NVIDIA Vera CPU Targets Agentic AI With 88-Core Design

Rongchai Wang Mar 16, 2026 20:06

NVIDIA launches Vera CPU with 88 custom cores and 1.2 TB/s memory bandwidth, claiming 50% faster performance than traditional CPUs for AI workloads.

NVIDIA unveiled its Vera CPU at GTC on March 16, positioning the 88-core processor as purpose-built for agentic AI and reinforcement learning workloads. The company claims Vera delivers 50% faster single-threaded performance and twice the efficiency of traditional rack-scale CPUs.

The announcement comes as NVIDIA stock trades at $180.25, with the company's market cap sitting at $4.43 trillion. Vera represents NVIDIA's bid to capture more of the data center CPU market currently dominated by Intel and AMD.

Technical Specifications

Vera features 88 custom-designed Olympus cores supporting NVIDIA Spatial Multithreading, enabling 176 threads per CPU. The LPDDR5X memory subsystem delivers up to 1.2 TB/s bandwidth while supporting 1.5 TB of memory—a threefold increase over the Grace CPU predecessor.

When paired with NVIDIA GPUs through NVLink-C2C interconnect, Vera achieves 1.8 TB/s of coherent bandwidth. That's 7x the bandwidth of PCIe Gen 6, according to NVIDIA.

A new Vera CPU rack configuration packs 256 liquid-cooled processors, capable of sustaining over 22,500 concurrent CPU environments running independently at full performance.

Major Customers Already Onboard

The deployment list reads like a who's who of cloud infrastructure: Alibaba, ByteDance, Meta, Oracle Cloud Infrastructure, CoreWeave, Lambda, Nebius, and Nscale. Manufacturing partners include Dell Technologies, HPE, Lenovo, Supermicro, ASUS, Foxconn, and Quanta Cloud Technology.

"Vera is arriving at a turning point for AI. As intelligence becomes agentic—capable of reasoning and acting—the importance of the systems orchestrating that work is elevated," said Jensen Huang, NVIDIA's CEO. "The CPU is no longer simply supporting the model; it's driving it."

Early Benchmark Claims

Redpanda, a streaming data platform, reported testing results showing up to 5.5x lower latency running Apache Kafka-compatible workloads compared to other benchmarked systems.

The Texas Advanced Computing Center noted "impressive early results" across six scientific applications while preparing for deployment in their upcoming Horizon system. National laboratories including Los Alamos and Lawrence Berkeley's NERSC also plan deployments.

Availability and Market Context

Vera is in full production with partner availability expected in the second half of 2026. The processor slots into NVIDIA's broader Vera Rubin platform strategy, competing directly with AMD's EPYC and Intel's Xeon lines in the AI-optimized data center segment.

For infrastructure investors watching AI compute buildouts, Vera's adoption rate among hyperscalers will signal whether NVIDIA can extend its GPU dominance into the CPU layer of AI factories.

• nvidia
• vera cpu
• agentic ai
• data center
• ai infrastructure