NVIDIA’s push into the era of agentic AI is no longer limited to GPUs or even rack-scale systems. With its latest wave of announcements, the company is now building what it calls a full-stack “AI factory” architecture, spanning CPUs, networking, storage, security, and system-level orchestration.
At the center of this expansion is NVIDIA Vera, a CPU designed specifically for agentic AI workloads, alongside the Vera Rubin platform, BlueField-4 STX, and the DSX AI Factory reference architecture.
Together, these components mark NVIDIA’s most aggressive attempt yet to redefine how modern data centers are built and operated.
Vera CPU: Built for Agentic Workloads
First introduced earlier this year, NVIDIA Vera is positioned as a CPU purpose-built for agentic AI, reinforcement learning, and large-scale orchestration workloads. Unlike traditional server CPUs that focus on general-purpose compute, Vera is designed to continuously handle the “behind-the-scenes” work of AI systems, such as tool execution, data routing, and multi-step reasoning workflows.
NVIDIA claims Vera delivers up to 1.8x performance compared to conventional x86 CPUs while significantly improving efficiency in AI-heavy environments. The chip features NVIDIA’s custom Olympus cores, spatial multithreading, and a high-bandwidth LPDDR5X memory subsystem delivering up to 1.2TB/s of bandwidth.
More importantly, Vera is designed to operate as a core component across NVIDIA’s entire AI stack. It serves as the host CPU in Vera Rubin systems, standalone Vera servers, and storage-centric AI platforms such as BlueField-4 STX.
Early ecosystem adoption is already broad, with cloud providers and AI labs including OpenAI, Anthropic, Oracle Cloud Infrastructure, ByteDance, and CoreWeave preparing deployments, alongside OEMs such as Dell, HPE, Lenovo, Supermicro, ASUS, and others.
BlueField-4 STX: Bringing Intelligence to Storage
While GPUs and CPUs handle compute, NVIDIA is increasingly focusing on what it considers the next major bottleneck in AI systems: data movement and storage.
The newly introduced BlueField-4 STX platform extends NVIDIA’s DPU strategy into AI-native storage processing. It integrates Vera CPUs, BlueField networking, and the DOCA software stack into a unified architecture designed to handle storage, security, and data orchestration directly within the infrastructure layer.
NVIDIA positions STX as a platform for “agentic AI storage processing,” where AI systems constantly retrieve and update information from enterprise datasets while maintaining strict security and performance constraints.
A key feature of the platform is in-silicon security, which allows policy enforcement, runtime threat detection, and encryption at line speed without relying on host CPU overhead. NVIDIA claims this approach can significantly improve both performance and data protection for enterprise AI workloads.
Vera Rubin Moves Into Full Production
NVIDIA has now confirmed that the Vera Rubin platform is entering full production, marking a major milestone in its transition from announcement to deployment.
The Vera Rubin architecture combines multiple chips into a unified rack-scale system, including Vera CPUs, Rubin GPUs, networking fabrics, and BlueField-4 STX storage systems. These components are designed to operate as a single AI supercomputer rather than discrete hardware blocks.
According to NVIDIA, Vera Rubin systems deliver up to 10x higher agent throughput compared to previous-generation platforms, enabled by tighter integration across compute, networking, and storage layers.
Ecosystem support is already extensive, with more than 150 partners in Taiwan alone and deployments spanning hyperscalers, AI labs, and cloud providers worldwide. Manufacturing is being led by major OEMs including Dell Technologies, HPE, Lenovo, Supermicro, ASUS, Foxconn, Quanta, and Wistron.
DSX AI Factory: Turning Infrastructure Into a Blueprint
Beyond hardware, NVIDIA is also pushing a full system-level framework through its DSX AI Factory initiative.
The DSX platform serves as a reference architecture for building and operating large-scale AI factories. It integrates hardware design, system simulation, digital twins via Omniverse, and infrastructure orchestration into a single blueprint for enterprise deployment.
In essence, DSX is NVIDIA’s attempt to standardize how AI data centers are designed from the ground up, optimizing for token efficiency, energy efficiency, and agent-scale workload management.
The platform is already supported by a wide ecosystem including industrial and infrastructure partners such as Siemens, Schneider Electric, GE Vernova, and Vertiv, highlighting NVIDIA’s push into not just computing, but physical data center design itself.
A Full Stack AI Infrastructure Play
Taken together, Vera, BlueField-4 STX, Vera Rubin, and DSX represent a clear shift in NVIDIA’s strategy. The company is no longer positioning itself purely as a GPU vendor. Instead, it is evolving into a full-stack infrastructure provider for AI factories.
This shift aligns with the broader industry transition from static AI models to agentic systems, where workloads are no longer single inference calls but continuous, multi-step reasoning processes involving tools, memory, and external systems.
In such an environment, performance is no longer defined solely by GPU compute. CPU orchestration, memory bandwidth, storage intelligence, and system-level coordination all become equally critical.
Availability
NVIDIA confirmed that Vera-based systems are already being manufactured by major OEM partners, with standalone Vera CPU servers and integrated systems expected to roll out in the second half of 2026.
BlueField-4 STX platforms are also expected to become available through enterprise infrastructure partners within the same timeframe, targeting large-scale AI deployments in cloud and enterprise environments.
Meanwhile, Vera Rubin systems are already in full production and scaling across global supply chains, forming the backbone of NVIDIA’s next-generation AI factory rollout.
If NVIDIA’s execution matches its ambition, the Vera ecosystem could become the foundation for how enterprise AI infrastructure is built for the next decade.
