Nvidia's Vera Rubin rack is priced at $7.8 million, twice that of Blackwell! It's not just memory—almost all components are getting more expensive.

The price of NVIDIA’s next-generation AI server racks is rapidly soaring.

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

Notably, after NVIDIA’s earnings announcement that day, its stock price closed down nearly 2%, but memory-related stocks surged by 6%-10%. The reason is the logic revealed by this report: the beneficiaries of the price hikes are not just NVIDIA itself.

Memory: From “Supporting Role” to “Leading Role”

Memory is the most aggressive driver in this round of price increases.

The analyst noted that since NVIDIA launched GB200 NVL72, memory prices have surged. In the old pricing system, memory only accounted for 5%-10% of the bill of materials (BOM) in the GB200 NVL72 rack; but in the VR200, memory’s proportion has skyrocketed to 25%-30%, with costs rising as much as 435%.

The surge in memory’s share directly compresses the proportion of GPUs in the rack’s overall cost—GPU’s share dropped from about 65% in GB200 to about 51% in VR200.

Not Just Memory—Almost Every Component is Increasing

The bank’s supply chain survey shows that price increases are widespread:

PCB (Printed Circuit Board): Cost up 233%

PCB content jumped from about $35,000 in GB300 to about $117,000. The reason is Rubin’s introduction of new modules (such as ConnectX modules and mid-board PCBs), as well as increased PCB layers and material grade. For example, the computation board upgraded from GB300’s 22-layer HDI PCB to 26 layers, material grade from M7 to M8; switch tray PCB increased from 24 to 32 layers. Also, a new 44-layer mid-board PCB was added in the compute tray, which didn’t exist in GB300.

MLCC (Multilayer Ceramic Capacitor): Cost up 182%

MLCC content per VR200 rack is about $4,300, compared to about $1,500 for GB300. The newly introduced BlueField and ConnectX modules also add to MLCC demand. The analyst believes this is the direct reason for the current strong demand for high-end AI server MLCCs, and ODMs compete to stockpile. Rubin racks are expected to enter mass production in the second half of 2026.

ABF Substrate: Cost up 82%

The number of NVLink and ConnectX chips in VR200 is twice that of the Blackwell system, and the substrate usage increases accordingly. The ABF substrate unit price for Rubin GPUs is about $200 apiece, compared to about $100 for Blackwell, an increase of 100%.

Power Supply cost up 32%; cooling materials up 12%.

Hyperscale Cloud Providers May Purchase Memory Directly

The analyst identified a key variable: the purchasing method for SOCAMM (Small Outline Compression Attached Memory Module). If hyperscale cloud providers purchase memory modules directly, rack price may drop to about $6.7 million.

In the base scenario, NVIDIA is responsible for purchasing SOCAMM and resells it with a gross margin of 70%, putting rack prices at about $7.8 million. But if hyperscale cloud providers (such as Microsoft, Google, Amazon, etc.) choose to directly purchase SOCAMM, bypassing NVIDIA’s markup, rack prices will decrease to around $6.7 million.

This variable directly affects NVIDIA’s memory business revenue and is a key area the market needs to keep tracking.

ODM Manufacturers: Profit Margins Down, But Absolute Profits Up

The prior market expectation was that, as compute tray design tends toward "standardization," ODMs would have less value-added space in Rubin racks. But the analyst concluded the opposite.

The report estimates that ODM value-added portion will rise 35%-40%, from about $108,000 per rack with GB300 to about $149,600 per rack with VR200. This aligns with Wistron’s management statement in the Q4 2025 earnings call—Wistron clearly stated that ODM’s dollar value-added on Rubin racks will increase.

But profit margins are indeed declining: ODM gross margin for GB300 was about 2.7%, while VR200 drops to about 1.9%. Analyst Kao commented, “Investors should focus on the growth in absolute dollar profit, rather than the decline in margin.”

Consignment Model Quietly Spreading, Long-Term Risks Should Not Be Ignored

The analyst also mentioned a noteworthy trend: more ODMs are now talking about consignment business models.

Hon Hai (Foxconn) was the first to mention this model, during its Q4 2025 earnings call. Quanta later indicated during its Q1 2026 earnings call that some projects are expected to shift to the consignment model in the second half of 2026.

The consignment model means customers (cloud providers) purchase the core components themselves, and ODMs only handle assembly, thus relieving their working capital pressure but also compressing revenue scale for ODMs.

The analyst pointed out it’s unclear how large a proportion of projects will turn to consignment, but if the promised “absolute profit growth” is not realized, this trend poses long-term concerns.

Power and Liquid Cooling: The Next Upgrade Direction

Supply chain survey shows that the Vera Rubin platform comes standard with a 110kW power module, but at least one US cloud service provider has adopted HVDC (High Voltage DC) independent power rack on the Vera Rubin platform. The analyst expects that 800V DC power will be widely adopted in the Rubin Ultra platform launching in the second half of 2027. Delta has partnered with at least three US cloud service providers to advance HVDC platform deployment, with initial rollout slated for the second half of 2026.

On cooling, Vera Rubin racks will use all-liquid cooling design (no fans), with total value of liquid cooling components per rack at about $72,080.

Risk Warning and DisclaimerThe market involves risk, and investments should be cautious. This article does not constitute personal investment advice, nor does it consider users’ particular investment goals, financial position, or needs. Users should consider whether opinions, views or conclusions in this article suit their circumstances. Investment made accordingly is at their own risk.