NVIDIA Ramps Vera Rubin Production Across 350 Factory Sites
NVIDIA said on July 21 that Vera Rubin NVL72 production is ramping across more than 350 factory sites in 30 countries, with racks running at four cloud providers. Separately, Wccftech reported that hardware engineering chief Andrew Bell said about a dozen partners could produce up to 1,000 Vera racks a day; that is stated capacity, not documented output or customer availability.
NVIDIA said on July 21 that Vera Rubin NVL72 production is ramping across more than 350 factory sites in 30 countries. The company said racks are running at CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure, supported by more than 300 global partners.
Separately, Wccftech reported that NVIDIA hardware engineering senior vice president Andrew Bell said about a dozen manufacturing partners could produce up to 1,000 Vera racks per day. Wccftech attributed that statement through The Information. The figure describes potential partner capacity; the retrieved sources do not show that 1,000 completed racks are currently leaving factories each day.
Production scale and deployment are different measures
NVIDIA's announcement documents a broad supply-chain ramp and early systems operating at named cloud providers. It does not publish daily completed-rack output, order volume, delivery schedules, or generally available customer capacity. Treating the 1,000-rack figure as current production would overstate the evidence.
Tom's Hardware separately reported that NVIDIA hyperscale and high-performance computing vice president Ian Buck said the company had shipped hundreds of thousands of standalone Grace servers. That installed base provides context for NVIDIA's CPU ambitions, but it is not a Vera Rubin shipment count and should not be combined with the reported rack-capacity figure.
What the rack contains
Vera Rubin NVL72 is an integrated platform rather than a standalone processor launch. NVIDIA describes a system spanning seven chips and five rack trays, with Vera CPUs, Rubin GPUs, networking, storage processing, and liquid-cooling infrastructure designed together. The company also cited a CoreWeave benchmark showing 10 times more DeepSeek-R1 throughput per megawatt than Grace Blackwell NVL72. That result is a partner benchmark on early hardware, not an independent market-wide comparison.
For infrastructure teams, the important signal is that NVIDIA and its partners are preparing rack-scale manufacturing and deployment capacity before broad availability. Actual usable capacity will still depend on completed systems, site power, cooling, networking, validation, and delivery schedules. The factory-site count and reported daily ceiling indicate supply-chain ambition; they do not guarantee immediate access to production clusters.
Key Points
- 1NVIDIA says Vera Rubin NVL72 production is ramping across more than 350 factory sites in 30 countries.
- 2Wccftech reports potential capacity of up to 1,000 Vera racks per day across about a dozen partners, not verified current output.
- 3Racks are operating at four named cloud providers, but NVIDIA has not disclosed daily shipments or broad customer availability.
Scoring Rationale
The documented global production ramp and reported partner capacity matter for teams planning large AI clusters. The impact remains bounded because daily output and availability are not disclosed, and several performance and capacity figures are company or trade-publication claims.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
