NVIDIA has begun shipping its first CPU designed specifically for AI agent workloads. On Friday, Vice President of Hyperscale and High-Performance Computing Ian Buck personally delivered Vera systems to Anthropic in San Francisco and OpenAI in Mission Bay, while SpaceXAI in Palo Alto received units the same day. Oracle Cloud Infrastructure in Santa Clara took delivery on Monday.
The hand-delivered units mark the transition of Vera from announcement to production availability for NVIDIA's most strategic AI customers. The CPU targets inference patterns where agents spend more time reasoning, retrieving context, and executing tool calls than performing the massive matrix multiplications that dominate training workloads.
What's new
Vera is NVIDIA's first CPU built from the ground up for agentic AI rather than general-purpose server workloads. The architecture emphasizes memory bandwidth and low-latency interconnects over raw core count, optimizing for the data-shuttling patterns between GPU accelerators, vector databases, and multi-step agent control flow. NVIDIA has not published detailed specifications, core counts, or clock speeds for Vera in this announcement.
The customer list signals where NVIDIA believes the agent infrastructure market is heading. Anthropic's Claude and OpenAI's GPT models already power agent frameworks that autonomously browse, write code, and manage complex workflows. SpaceXAI, Elon Musk's AI venture, is racing to establish its own agent capabilities. Oracle Cloud Infrastructure gains early access to differentiate its AI infrastructure offerings against AWS, Google Cloud, and Microsoft Azure in the emerging agent-as-a-service market.
Why it matters
Vera represents a direct challenge to AMD's Instinct MI300A, which combines CPU and GPU cores on a single package for AI workloads, and to Intel's Gaudi processors already deployed at scale for inference. By controlling both the CPU and GPU layers, NVIDIA can offer a vertically integrated stack — Vera paired with next-generation Blackwell GPUs — that promises tighter hardware-software co-optimization for agent workloads. Enterprise adoption of AI agents is accelerating faster than analysts predicted six months ago, with production deployments handling customer service, report generation, infrastructure management, and code review.
The deliveries this week are likely pre-production samples for integration testing and benchmarking. Cloud instance types featuring Vera are not expected immediately. Performance benchmarks comparing Vera against established Xeon, EPYC, and MI300A alternatives have not been published.
Our take
NVIDIA's white-glove delivery to the labs defining the agent frontier is a calculated signal: the company intends to own the full inference stack the way it owns training. The real test will come when independent benchmarks arrive and cloud providers publish instance pricing — until then, Vera's advantage over well-tuned Xeon or EPYC configurations paired with Blackwell GPUs remains unproven.
Series: 1. xAI Ships Grok Imagine Image 2.0 With Precise Editing and Top Arena Ranking · 2. SpaceXAI Launches Grok Bot, Persistent AI Teammates With Their Own Cloud Computers · 3. Cursor Acquired by SpaceX, Grok 4.6 Released as First Joint Model · 4. X Open-Sourced the For You Algorithm: Not a Leak — a Transparency Push · 5. NVIDIA ships Vera CPU to Anthropic, OpenAI, Oracle, and SpaceXAI · xAI Grok
