Goldman Sachs estimates memory chips will make up about 62% of the material costs on Nvidia's upcoming Vera Rubin superchip, up from roughly 53% on the current GB300 generation. The dollar figures behind that shift concentrate cost and supply leverage in the hands of memory makers like SK Hynix, Samsung, and Micron.
Memory chips will account for roughly 62% of the material costs on Nvidia's next-generation Vera Rubin superchip, according to Goldman Sachs — up from about 53% on the current GB300 systems, a nine-percentage-point jump. The dollar figures show the scale of the shift: Goldman estimates the bill of materials for a full Vera Rubin NVL72 rack system at approximately $7.8 million, with memory costs alone approaching $2 million. Some analyses peg the year-on-year increase in memory spending at over 435%.
What is driving the memory cost surge
The Vera Rubin platform pairs two new chips. Nvidia's Rubin GPU packs up to 288 GB of HBM4 memory, capable of delivering 22 TB/s of bandwidth, on a 336-billion-transistor design built on TSMC's 3-nanometer process. The companion Vera processor houses 88 Olympus cores alongside up to 1.5 TB of LPDDR5X memory, linked to the GPU through a memory-coherent NVLink-C2C connection running at 1.2 TB/s.
Memory makers gain leverage
Because memory now represents 62% of a system's material cost, the fortunes of memory manufacturers become directly tied to the AI infrastructure buildout. Samsung, SK Hynix, and Micron are the three dominant producers of HBM and LPDDR chips, and that concentration puts them in an unusually strong position. SK Hynix in particular has been the leading supplier of HBM chips to Nvidia, and the move to HBM4 brings both an opportunity and a manufacturing challenge, since the advanced packaging capacity it requires is in limited global supply. That scarcity creates a natural constraint on how quickly Nvidia can scale production.
The cost pressure is already pushing Nvidia and its partners toward possible fixes. These could include adjusted memory capacities on certain SKUs, more aggressive binning of memory chips, or architectural changes that improve memory utilization so each unit of compute needs less raw capacity.
Nvidia announced the Vera Rubin platform in 2025, with production ramping set to begin in 2026. The system is designed for high-performance agentic AI workloads.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
