Nvidia begins shipping Vera CPU for AI data centres

Nvidia has begun delivering its Vera central processing unit, marking the company's latest move into data centre computing.

Vera is a custom Arm-based server processor for AI systems and broader data centre workloads. It is designed to work alongside Nvidia's graphics processors in servers used for model training, inference and other computing tasks.

Vera is part of the GB300 NVL72 platform, which combines the new CPU with Nvidia's Blackwell Ultra graphics processors and networking components. The chip includes custom cores and memory features designed to move data more efficiently across systems handling large AI workloads.

The processor is part of a broader push by Nvidia to deepen its role in the server stack beyond graphics chips. While the group dominates the market for AI accelerators, it faces growing competition and customer demand for more tightly integrated systems that combine processors, interconnects and software.

Chief Executive Officer Jensen Huang said the processor had moved from development into shipment. "Vera is now in full production and on its way to customers," Huang said.

Arm architecture has become increasingly important in the data centre market as cloud groups and hardware vendors look for alternatives to x86 processors from Intel and AMD. Several large operators now deploy Arm-based chips in their own infrastructure, and chip designers have expanded their use in AI servers, where power use, system design and workload balance have become more important.

System design

Nvidia has been building a broader product line that more closely links chips, networking and system hardware. Its earlier Grace CPU targeted high-performance computing and AI applications, and Vera extends that strategy with a newer design for next-generation server platforms.

Nvidia argues that moving data quickly between CPUs, GPUs and memory has become critical as AI models grow larger and more complex. Owning more of the computing architecture gives the company greater influence over how customers build and scale AI clusters.

Major cloud providers, sovereign AI projects and enterprise customers have been among the main buyers of Nvidia's systems in recent years. Demand for complete racks and pre-integrated platforms has risen as customers seek faster deployment of AI infrastructure and tighter control over system performance, energy use and networking.

Market pressure

Vera deliveries begin as rivals pursue a larger share of AI infrastructure spending. AMD has expanded its own data centre portfolio, Intel is trying to regain momentum in servers, and several hyperscale operators continue to develop in-house silicon for selected workloads.

Nvidia's advantage has rested not only on chip design but also on its software tools and system integration. By shipping a CPU closely paired with its latest AI platforms, the company is seeking to make that broader ecosystem harder for rivals to displace.

Nvidia did not disclose shipment volumes, customer names or financial terms tied to the start of Vera deliveries in the source material. Even so, the announcement signals that one of its most closely watched server processors has entered the commercial supply chain.