heavily rely on CPU cycles and frequent CPU-GPU switching. In a 100-megawatt data center test, Vera CPU supported 1.6 times the number of agents per watt compared to the fastest x86 CPUs, and achieved a 2x speedup over the most energy-efficient x86 CPUs while maintaining top efficiency. Several industry participants validated these gains. Perplexity reported a 1.5x speed increase for agent coding tasks using Vera and plans to integrate it into production. DeepInfra observed up to a 2.2x speed advantage in agent orchestration tasks over top x86 processors. Furthermore, Daytona demonstrated Vera launching 1,000 agent sandboxes in 10.8 seconds versus 88.4 seconds on x86, and Modal reported a 2x per-core sandbox throughput compared to similar x86 hardware.