NVIDIA Vera CPU

NVIDIA Vera CPU is NVIDIA’s CPU line aimed at AI factories and agent workloads. Official blog (2026-11-05): https://blogs.nvidia.com/blog/vera-cpu-highest-throughput-fastest-agents/ .

AI agents do more than generate answers—they search, query databases, write and test code, and take action, moving between GPU reasoning and CPU execution. As more agents run, AI factories need CPUs that make each agent fast and support many agents within power limits.

Highlights:

• Positioned for highest throughput and fastest agents in one CPU, so providers need not mix “fast” and “dense” x86 parts;
• In NVIDIA’s ~100 MW data-center comparison: vs the fastest x86, Vera enables about 1.6× more agents per watt while keeping leading agent speed; vs the most energy-efficient x86, about 2× faster agent speed with leading efficiency;
• Early adopters (per the blog): Perplexity ~1.5× on agentic coding; DeepInfra up to ~2.2× on agentic orchestration; ClickHouse standard binary scaled near-linearly across 176 cores and topped ClickBench hardware; DuckLabs ~1.5× median TPC-H SF1000 vs Intel Xeon 6; Daytona brought 1,000 agent sandboxes online in 10.8 s (vs ~88.4 s on x86) and 12.7 completed jobs/s at 2,000 concurrent sandboxes; Modal up to ~2× sandbox throughput per core.

Problem

Agent workflows lean heavily on the CPU (search, DB, sandboxes, orchestration) while AI factories are power-capped. Vera aims to cover both speed and density so operators need fewer CPU SKUs.

CPU vs GPU roles

• GPU: model inference / generation;
• CPU: tool calls, queries, code execution, sandbox lifecycle, orchestration;
• Agents bounce between both; the CPU often limits throughput and latency.

Limits

• Figures come from NVIDIA’s blog and partner production-style tests; baselines and configs follow each disclosure;
• “Fastest / highest-throughput” is product positioning—gains depend on workload, software stack, and system balance;
• Aimed at AI factories and cloud/inference providers, not consumer desktops.

Who it's for

• Cloud and inference providers building agent / AI factories;
• Platforms that need high-concurrency sandboxes, DB queries, and agent orchestration (e.g. coding agents, analytics DBs cited by NVIDIA).

Getting started

1. Read https://blogs.nvidia.com/blog/vera-cpu-highest-throughput-fastest-agents/ ;
2. Map your agent load (search, DB, code write/test, sandboxes) to see if the CPU is the bottleneck;
3. Optimize for agents-per-watt and per-task latency under a power wall, not peak core count alone;
4. Get Vera platform specs and deployment guidance from NVIDIA / partners (official channels).