Nvidia's Self-Developed CPU "Vera" Officially Delivered

Nvidia has taken a substantial step in the commercialization of its self-developed CPU. According to Nvidia's official website, the first batch of Vera CPUs was officially delivered to customers last Friday, with recipients including three top AI labs: Anthropic, OpenAI, and SpaceX AI.

Oracle Cloud Infrastructure (OCI) completed the handover on Monday. Ian Buck, Nvidia's Vice President of Hyperscale and High-Performance Computing, personally attended each delivery ceremony. This milestone marks the formal transition of the "next multi-billion dollar business" defined by Jensen Huang at the GTC conference in March this year from a strategic vision to a commercial reality.

The core logic driving this strategy is the structural migration of AI workloads. As large models evolve from question-and-answer reasoning to the Agentic paradigm of autonomous task execution, the computing bottleneck is shifting from the GPU side to the CPU side—tool invocation, code execution, sandbox operation, long context retrieval, and orchestration scheduling are all handled by the CPU.

Buck stated directly: "When an AI model faces a problem, the answer is not readily available. The model actually needs to generate Python code to derive the correct result. This is the fundamental reason for the sharp increase in CPU demand."

In terms of hardware specifications, Vera is equipped with 88 Nvidia-developed Olympus cores, with a memory bandwidth of 1.2 TB/s. In a fully loaded state, the single-core performance is 50% higher than the previous generation. As a core component of Nvidia's co-design strategy, Vera also serves as the master processor for the next-generation Vera Rubin NVL72 system, directly connected to two Rubin GPUs via second-generation NVLink-C2C to build a unified memory architecture, with approximately twice the energy efficiency compared to traditional heterogeneous solutions.

James Bradbury, Head of Compute at Anthropic, stated that computing power expansion is a core accelerating variable for model growth and looks forward to Vera demonstrating differentiated performance in Agentic workloads. Karan Batta, Head of Product Management, revealed that Oracle plans to deploy hundreds of thousands of Vera CPUs starting in 2026, becoming the first cloud service provider to deploy the chip on a large scale, stating that its architecture is well-suited in terms of efficiency, density, and space occupation to drive next-generation enterprise-level AI needs.

SpaceX AI is currently evaluating the introduction of Vera into reinforcement learning and Agent-based simulation training pipelines. It is reported that Elon Musk conducted in-depth inquiries on technical details such as core architecture, memory layout, and heat dissipation design at the delivery site.

The delivery of Vera means that Nvidia is extending its territory from the monopolistic advantage of GPUs to the CPU market, which has long been dominated by Intel and AMD. Under the narrative framework of Agentic AI reconstructing the data center architecture, Nvidia is attempting to strengthen customer lock-in with a "GPU+CPU+DPU+network" full-stack solution. OCI's large-scale procurement commitment has now become the first heavyweight validation of this strategy's implementation.