NVIDIA Vera CPU Accelerates AI Agents and Raises Throughput per Watt

NVIDIA has positioned its Vera CPU as a platform for AI factories that must support more AI agents without sacrificing response speed. AI agents increasingly move between GPU reasoning and CPU-based execution, including database queries, code testing, tool invocation and orchestration. As agent workloads scale, infrastructure operators face a practical constraint: power budgets limit how many agents can run concurrently and how quickly they can be served. NVIDIA said Vera addresses this challenge by combining high agent throughput with strong energy efficiency in a single CPU architecture. In a 100-megawatt data center comparison, NVIDIA said a Vera CPU outperformed both the fastest x86 CPU and the most energy-efficient x86 CPU on agent workloads. The company reported 1.6x more agents per watt than the fastest x86 CPU and 2x faster agent speed with leading energy efficiency compared with the most energy-efficient x86 CPU. NVIDIA presented the result as a way to reduce the need to mix different CPU classes to balance speed and capacity. Early adopters reported production-style gains across agent-related workloads. Perplexity said Vera ran agentic coding tasks 1.5x faster than traditional CPUs and plans to deploy the processor in an upcoming production system. DeepInfra reported up to 2.2x speed over the best x86 processor in agentic orchestration for its production workload. ClickHouse said its standard binary, without Vera-specific tuning, scaled near-linearly across 176 cores and topped the ClickBench hardware leaderboard. Further tests showed similar patterns in sandboxing, analytics and parallel agent execution. DuckLabs achieved about 1.5x the median TPC-H SF1000 performance of an Intel Xeon 6 baseline. Daytona brought 1,000 agent sandboxes online in 10.8 seconds, compared with 88.4 seconds on x86, and completed 12.7 jobs per second at 2,000 concurrent sandboxes. Modal reported up to 2x sandbox throughput per core versus comparable x86 hardware, with gains increasing under heavy load. The broader implication is that AI infrastructure providers may increasingly evaluate CPUs not only by peak core count or clock speed, but by how much useful agent work they can complete within fixed power envelopes. NVIDIA is presenting Vera as a single CPU architecture capable of serving a range of agent workloads, aiming to help AI factories scale more efficiently as agentic systems become more common.