NVIDIA crossed a significant inflection point this week with the arrival of its first custom-designed CPU, Vera, at three of the world's leading AI research labs: Anthropic, OpenAI, and SpaceX AI, followed by Oracle Cloud Infrastructure. The move represents NVIDIA's most direct challenge to traditional CPU vendors like Intel and AMD, and signals the company's confidence in building complete computing systems rather than remaining purely a GPU supplier. CEO Jensen Huang emphasized the urgency of this transition during his Dell Technologies World keynote, declaring that "demand is going parabolic, utterly parabolic," and positioning Vera as essential infrastructure for the next compute paradigm.
Vera is purpose-built for agentic AI inference, addressing a specific bottleneck in the emerging category of reasoning AI systems that make decisions and take actions autonomously. According to NVIDIA's specifications, the NVL72 variant delivers agentic AI inference at one-tenth the cost per token compared to traditional approaches, while agent sandboxes run 50% faster on Vera than on conventional CPUs. Enterprise data queries are up to 3x faster, metrics that appeal directly to the 5,000 enterprises—including Lilly, Samsung, and Honeywell—already running AI workloads. This performance-per-watt advantage matters enormously in hyperscale data centers facing staggering power constraints.
The Vera launch reinforces NVIDIA's transformation from a GPU-centric vendor into a full-stack AI infrastructure provider, a strategy being accelerated across multiple fronts. NVIDIA is simultaneously deepening partnerships with cloud providers like Google Cloud, expanding developer ecosystems through events at GTC Taipei and Google I/O, and rolling out specialized architectures including the Blackwell GPU line. Analysts suggest this vertical integration could create a hidden $60 billion business potentially overtaking Broadcom, cementing NVIDIA's dominance not just in accelerators but across the entire AI compute stack.