NVIDIA’s upcoming Vera Rubin NVL72 rack is expected to carry a much higher bill of materials as memory prices rise sharply across the AI hardware market. According to Morgan Stanley Research estimates cited in the reference, the rack’s total cost could reach about $7.8 million, with memory alone making up around 26 percent of the full system cost.
The Vera Rubin platform is already in production, with first shipments expected in the third quarter of 2026 and volume ramp planned for the fourth quarter. The rack, also called Oberon, is designed for large AI infrastructure and uses 72 Rubin GPUs and 36 Vera CPUs.
HBM4 and LPDDR5X costs are becoming a major part of NVIDIA’s next AI platform
Each Rubin GPU is expected to include 288GB of HBM4 memory, while each Vera CPU carries 1.5TB of LPDDR5X memory. Across a full NVL72 rack, that adds up to 20.7TB of HBM4 and 54TB of LPDDR5X. This explains why memory has become one of the largest cost drivers in the platform.
The estimated memory cost has reportedly jumped from $373,939 on Grace Blackwell to more than $2 million on Vera Rubin. That is a 435 percent increase, covering both HBM4 and LPDDR5X. GPU costs are also expected to rise, with the 72 Rubin GPUs estimated at nearly $4 million in total, or about $55,000 per GPU.
| Component or area | Estimated detail |
|---|---|
| Platform | NVIDIA Vera Rubin NVL72 |
| Rack name | Oberon |
| GPUs | 72 Rubin GPUs |
| CPUs | 36 Vera CPUs |
| HBM4 memory | 20.7TB per rack |
| LPDDR5X memory | 54TB per rack |
| Estimated memory cost | More than $2 million |
| Total estimated rack cost | About $7.8 million |
| Shipment target | Q3 2026, with volume ramp in Q4 |
The Vera CPUs are estimated at $180,000 for the full rack, or about $5,000 per chip. Other costs include NVLink switches, networking chips, cooling, power supplies, PCBs, ABF substrates, MLCCs, and other supporting parts. The PCB cost is also expected to rise sharply, increasing from $35,100 on Blackwell to $116,730 on Rubin.
This is a separate NVIDIA story from the earlier China H200 revenue update and the gaming revenue reporting change. This report focuses on the cost of NVIDIA’s next major AI rack and shows how expensive memory has become as demand for AI systems keeps rising. For NVIDIA, Vera Rubin may deliver a major performance step, but it also shows that the next phase of AI infrastructure will come with much higher hardware costs.
Discussion (0)
Be the first to comment.