All posts
AI / News

NVIDIA Says Its Vera CPU Delivers 1.6x More AI Agents per Watt Than the Fastest x86 Chip

NVIDIA: Vera CPU rack for agentic AI

NVIDIA published its case for the Vera CPU on 5 November 2026, positioning it as the processor for AI agents. The argument is that agent workloads alternate between GPU reasoning and CPU execution, so the CPU side has become a bottleneck worth designing for.

The headline comparison is for a 100-megawatt data centre. NVIDIA claims 1.6 times more agents per watt than the fastest x86 CPU. Against the most energy-efficient x86 CPU, it claims twice the agent speed while keeping leading energy efficiency. These are NVIDIA's own figures, so treat them as a claim until independent tests appear.

The partner results are more specific. Perplexity reports agentic coding tasks running 1.5 times faster than on traditional CPUs and plans to use Vera in an upcoming production system. DeepInfra reports up to 2.2 times the speed of the best x86 processor on agent orchestration. Modal reports up to twice the sandbox throughput per core.

Two results stand out. Daytona brought 1,000 agent sandboxes online in 10.8 seconds, against 88.4 seconds on x86. ClickHouse scaled close to linearly across 176 cores with a standard binary and topped the ClickBench hardware leaderboard.

NVIDIA does not give a general availability date in the post. The only deployment timing mentioned is Perplexity's planned production use, so this is an announcement of direction rather than a product you can buy this month.

For a studio, the interest is in what it signals. Much of the AI tooling we use, from image generation to coding assistants, now runs as agents that call tools, run code and wait. The more that infrastructure improves, the cheaper and faster those workflows become, though not usually on a timescale that changes our week.

Our reading is that the real shift is the framing. For two years the story was about GPUs. NVIDIA is now telling the market that the rest of the machine matters for agents too, and that is a useful thing to keep in mind when reading any cost or speed claim about agentic tools.