Nvidia's Vera Rubin enters full production with 10x agent throughput

Nvidia's Vera Rubin platform — delivering 10x agent throughput over Grace Blackwell — is now in full production across 350 factories worldwide.

Nvidia's Vera Rubin platform, offering 10x the agent throughput of its predecessor, is ramping across 350-plus factories in 30 countries, giving cloud providers a purpose-built engine for million-GPU 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," Jensen Huang, founder and CEO of Nvidia, said at GTC Taipei. "Vera Rubin was built for this moment."

The five-rack POD system integrates Vera CPUs, Rubin GPUs, BlueField-4 DPUs and Spectrum-X Ethernet Photonics switches with co-packaged optics at 200Gb/s SerDes — delivering 5x better power efficiency and 5x longer AI uptime than traditional transceiver-based networks, Nvidia said. Production shipments begin this fall.

Nvidia, which reported $81.6 billion in Q1 revenue — up 85% year over year — trades at roughly 30x forward earnings. The Vera Rubin ramp extends its lead in AI infrastructure at a time when rivals including AMD and custom-chip efforts from Amazon and Google are targeting the same data center wallet.

The Vera Rubin ramp marks the third generation of Nvidia's MGX rack-scale systems. More than 150 supply chain partners in Taiwan alone — including Dell Technologies, HPE, Lenovo, Supermicro, Foxconn, Quanta Cloud Technology and Wistron — are manufacturing Vera Rubin-based systems at scale. The platform's open-source MGX design allows hundreds of ecosystem partners across 350-plus factories in 30 countries to participate, Nvidia said.

Spectrum-X and the Million-GPU Fabric

To connect Vera Rubin PODs into million-GPU clusters, Nvidia introduced Spectrum-X Ethernet Photonics, the industry's first co-packaged-optics-based switches now in production. The technology frees power typically consumed by traditional transceivers, enabling 5x longer AI uptime and 1.3x faster deployment. CoreWeave, Lambda and Oracle Cloud Infrastructure are among the first adopters.

The platform also integrates BlueField-4 DPUs with software-defined networking at 800Gb/s and full-stack Confidential Computing, encrypting data across high-speed interconnects at rack scale. Cloud providers including Microsoft Azure, IBM Cloud and CoreWeave are adopting Nvidia's confidential computing framework, the company said.

DSX: The AI Factory Playbook

Alongside the Vera Rubin ramp, Nvidia unveiled the DSX platform — a complete design and operations framework for AI factories. DSX MaxLPS software maximizes token performance per megawatt by combining 45-degree-Celsius liquid cooling with in-rack optimization, letting operators run up to 40% more GPUs at their most efficient operating point. DSX OS provides open-source lifecycle management, multi-tenant operations and health automation.

Dell Technologies, HPE, Lenovo and Supermicro, along with ASUS, Foxconn, GIGABYTE and QCT, are building DSX-ready systems. The DSX Flex pilot with Emerald AI and Silicon Valley Power demonstrates grid-responsive AI factories that adjust power consumption in response to utility signals.

Nvidia's data center revenue has been the primary driver of its 85% year-over-year growth, and the Vera Rubin cycle represents the next catalyst for that segment. With a median analyst price target of $303.50 — implying roughly 44% upside from the $211.26 close — the market is pricing in sustained demand for Nvidia's full-stack AI infrastructure. The DSX platform, by lowering total cost of ownership through token-per-megawatt optimization, could accelerate enterprise adoption beyond the hyperscaler base.

This article is for informational purposes only and does not constitute investment advice.