NVIDIA Vera CPU Ships to OpenAI, xAI as AI Server Orders Surge

According to NVIDIA, the company has officially begun shipping its next-generation Vera CPU platform to major North American cloud and AI companies, marking an important milestone for the launch of the new Vera Rubin AI server ecosystem and creating fresh momentum across the AI hardware supply chain.

NVIDIA announced on May 19 that the first batches of Vera CPUs were delivered on May 15 and May 18 to leading AI and cloud computing customers including Anthropic, OpenAI, xAI, and Oracle. NVIDIA’s Vice President of Hyperscale and High-Performance Computing, Ian Buck, personally delivered the new processors to several customers, with xAI founder Elon Musk reportedly receiving the shipment in person.

The shipments signal the beginning of a new AI server deployment cycle centered around NVIDIA’s Vera Rubin platform. Major Taiwan-based electronics manufacturing companies, including Foxconn, Quanta Computer, and Wistron, are expected to benefit from increasing AI server production demand as cloud service providers accelerate infrastructure expansion.

To support growing AI server orders, Quanta Computer recently announced plans to lease a new manufacturing facility in Fremont, California, with the total contract value reaching approximately US$61.71 million. The site will serve as an expanded production base for AI server manufacturing in the United States.

NVIDIA also revealed that Ian Buck demonstrated the internal architecture of the Vera Rubin server platform during a visit to OpenAI’s office. The presentation included detailed explanations of the system’s compute core layout, memory capacity, and platform design, highlighting the company’s strong focus on the rollout of its latest AI infrastructure products.

Industry sources indicate that the new Vera CPU is built on an Arm-based architecture, representing NVIDIA’s first major CPU architecture update since the launch of the Grace CPU platform in 2021.

The Vera CPU features 88 compute cores and supports NVLink interconnect bandwidth of up to 1.8TB per second. The processor also supports 1.5TB of system memory and up to 1.2TB per second of SOCAMM LPDDR5X memory bandwidth, providing optimized performance for Rubin architecture GPUs and next-generation AI agent workloads.

Manufactured using TSMC 3nm process technology, the Vera CPU reportedly uses a simplified 2.5D/3D advanced packaging design compared with the Rubin GPU platform. The streamlined packaging approach has shortened production lead times, allowing the Vera CPU to reach customers ahead of full Rubin platform deployments.