Nvidia released performance data for its upcoming Vera CPU, scheduled for launch on March 16, 2026. The company claims the processor optimizes the balance between agent execution speed and power efficiency, addressing bottlenecks in data centers that deploy AI agents.
In a simulated 100-megawatt data center environment, the Vera CPU outperformed top-tier x86 processors. It supports 1.6 times more agents per watt than the fastest x86 chips and delivers double the speed of the most energy-efficient x86 alternatives while maintaining superior efficiency.
Early testing by industry partners confirms these gains. Perplexity reported a 1.5 times speed increase for coding tasks, while DeepInfra observed up to 2.2 times faster performance in agent orchestration. ClickHouse achieved near-linear scaling on 176 cores without specific software optimizations.
Concurrency benchmarks highlight significant improvements in sandbox environments. Daytona used Vera to launch 1,000 agent sandboxes in 10.8 seconds, compared to 88.4 seconds on x86 platforms. Additionally, Modal reported that Vera achieves double the sandbox throughput per core compared to equivalent x86 hardware.
