NVIDIA moved the NVIDIA Vera Rubin platform into full production, marking the company’s next major step in scaling infrastructure for agentic AI workloads. Announced at NVIDIA GTC Taipei, Vera Rubin is now being manufactured by Taiwan’s leading server builders and NVIDIA’s broader global supply chain ecosystem. NVIDIA positions the platform as a rack-scale foundation for “AI factories,” targeting hyperscalers, cloud providers, and AI labs building large-scale infrastructure for autonomous reasoning and inference workloads.
Vera Rubin is built as a five-rack POD architecture combining NVIDIA Vera CPU, Rubin GPUs, NVIDIA BlueField-4 storage processing, NVLink interconnect, and NVIDIA Spectrum-X switching into a single integrated system. NVIDIA says the configuration delivers up to 10x higher agent throughput at scale compared with the prior-generation Grace Blackwell platform. Production partners include Dell Technologies, Hewlett Packard Enterprise, Lenovo, Supermicro, Foxconn, Quanta Cloud Technology, Wistron and Wiwynn. NVIDIA said more than 150 partners in Taiwan and over 350 factories across 30 countries are participating in the ramp.
A notable element of the rollout is the production launch of NVIDIA Spectrum-X Ethernet Photonics, which NVIDIA describes as the industry’s first co-packaged-optics Ethernet switching platform using 200Gb/s SerDes. The company says the system provides 5x better power efficiency, 5x longer AI uptime, and 1.3x faster deployment compared with networks based on traditional pluggable transceivers. Vera Rubin also integrates confidential computing and security features through BlueField-4 DPUs and the DOCA software stack to support multi-tenant isolation, zero-trust networking, runtime threat detection, and encrypted communications across large AI clusters. Production shipments are scheduled to begin this fall.
- NVIDIA says Vera Rubin is now ramping through 350+ factories in 30 countries
- Platform architecture combines Rubin GPUs, Vera CPUs, BlueField-4, NVLink, and Spectrum-X Ethernet
- NVIDIA claims up to 10x higher agent throughput versus Grace Blackwell
- CoreWeave, Lambda and Oracle Cloud Infrastructure are among early adopters of Spectrum-X Ethernet Photonics
- Cloud providers adopting NVIDIA Confidential Computing include Microsoft Azure, IBM Cloud, Nebius, Vultrand IREN
- NVIDIA DSX provides the design and operations framework for deployment of Vera Rubin AI factories
“Agentic AI is a new kind of workload. One prompt can launch a thousand-step journey of reasoning, retrieval, tool use and response generation,” said Jensen Huang. “Vera Rubin was built for this moment — an AI factory engine that delivers intelligence at scale, with the performance, efficiency and security needed to power the next industrial revolution.”
🌐 Analysis: Vera Rubin extends NVIDIA’s strategy from selling GPUs toward delivering complete AI factory infrastructure, spanning silicon, networking, optics, security, and reference data center design. The introduction of Spectrum-X Ethernet Photonics is especially significant because it brings co-packaged optics directly into production AI networking, an area the industry has viewed as a long-term roadmap item but is now moving toward deployment much faster due to AI cluster power and scaling demands.
