NVIDIA's next-generation Vera Rubin server system is progressing smoothly in mass production. Dozens of customers, including CoreWeave, Microsoft, OpenAI, Anthropic, and SpaceX AI, have already received test racks, each equipped with 72 GPUs and priced between $7 million and $8 million. Compared to the Blackwell system, the new design eliminates cables and incorporates robotic assembly, significantly improving production efficiency.
NVIDIA's next-generation Vera Rubin server system is accelerating toward production and is expected to achieve large-scale manufacturing next year, drawing keen anticipation from AI developers, cloud service providers, and Wall Street investors.
On July 22, according to The Information, several NVIDIA executives recently hosted on-site media visits, providing detailed disclosures on the production progress and technical specifications of the Vera Rubin system.
Andrew Bell, Senior Vice President at NVIDIA, stated that the manufacturing process for Vera Rubin is significantly more efficient than that of the previous-generation Blackwell product and expressed optimism about the smooth progress of this rollout.
The production ramp-up pace of the Vera Rubin racks will be pivotal to NVIDIA’s future revenue. At maximum capacity, the potential revenue scale could far exceed NVIDIA’s current quarterly performance, further intensifying market attention on its commercial prospects.
Dozens of customers have already received test racks
According to The Information, dozens of customers—including CoreWeave, Microsoft, OpenAI, Anthropic, and SpaceX AI—have already received initial shipments of a limited number of Vera Rubin test racks, each equipped with 72 GPUs.
In terms of pricing, one buyer revealed that each Vera Rubin rack is priced between $7 million and $8 million, compared to approximately $5 million for the current flagship Grace Blackwell 300 rack—a significant increase.
During their visit to NVIDIA’s headquarters in Santa Clara, journalists also toured a semi-confidential testing facility where they observed around 30 server racks in operation, many of which were Vera Rubin systems. NVIDIA executives noted that OpenAI is already using some of these racks.
Visually, a single Vera Rubin rack is roughly the size of a tall filing cabinet but weighs as much as 4,000 pounds—equivalent to a pickup truck.
Production has progressed more smoothly than with Blackwell, and the cable-free design has improved assembly efficiency
Compared to the numerous setbacks encountered during the launch of Blackwell, the manufacturing progress of Vera Rubin has been significantly smoother.
In an interview, NVIDIA Senior Vice President Andrew Bell candidly admitted that Blackwell initially "nearly broke everything"—with issues spanning hardware, software, diagnostic systems, and manufacturing, ultimately requiring a complete restart.
Bell noted that Vera Rubin incorporates significant design improvements: racks are nearly cable-free and rely on robotic systems to handle most assembly tasks, which will substantially reduce assembly complexity and thereby enhance production efficiency and yield.
He expressed cautious optimism that Vera Rubin will avoid the technical bottlenecks that plagued Blackwell.
Maximum daily production capacity reaches 1,000 racks, with potential quarterly revenue exceeding $630 billion
Regarding production planning, Bell revealed that NVIDIA’s current network of more than ten manufacturing partners will eventually be capable of producing up to 1,000 Vera Rubin racks per day.
Based on this capacity, potential quarterly revenue would amount to at least $630 billion, compared to NVIDIA’s actual revenue of $82 billion for the quarter ending in April this year.
However, achieving maximum capacity does not guarantee delivery at that rate; actual shipments will still depend on customer demand and installation timelines.
NVIDIA Vice President Ian Buck stated that the company’s GPU allocation decisions are directly tied to whether customers are physically ready to install servers, and jokingly suggested that the final allocation authority rests with CEO Jensen Huang.
Editor/lambor