NVIDIA Rubin Latest Update: What Does the Delivery of Test Cabinets with a Daily Capacity of 1,000 Mean?

TL;DR
· According to The Information, Vera Rubin test cabinets have been delivered to dozens of customers, with each cabinet priced at approximately $7 million to $8 million.
· Each cabinet contains 72 Rubin GPUs and 36 Vera CPUs, with manufacturing partners aiming for a maximum production capacity of 1,000 cabinets per day.
· The 1,000 cabinets represent the production capacity limit, not actual orders, as customer data centers' power supply, liquid cooling, and deployment capabilities remain constraints.

According to The Information, NVIDIA's next-generation Vera Rubin server system has delivered a small number of test cabinets to dozens of customers, with each cabinet priced at approximately $7 million to $8 million, and manufacturing partners ultimately aiming for a maximum production capacity of 1,000 cabinets per day.
> NVIDIA Vera Rubin NVL144 CPX Rack and Tray
These figures clearly indicate the focus of NVIDIA's next round of AI hardware upgrades: it is not just selling stronger GPUs, but rather more expensive and complex complete cabinet server systems. NVIDIA's official documentation shows that the Vera Rubin NVL72 is a rack-level system that includes 72 Rubin GPUs and 36 Vera CPUs. Compared to the current flagship Grace Blackwell 300 rack priced at about $5 million, the price of the Rubin cabinet has further increased.
CoreWeave announced in June that it has completed the bring-up and validation of the Vera Rubin NVL72. NVIDIA has also recently stated that Vera Rubin has entered the full-capacity ramp-up phase, with racks already operating at partners such as CoreWeave, Google Cloud, Microsoft Azure, and OCI. This means that Rubin is no longer just a theoretical product, but the $7 million to $8 million cabinet price, dozens of test customers, and the daily target capacity of 1,000 cabinets are still not part of NVIDIA's official revenue guidance.
A cabinet priced at up to $8 million buys a complete computing unit =======================
The price of Rubin is not merely due to chip price increases, but rather a continuation of NVIDIA's complete cabinet system approach.
After Blackwell, NVIDIA's core products sold to major customers are increasingly not isolated GPUs, but systems that package GPUs, CPUs, networking, cooling, power, software, and rack-level interconnects together. Customers are purchasing a computing unit that can be integrated into data center planning, rather than assembling components from scratch themselves.
This is also the reason why the price of a single cabinet can reach $7 million to $8 million. NVIDIA's technical documentation indicates that the Vera Rubin NVL72 weighs about 4,000 pounds, close to the weight of a pickup truck. For cloud vendors and AI companies, purchasing Rubin is not just about ordering chips, but also requires simultaneous preparation for data center load-bearing, power supply, liquid cooling, network connections, and deployment debugging.
The Information quoted NVIDIA's Vice President of High-Performance Computing Ian Buck, stating that the company hopes to sell to all customers, but actual distribution will be linked to whether customers have the physical installation capabilities and can bring the servers online. Those who can actually connect these cabinets to data centers are more likely to receive more shipments.
1,000 cabinets per day is impressive, but not actual orders ==================
The most market-inspiring figure is "1,000 cabinets per day."
The Information quoted NVIDIA's Senior Vice President of Hardware Engineering Andrew Bell, stating that the more than ten manufacturing partners collaborating with NVIDIA to produce Rubin racks will ultimately be able to produce up to 1,000 racks per day. Based on a price of $7 million to $8 million per cabinet, this corresponds to a significant potential revenue space.
According to this production capacity estimate, if 1,000 cabinets are produced daily, theoretically, it could correspond to at least $630 billion in revenue for a quarter. In comparison, NVIDIA's revenue for the quarter ending April 26, 2026, was $81.6 billion.
However, this figure can only be understood as a theoretical estimate based on "production capacity multiplied by unit price." It is not revenue guidance provided by NVIDIA's management, nor does it represent confirmed orders, and it does not mean that manufacturing partners will operate at full capacity for an extended period. Actual shipments will depend on customer orders, supply chains, data center construction, acceptance rhythms, and revenue recognition rules.
For investors, 1,000 cabinets per day is more like NVIDIA showcasing the upper limit of next-generation system manufacturing, rather than a short-term revenue figure that can be directly incorporated into profit models. What truly changes the revenue rhythm is whether these high-priced racks can be delivered consistently, installed in customer facilities, and begin operations.
Lessons from Blackwell: Rubin Must First Solve Manufacturing Challenges =============================
The first thing Rubin needs to prove is not just performance, but whether it can be manufactured, assembled, and operated more smoothly than Blackwell.
Bell recalled the issues encountered with Blackwell racks last year, saying, "We almost messed everything up." Problems arose in hardware, software, diagnostic systems, and manufacturing processes. After NVIDIA entered a rack scale that it had never reached before, it almost had to rebuild the system from scratch.
This experience explains why the manufacturing details of Rubin are emphasized by executives. The new generation of racks does not mean "no cables." NVIDIA's technical blog shows that the NVLink spine's rear has pre-integrated cable cartridges containing about 5,000 copper cables. The change is that more cables and components are made modular and pre-integrated, reducing manual wiring work on-site and on the production line, improving consistency and assembly speed.
This is crucial for NVIDIA. AI customers buy Rubin not just for the theoretical computing power, but for stable delivery and quick deployment. If the new generation of racks encounters hardware, software, diagnostic, or manufacturing issues like Blackwell again, the pace of customer capital expenditures will slow down, and NVIDIA's shipments and revenue recognition will also be affected.
NVIDIA is also locking customers into a complete data center system =============================
Another layer of meaning for Rubin is that NVIDIA is pushing its position from a GPU supplier to an AI data center system supplier.
In communications at headquarters, NVIDIA executives emphasized that the company is not just selling GPUs, but also selling CPUs, network switches, cables, storage, and server cooling technologies, among other non-GPU components. This is not a simple expansion of product categories, but a way to maintain a critical role in data centers while cloud vendors and large model companies seek alternative AI chips.
Competition is becoming more complex. Google's public information shows that the eighth-generation TPU is divided into training TPU 8t and inference TPU 8i, with the inference chip and training chip routes diverging. Cloud vendors and large model companies are also looking for cheaper and more controllable computing solutions. If customers partially use non-NVIDIA AI chips in the future, NVIDIA still hopes to sell them networking, interconnect, cooling, and other server components.
This is also why the price of a single Rubin cabinet is attracting attention. It is not just a quote for a chip, but a new pricing benchmark formed after NVIDIA packages critical elements of AI data centers. The more customers purchase complete cabinet systems, the more NVIDIA can gain additional revenue from networking, CPUs, cooling, and system integration.
High-priced racks must truly operate, but face data center constraints ====================
Rubin has entered the capacity ramp-up phase, but large-scale commercial deployment may still extend to 2027. Customers need to complete power supply, liquid cooling, rack installation, network debugging, and software adaptation; any delay in any of these links will affect the actual usage rhythm of Rubin.
Supply chain constraints are not limited to advanced chip manufacturing. NVIDIA executives mentioned that the company has teams continuously tracking the supply chain and establishing contacts with companies involved in natural resources such as aluminum and indium phosphide. These materials are related to servers and network devices, indicating that after the complete cabinet system is ramped up, bottlenecks may arise in more traditional industrial links, not just in wafers and packaging.
The goal of 1,000 cabinets per day provides the market with a huge imagination space, but what Rubin truly needs to navigate are a series of real constraints: whether manufacturing partners can assemble stably, whether customers can receive and install, whether data centers can provide sufficient power and cooling, and whether non-GPU components can ramp up simultaneously.
NVIDIA has pushed the next generation of AI hardware business towards higher prices and greater system complexity. What determines the quality of Rubin is not just the computing power of 72 GPUs per cabinet, but whether these multi-million dollar racks can enter data centers as planned and truly operate.

Disclaimer: This content is provided for general branding and informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online events, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets or to use any services. Crypto assets are highly volatile and may result in loss. WEEX services and online events may not be available in all regions and are subject to applicable laws, regulations, and eligibility requirements. You are responsible for ensuring that your use of WEEX services complies with local laws and for carefully assessing the risks before participating in any crypto-related activities.