Nvidia’s Vera CPU shifts the server fight to systems

Nvidia’s Vera server CPU moves the company deeper into territory long led by Intel Xeon and AMD EPYC processors. The strategy is less about winning every head-to-head CPU benchmark and more about making Nvidia-designed AI systems the default purchasing choice, with the CPU already integrated alongside GPUs, networking, software, and rack-scale infrastructure.

The company’s CPU push builds on years of Arm-based development. Nvidia agreed in September 2020 to acquire Arm from SoftBank for $40 billion, but regulators opposed the deal and it was terminated in February 2022. Nvidia kept a 20-year Arm architecture license, later developed Grace, and now has Vera, which includes 88 custom Olympus cores designed for workloads around AI, including agent orchestration, data processing, analytics pipelines, sandboxed code execution, and CPU-bound support for accelerated computing.

Early support spans major AI and cloud customers, including Anthropic, OpenAI, SpaceXAI, ByteDance, CoreWeave, Oracle Cloud Infrastructure, and AWS. AWS and Nvidia announced plans for AWS to deploy two million additional Nvidia GPUs during 2027 and 2028 while bringing Vera CPU-based infrastructure to AWS, signaling that Vera can coexist with in-house chips such as AWS Graviton when workloads differ.

The opportunity is expanding as AI reshapes data-center spending. The Information Network estimates AI data-center systems revenue will rise from $560 billion in 2026 to $1.68 trillion in 2030, while AI CPU revenue grows from $38 billion to $155 billion. Vera lets Nvidia pursue a larger share of each AI installation without relying on GPU demand to weaken.