Vera Combines Peak Throughput with the Fastest Agents.
Vera CPU targets AI agent workloads, combining faster execution of lookups, code, and actions with more concurrent agents per watt than x86 CPUs.
Key Takeaways
- NVIDIA presents Vera as a CPU optimized for AI agent workloads, where CPU execution steps such as lookups, database queries, code testing, and follow-on actions matter as much as GPU reasoning.
- Vera is positioned as a single CPU that combines fast agent speed with high capacity per watt, avoiding the need to pair separate x86 parts optimized for speed or density.
- NVIDIA claims Vera delivers 1.6x more agents per watt than the fastest x86 CPU and 2x faster agent speed than the most energy-efficient x86 CPU in power-limited data center scenarios.
- Early partner tests show gains in agentic coding, orchestration, analytics, and sandbox creation, suggesting Vera may help data centers fit more agents into fixed power budgets.
For readers evaluating AI infrastructure, NVIDIA says its Vera CPU is aimed at AI agents that perform lookups, database queries, code writing and testing, and follow-on actions, with the goal of moving more work through power-limited AI factories. In a 100-megawatt data center comparison, NVIDIA found a single Vera CPU performs better than the fastest x86 CPU and the most energy-efficient x86 CPU, while early adopter tests covered agentic coding, orchestration, analytics, benchmarks, and sandboxes.

A CPU built for many agent steps
AI agents are not limited to producing answers. A single task can alternate between GPU reasoning and CPU execution. That makes the CPU important because it handles the execution steps that turn a model output into action: checking information, querying data, running code, and carrying out operations.
As businesses deploy more agents, the infrastructure behind them has to do two things at once. It must finish each step quickly and support many concurrent agents without increasing per-agent wait times. The constraint is power. An AI factory has a fixed amount of electricity available, so operators need infrastructure that gets the most useful work out of each watt.
The pitch: one CPU for speed and capacity
Vera is presented as a single CPU that combines fast agent performance with high capacity per watt. The pitch is to avoid pairing separate x86 parts optimized for speed or density to balance those trade-offs, and to use one CPU across many agent workloads.
The claim matters because agent workloads often combine short, sequential operations with high concurrency. A CPU that is fast for one agent may not scale well if many agents run at the same time, while a dense CPU may keep power low but leave work queued. The comparison is meant to address that tension.
NVIDIA reports that Vera provides industry-leading agent speed and 1.6x more agents per watt than the fastest x86 CPU. It also reports 2x faster agent speed and leading energy efficiency against the most energy-efficient x86 CPU.
Those figures point to the central question for infrastructure buyers: whether one processor family can handle a range of agent workloads inside the same power envelope, instead of requiring separate parts for different tasks.
Early adopter tests cover several agent patterns
Partners are already using Vera, and NVIDIA reported similar results in production-style tests. The examples show where the gains may be most visible: coding pipelines, orchestration layers, query engines, benchmarking, and sandbox creation.
Perplexity reported 1.5x faster agentic coding on Vera than on traditional CPUs, and it plans to use Vera in an upcoming production system.
For its production agent workload, DeepInfra measured agentic orchestration at up to 2.2x the speed of the best x86 processor. It was the fastest of the four architectures tested.
ClickHouse used its standard binary, without Vera-specific tuning, and reported near-linear scaling across 176 cores plus a top spot on the ClickBench hardware leaderboard.
DuckLabs reported median TPC-H SF1000 performance about 1.5x that of an Intel Xeon 6 baseline.
Daytona reported bringing 1,000 agent sandboxes online in 10.8 seconds on Vera, versus 88.4 seconds on x86, and completing 12.7 jobs per second at 2,000 concurrent sandboxes.
Modal reported up to 2x sandbox throughput per core versus comparable x86 hardware, with the advantage expanding under heavy load.
What this means for AI infrastructure
The reported results suggest that agent workloads may reward a CPU design focused on execution speed and power efficiency together. For providers, the value is not only making a single agent respond faster. It is fitting more agents into the same megawatt budget while keeping queues short.
The 100-megawatt comparison, the partner deployments, and the per-core and per-watt claims make this a test of whether Vera can serve as a common CPU foundation for agent-heavy environments. The next question is how broadly these gains hold as production systems scale beyond controlled tests. NVIDIA also points to additional information about the Vera CPU.
Meta description: NVIDIA Vera CPU helps AI agents run faster and more efficiently in 100-megawatt data centers, based on partner tests across coding, orchestration and sandboxes.
Tags: AI agents, NVIDIA Vera CPU, data center efficiency, agentic orchestration, sandbox throughput
Featured image: Abstract friendly illustration of a CPU guiding many small agent workflows through power-limited data center racks, no real people, no logos, no text.
Comments (0)