NVIDIA Vera Rubin Rack Priced at $7.8 Million, Double That of Blackwell! Not Just Memory, Nearly All Components Are Seeing Price Hikes

The price of NVIDIA's next-generation AI server racks is rising rapidly.

According to a report released on May 22 by Morgan Stanley analyst Howard Kao, NVIDIA's upcoming Vera Rubin (VR200) rack costs approximately $7.8 million when procured from ODMs (Original Design Manufacturers), while the current GB300 Blackwell rack costs less than $4 million—meaning the price per rack has nearly doubled between generations.

Notably, NVIDIA's stock closed down nearly 2% after its earnings release on the same day, but memory-related stocks surged 6%-10%. The reason lies in the logic revealed by this report: the beneficiaries of the price hikes are not limited to NVIDIA itself.

Memory: From "Supporting Role" to "Lead Actor"

Memory is the most significant driver of this round of price increases.

Analysts pointed out that memory prices have risen substantially since NVIDIA launched the GB200 NVL72. Under the old pricing structure, memory accounted for only 5%-10% of the bill of materials (BOM) cost for the GB200 NVL72 rack; however, for the VR200, this proportion has soared to 25%-30%, representing a cost increase of up to 435%.

The sharp rise in the proportion of memory costs has directly compressed the GPU's share of the total rack cost—the GPU's share dropped from about 65% in the GB200 to about 51% in the VR200.

Not Just Memory, Nearly All Components Are Rising

The firm's supply chain survey indicates that the price increases are comprehensive:

PCB (Printed Circuit Board): Cost increased by 233%

PCB content jumped from approximately $35,000 for the GB300 to about $117,000. This is because Rubin introduced new modules (such as ConnectX modules and midplane PCBs), while both the number of board layers and material grades have improved. For example, the compute board upgraded from a 22-layer HDI PCB in the GB300 to 26 layers, with the material grade rising from M7 to M8; the switch tray PCB increased from 24 layers to 32 layers. Additionally, a 44-layer midplane PCB was added to the compute tray, which was not present in the GB300.

MLCC (Multilayer Ceramic Capacitors): Cost increased by 182%

MLCC content per VR200 rack is approximately $4,300, compared to only about $1,500 for the GB300. The newly introduced BlueField and ConnectX modules also brought additional MLCC demand. Analysts believe this is the direct reason for the unusually strong demand for high-end AI server MLCCs currently, with ODM manufacturers rushing to stockpile inventory—mass production of Rubin racks is expected to begin in the second half of 2026.

ABF Substrate: Cost increased by 82%

The number of NVLink and ConnectX chips in the VR200 is double that of the Blackwell system, leading to increased substrate usage. The unit price of ABF substrates for Rubin GPUs is approximately $200 per chip, a 100% increase from Blackwell's approximately $100.

Power supply costs increased by 32%; cooling materials also increased by 12%.

Hyperscale Cloud Providers May Bypass NVIDIA to Buy Memory Directly

Analysts proposed a key variable: the procurement method for SOCAMM (Small Outline Compression Attached Memory Module). If hyperscale cloud providers purchase memory modules directly, the rack price could drop to approximately $6.7 million.

In the base case scenario, NVIDIA is responsible for procuring SOCAMM and reselling it with a 70% gross margin, resulting in a rack price of approximately $7.8 million. However, if hyperscale cloud providers (such as Microsoft, Google, Amazon, etc.) choose to procure SOCAMM directly, bypassing NVIDIA's markup, the rack price will drop to approximately $6.7 million.

This variable directly impacts NVIDIA's memory business revenue and is a key point for the market to continue tracking.

ODM Manufacturers: Margins Declining, But Absolute Profits Rising

The market previously generally expected that ODM manufacturers' value-add space on Rubin racks would shrink due to the trend toward "standardization" in compute tray design. However, analysts reached the opposite conclusion.

The report estimates that the ODM value-add portion will rise by 35%-40%, increasing from approximately $108,000 per rack for the GB300 to about $149,600 per rack for the VR200. This assessment aligns with statements made by Wistron management during their Q4 2025 earnings conference call—Wistron explicitly stated that the USD value-add for Rubin rack ODMs would increase.

However, gross margins are indeed declining: the ODM gross margin for the GB300 was approximately 2.7%, dropping to about 1.9% for the VR200. Regarding this, analyst Kao stated, "Investors should focus on the growth in absolute USD profits rather than the decline in margins."

Consignment Model Quietly Spreading, Long-Term Risks Cannot Be Ignored

Analysts also mentioned a trend worth noting: an increasing number of ODM manufacturers are beginning to discuss the consignment business model.

Hon Hai was the first to mention this model, during its Q4 2025 earnings conference call. Quanta subsequently stated in its Q1 2026 earnings conference call that it expects some projects to shift to the consignment model in the second half of 2026.

The so-called consignment model refers to customers (cloud providers) purchasing core components themselves, with ODMs only responsible for assembly. This alleviates working capital pressure for ODMs but also compresses their revenue scale.

Analysts pointed out that it is currently unclear what proportion of projects will shift to consignment, but if the aforementioned "absolute profit growth" fails to materialize, this trend poses long-term concerns.

Power Supply and Liquid Cooling: The Next Upgrade Direction

Supply chain surveys show that the Vera Rubin platform comes standard with 110kW power modules, but at least one US cloud service provider has adopted HVDC (High Voltage Direct Current) independent power racks on the Vera Rubin platform. Analysts expect that 800V DC power will be widely adopted on the Rubin Ultra platform launching in the second half of 2027. Delta Electronics has already collaborated with at least three US cloud service providers to advance the implementation of HVDC platforms, with initial deployments expected to begin in the second half of 2026.

Regarding cooling, the Vera Rubin rack will adopt a full liquid-cooling design (fanless), with the total value of liquid cooling components per rack being approximately $72,080.