en.Wedoany.com Reported - Recently, during related events in Taipei, NVIDIA announced that the NVIDIA Vera Rubin platform is entering full production, set to support the global construction of next-generation agentic AI factories. Targeting AI labs, cloud service providers, and hyperscale data center customers, production shipments are planned to begin this fall.

Vera Rubin is NVIDIA's POD-level platform for next-generation AI infrastructure, shifting its core focus from individual GPUs or single servers to system-level delivery of entire racks, clusters, and full AI factories. NVIDIA states that the Vera Rubin NVL72 system integrates components such as the Vera CPU, Rubin GPU, NVLink high-speed interconnect, BlueField-4 data processor, and Spectrum-X Ethernet photonic network into a unified architecture, with five dedicated racks operating collaboratively to form a massive computing unit for agentic AI workloads. Compared to the previous generation Grace Blackwell platform, Vera Rubin can achieve up to a 10x improvement in scaled agent throughput, handling more complex AI task chains including inference, retrieval, tool invocation, multi-step task planning, and response generation.

The key context behind this accelerated production is the transition of enterprise infrastructure from traditional data centers to the "AI factory" model. AI factories require not only computing chips but also complete server systems, liquid cooling systems, optical interconnect networks, storage, data processors, security isolation, scheduling software, and operational support. In its announcement, NVIDIA noted that its MGX open rack design has formed a mature supply chain ecosystem, with hundreds of global partners advancing Vera Rubin production. In Taiwan alone, approximately 150 ecosystem partners are involved, with related production activities spanning 30 countries and over 350 factories. Dell, HPE, Lenovo, Supermicro, as well as ASUS, Foxconn, Gigabyte, Quanta Cloud Technology, Wistron, and Wiwynn, are among the system and manufacturing companies on the ecosystem list.

The Taiwan supply chain is playing a more concentrated manufacturing role in this round of AI infrastructure expansion. According to NVIDIA's blog, Taiwan has over 500 NVIDIA ecosystem partners, and more than one million MGX rack components for Vera Rubin infrastructure will be integrated across 25 factory sites in Taiwan. Companies such as TSMC, ASE Technology Holding (including its packaging, testing, and substrate-related subsidiaries), Foxconn, Quanta, Wistron, and Inventec cover areas including wafer manufacturing, advanced packaging, substrates, server manufacturing, system integration, and factory operations. The entry of Vera Rubin into production means that AI chip competition is extending to the delivery capabilities of entire infrastructure sets, with server manufacturing, optical communications, liquid cooling, power supply, cabinets, storage, and networking equipment companies entering a higher-intensity cycle of capacity coordination.

Network architecture is also a significant change in the Vera Rubin platform. NVIDIA simultaneously launched Spectrum-X Ethernet Photonics, describing it as a switching platform based on co-packaged optics technology, building the network foundation for AI factories with millions of GPUs. As AI cluster scales expand, traditional data center networks face increasing pressure in terms of power consumption, latency, reliability, and the complexity of optical module deployment. Co-packaged optics bring optical interconnect capabilities closer to the switching chip, helping to reduce energy consumption and improve connection efficiency for large-scale GPU clusters. For cloud service providers and hyperscale data center customers, the bottleneck of AI factories has shifted from "having enough GPUs" to "whether computing, networking, storage, power supply, and cooling can be stably organized into a sustainably operating production system."

The full production of Vera Rubin will also strengthen NVIDIA's platform control in enterprise AI infrastructure. As agentic AI enters enterprise software, R&D design, financial risk control, manufacturing operations, medical research, and automated office scenarios, enterprises need to handle more proprietary data, long-term context, and multi-tenant tasks. The security, isolation, encryption, and operational capabilities of computing platforms will become procurement priorities. NVIDIA is bundling Vera Rubin with the DSX data center design and operations platform, BlueField-4 security and networking capabilities, and the DOCA software stack, aiming to upgrade chip sales into the output of a complete AI factory architecture. In the coming years, AI infrastructure competition will increasingly focus on platform generations, supply chain delivery, energy efficiency, network scale, and enterprise-grade security capabilities.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com