NVIDIA launches 88-core Vera CPU, targets AI data center dominance

- Vera is NVIDIA’s first CPU built for AI agent workloads, featuring 88 custom cores and debuting with OpenAI, Anthropic, and Oracle Cloud - Early benchmarks indicate up to 1.8x faster AI agent task completion than leading x86 chips, challenging Intel and AMD On May 31, 2026, Tom’s Hardware reported that NVIDIA had announced its Vera CPU, describing it as the company’s first processor designed specifically for AI agent workloads in data centers. The outlet noted that Vera features 88 custom Olympus cores and introduces Spatial Multithreading technology, representing NVIDIA’s most assertive entry into the CPU market and a direct challenge to Intel and AMD’s dominance. According to the same report, initial adopters include Anthropic, OpenAI, Oracle Cloud Infrastructure, SpaceXAI, ByteDance, and CoreWeave, while enterprise and OEM partners such as Dell, HPE, Lenovo, Supermicro, ASUS, Foxconn, and Quanta are expected to begin shipping Vera-based systems, with customer deployments scheduled for the third quarter of 2026. On 2026-06-01, Tom’s Hardware reported that NVIDIA’s internal benchmarks showed Vera achieving up to 1.8x faster completion of agentic AI tasks compared with leading x86 chips, suggesting a significant performance uplift for AI-centric workloads. In addition, Phoronix reported on June 1, 2026, that its initial performance analysis indicated Vera outperformed AMD’s EPYC 9575F by approximately 10% in Linux kernel compilation tasks, which suggests growing ARM CPU competitiveness for data centers. However, both outlets emphasized that independent third-party evaluations are still forthcoming as the industry assesses real-world workload performance and application compatibility. On 2026-06-01, Tom’s Hardware further reported that Vera plays a central role in NVIDIA’s expanding “AI factory” ecosystem and serves as the CPU foundation for the Vera Rubin platform and associated networking solutions. According to the same coverage, the chip delivers up to 1.2 TB/sec LPDDR5X memory bandwidth while keeping power consumption below 30W, and NVIDIA is therefore targeting high throughput and efficiency in AI computing with this design. According to NVIDIA’s press materials cited by Tom’s Hardware on 2026-06-01, the company is positioning Vera’s launch as a pivotal move to redefine hyperscale computing. The reporting explained that NVIDIA aims to control both AI acceleration and the underlying CPU infrastructure for next-generation data centers, and that customer deployments and comprehensive performance reviews are planned for the third quarter of 2026, with industry attention focused on workload performance and x86 compatibility.