APREX · EST. 2026
NVIDIA has completed delivery of its first Vera processors to some of the world's most prominent AI organizations. According to NVIDIA's own blog post, the units were personally delivered by Ian Buck, the company's Vice President of Hyperscale and High-Performance Computing, to Anthropic in San Francisco, OpenAI in Mission Bay, and SpaceXAI in Palo Alto — followed by a delivery to Oracle Cloud Infrastructure in Santa Clara the following Monday.
A CPU Designed for a New Era
Vera represents a strategic move for NVIDIA: the company is transitioning from the Grace architecture toward a processor explicitly built for agentic AI — systems in which artificial intelligence operates autonomously, executes tasks in sequences, and coordinates complex workflows without continuous human oversight.
The processor is built around 88 in-house-designed "Olympus" cores compatible with the Armv9.2 instruction set. According to NVIDIA's technical documentation, the core design is optimized for branch-intensive code, sandboxed execution, and orchestration — workloads that are characteristic of AI agents.
Vera is the first CPU NVIDIA has built specifically to direct AI agents — not to train models, but to let them act.
Technical Specifications
The memory subsystem stands out in particular: according to NVIDIA, the LPDDR5X solution delivers twice the bandwidth of traditional CPU solutions at under one-third of the power consumption. Combined with the NVLink Chip-to-Chip interconnect, the Vera processor can access the GPU's HBM4 memory directly within a unified address space — eliminating the need for costly data copying between components.
Performance in Practice
NVIDIA states that Vera delivers 1.5 times higher performance than a modern 128-core x86 processor on agentic workloads. It is worth noting that this figure comes from NVIDIA's own benchmarks, and independent verification has not yet been published.
In a rack configuration, 256 liquid-cooled Vera processors can run more than 22,500 concurrent CPU environments at full performance — a density NVIDIA describes as critical for AI factories that must handle tens of thousands of parallel agent instances.
Part of a Larger Platform
The Vera processor is a central component of NVIDIA's Vera Rubin platform, which also includes Rubin GPUs, HBM4 memory, NVLink 6, ConnectX-9 SuperNIC, and BlueField-4 DPUs. NVIDIA refers to this as a "six-chip" ecosystem tailored for models with trillions of parameters.
Among the companies collaborating with NVIDIA on the Vera rollout are Alibaba Cloud, ByteDance, Meta, Oracle Cloud Infrastructure, CoreWeave, Lambda, Nebius, and Nscale. The processor is now in full production, and broader availability through partner channels has been announced for the second half of 2026, according to NVIDIA's blog.