NVIDIA Vera CPU offers 1.6x more agent density per watt

NVIDIA announced the Vera CPU on 5 November 2026, claiming the architecture delivers 1.6x more agents per watt than the fastest x86 CPU and 2x faster agent speed compared to the most energy-efficient x86 option. (x86: a common family of instruction set architectures used in most standard personal computers and servers.) The hardware targets AI factories where power constraints limit the density of agentic AI workloads, requiring a balance of high throughput and high capacity per watt.

How AI agents move beyond generation to CPU execution

AI agents perform more than simple text generation; they search for information, query databases, write and test code, and execute specific actions. Each of these steps relies on a CPU because the agents move between GPU reasoning and CPU execution throughout a single task. This shift means that AI infrastructure providers must support a high volume of concurrent agents without increasing wait times for any single task. Because AI factories are power limited, the primary constraint for operators is maximizing the work done for every watt of electricity provided to the hardware.

The NVIDIA Vera CPU addresses this by providing both fast agent performance and high

The NVIDIA Vera CPU addresses this by providing both fast agent performance and high capacity per watt within a single processor. By achieving both metrics simultaneously, Vera allows providers to use one CPU for a range of agent workloads rather than having to mix and match different x86 parts to balance speed and capacity. This simplification reduces the complexity of hardware selection while maintaining the performance necessary for real-time agentic responses.

NVIDIA Vera achieves 1.6x more agents per watt

NVIDIA found that a single Vera CPU performs better than the fastest x86 CPU and the most energy-efficient x86 CPU in a 100-megawatt data center comparison. The platform delivers 1.6x more agents per watt compared to the fastest x86 CPU, which directly influences the power usage effectiveness (PUE) of an AI factory. (power usage effectiveness (PUE): a measure of how efficiently a data center uses energy to power computing equipment versus overhead.) Furthermore, Vera delivers 2x faster agent speed with leading energy efficiency when compared to the most energy-efficient x86 CPU.

The architecture's efficiency

The architecture’s efficiency is validated by several production-style tests from early adopters. Perplexity reported that agentic coding tasks ran 1.5x faster with Vera than on traditional CPUs. DeepInfra achieved up to 2.2x speed of the best x86 processor on agentic orchestration for their production agent workload, marking it as the fastest of the four architectures tested. These results indicate that the Vera CPU handles the execution-heavy “actions” of AI agents more efficiently than standard x86 alternatives.

High-concurrency throughput and 176-core scaling

The NVIDIA Vera CPU supports massive concurrency through high sandbox throughput—the speed of work completed in a given time period. Daytona demonstrated that Vera could bring 1,000 agent sandboxes online in 10.8 seconds, a significant improvement over the 88.4 seconds required by x86 hardware. Under heavy load, Vera achieved 12.7 completed jobs per second at 2,000 concurrent sandboxes. Additionally, Modal reported that the CPU achieved up to 2x sandbox throughput per core compared to comparable x86 hardware, with those gains increasing as the load becomes heavier.

Standard software performance also scales effectively on the 176-core architecture.

Standard software performance also scales effectively on the 176-core architecture. ClickHouse reported that its standard binary, which required no Vera-specific tuning, scaled near-linearly across the 176 cores and topped the ClickBench hardware leaderboard. (near-linearly: scaling where the output increases in direct proportion to the increase in the number of processing cores.) This near-linear scaling allows standard binaries to perform at the top of the hardware leaderboard without requiring specific software engineering to optimize for the high core count. Because Vera achieves both fast agent speed and high capacity per watt, providers can use one CPU for a range of workloads instead of mixing and matching different x86 parts.

Vera vs. x86 performance highlights

Simplifying AI factory infrastructure

NVIDIA Vera simplifies the bill of materials (BOM) by allowing providers to avoid mixing x86 parts to balance density and speed. (bill of materials (BOM): a comprehensive list of every raw material and component required to manufacture or assemble a finished product.) Because the chip provides both high capacity and high speed, providers stop needing to manage complex hardware configurations that pair fast CPUs with dense ones. This reduces the risk of performance bottlenecks in high-concurrency environments while staying within the power limits of the data center.

This architecture’s advantage relies on the assumption that agents will continue to move between GPU reasoning and CPU execution. If the workload shifts such that GPU execution handles the majority of action steps, the relative advantage of Vera’s CPU-specific optimizations would diminish. However, as long as agentic orchestration—the process of managing multiple AI bots performing multi-step tasks—requires high-throughput CPU execution, Vera remains a primary candidate for power-constrained AI factories.

Who this is for: Data center architects

Data center architects who manage 100-megawatt sites should monitor the power usage effectiveness (PUE) to ensure NVIDIA Vera meets the per-watt efficiency targets, as it delivers 1.6x more agents per watt than the fastest x86 CPU. AI infrastructure providers who manage VMware ESXi hypervisors should configure the environment to support agents on NVIDIA Vera, as it provides the high throughput and capacity required for agentic workloads. Cloud hardware specifiers who manage the bill of materials (BOM) should replace mixed x86 parts with NVIDIA Vera CPUs to simplify hardware selection and improve consistency. Backend developers building coding agents should verify that Vera provides the 1.5x faster performance for agentic coding tasks reported by Perplexity.

Who this is not for: Home PC gamers

Home PC gamers should skip this analysis as the source focuses on data center infrastructure and agentic workloads rather than gaming benchmarks; these users should look for consumer GPU and CPU gaming reviews instead. Front-end web developers can ignore this hardware scaling story, as it does not affect web performance or frontend frameworks; these users should monitor web performance and frontend framework news instead.

Cloud infrastructure scenario: Mei’s AI factory

Mei runs a cloud provider and is building an AI factory that must support 10,000 concurrent AI agents within a 100-megawatt power limit. She must choose between using the fastest x86 CPUs, which may exceed her power limits, or using the most energy-efficient x86 CPUs, which may be too slow to handle the 12.7 jobs per second required for her workload. Mei chooses NVIDIA Vera because it allows her to run 1.6x more agents per watt than the fastest x86 alternative, helping ensure she stays under her power budget while maintaining high throughput.

If Mei had instead chosen the fastest x86 CPUs, she would have been forced to reduce the number of agents to stay within the 100-megawatt limit. By choosing Vera, she successfully hosts more agents while maintaining the speed required for agentic orchestration.

What this can be used for

What this replaces: NVIDIA Vera reduces

NVIDIA Vera reduces the need for mixing and matching fast and dense x86 parts to balance speed and capacity. By providing a single CPU that achieves both high capacity and high speed, it simplifies the bill of materials (BOM) and reduces the engineering complexity of managing heterogeneous hardware for agentic AI. (heterogeneous hardware: a computing system that uses different types of processors or architectures to perform various tasks.)

Key terms

Agentic orchestration

Managing the execution of multiple AI agents performing multi-step tasks.

Throughput

The speed of work, or the amount of work completed in a given time period.

Power limited

Infrastructure constraints where the amount of work is capped by available electricity or heat dissipation.

Agent sandbox

A safe testing area, or an isolated environment where an AI agent can write and test code safely.

Source story: Vera Delivers the Best of Both Worlds: Highest Throughput and Fastest Agents, published by NVIDIA on 5 November 2026.