NVIDIA delivered its first Vera CPUs to Anthropic, OpenAI, and SpaceX AI last week, marking the company's entry into CPU design and reflecting a strategic shift toward infrastructure for agentic AI workloads. The move signals confidence that autonomous agents—AI systems that execute multi-step tasks with limited human intervention—represent a distinct compute problem requiring specialized hardware. Unlike traditional inference, agentic systems must handle extended reasoning chains, tool use, and real-time decision-making, creating different performance bottlenecks than standard large language model serving. By placing Vera units at three of the world's most influential AI research organizations, NVIDIA positioned itself to influence how the next generation of AI systems gets built and deployed.
According to NVIDIA leadership, Vera delivers measurable advantages for agentic workloads. During Dell Technologies World, CEO Jensen Huang highlighted that the Vera NVL72 achieves agentic AI inference at one-tenth the cost per token compared to traditional approaches. Additionally, NVIDIA reported that agent sandboxes run 50 percent faster on Vera than on traditional CPUs, while enterprise data queries execute up to 3x faster when paired with Vera processors. These claims position the chip as a practical alternative for production agentic systems, not merely research platforms. The subsequent delivery to Oracle Cloud Infrastructure on Monday extended reach into commercial infrastructure providers, suggesting NVIDIA expects rapid adoption.
The strategic significance extends beyond hardware specs. NVIDIA is responding to what CEO Huang called 'parabolic' demand for AI compute across thousands of enterprises—Lilly, Samsung, and Honeywell among them—already running AI workloads. However, the agentic CPU launch also reflects a business reality: building general-purpose agents requires rethinking compute architecture. By offering a purpose-built CPU alongside its dominant GPU ecosystem, NVIDIA increases switching costs and deepens enterprise dependency on its full-stack platform. The presence of Vera at OpenAI and Anthropic matters less as immediate revenue and more as validation that heterogeneous compute—combining GPUs, CPUs, and specialized processors—will define competitive AI infrastructure through 2025.