Ask OEMs for a month. NVIDIA's 31 May 2026 GTC Taipei newsroom post says Vera Rubin is ramping into full production and that shipments "are set to begin starting this fall." The same page says products remain when-and-if-available. A purchase order is not a delivery date.
The announcement
The NVIDIA Newsroom post "NVIDIA Vera Rubin Ramps Into Full Production to Power Agentic AI Factories Worldwide," dated 31 May 2026, datelined NVIDIA GTC Taipei, states that Vera Rubin "is ramping into full production to power agentic AI factories worldwide," with Taiwan server makers and global supply-chain leaders manufacturing systems at scale.
The news summary says Vera Rubin delivers a POD-scale foundation "with 10x agent throughput at scale compared with the previous-generation NVIDIA Grace Blackwell platform." It says hundreds of supply-chain partners, "150 in Taiwan alone," across "350+ factories and 30 countries" are ramping Vera Rubin, on an "open source MGX design."
Named system builders in full-scale production include Dell Technologies, HPE, Lenovo, and Supermicro, plus a longer list (ASUS, Foxconn, GIGABYTE, QCT, Wistron, Wiwynn, and others). The platform is described as five purpose-built racks: Vera Rubin NVL72, Vera CPU, Groq 3 LPX, Vera BlueField-4 STX storage, and Spectrum-6 SPX Ethernet.
The post says Vera Rubin introduces Spectrum-X Ethernet Photonics, "now in production," combining co-packaged optics with Spectrum-X switching, and claims "5x better power efficiency, 5x longer AI uptime and 1.3x faster time to deployment than networks using traditional transceivers." CoreWeave, Lambda, and Oracle Cloud Infrastructure are named among first adopters of that fabric.
Under Availability: "Production shipments of Vera Rubin are set to begin starting this fall."
The same page states that many products and features "remain in various stages and will be offered on a when-and-if-available basis," and that statements "are not intended to be, and should not be interpreted as a commitment, promise, or legal obligation."
Who NVIDIA named
NVIDIA shipped the announcement. Bound as named manufacturers are the listed OEMs and ODMs. Named clouds adopting Confidential Computing on the post include CoreWeave, Microsoft Azure, IBM Cloud, Lambda, Nebius, Nscale, and others listed there. Buyers who signed 2026 GPU supply deals are counterparties to those contracts, not to this press page.
What is actually new
Relative to CES-era "full production" chip language, this post is a GTC Taipei ramp: factories, OEM names, a fall shipment window, and a POD of five rack types including Groq 3 LPX and Spectrum-X Photonics "now in production." The operator fact is supply-chain, not a public benchmark PDF.
What the post does not settle
Date: 31 May 2026. Shipments: "this fall." Calendar month UNKNOWN. Price: none in the post. Region: global partner list; no export-license discussion on this page. Hardware: NVL72 plus four other rack types; Spectrum-X Photonics CPO switches "now in production" per NVIDIA. Data: 10x agent throughput vs Grace Blackwell is NVIDIA-reported; workload and SKU UNKNOWN. License: MGX design called open source; SPDX of that design is UNKNOWN from the newsroom page. When-and-if-available disclaimer applies.
What to do now
Cluster buyers with 2026 Rubin reservations should ask Dell, HPE, Lenovo, or Supermicro for a shipment month, not "fall," and should write the when-and-if-available clause next to any verbal date. Capacity planners comparing token cost to Blackwell should distrust 10x until they have a named workload, precision, and rack config.
Whether any OEM will quote a firm 2026 delivery week is UNKNOWN.