NVIDIA Cuts Vera CPU Memory Capacity Amid Shortage

According to global research firm TrendForce , NVIDIA has decided to reduce the SOCAMM memory capacity used in its next-generation Vera Rubin Superchip module by half, in response to expected DRAM supply constraints projected for early 2027, rather than cutting its overall memory demand outlook.

TrendForce noted that this adjustment reflects a supply-driven architectural compromise rather than a downward revision of total storage requirements. In practice, NVIDIA is lowering per-module capacity while increasing the number of modules shipped, a strategy aimed at sustaining platform scalability, strengthening market share, and addressing the persistent structural shortage in LPDDR5X supply.

Although major memory manufacturers—including Samsung Electronics, SK hynix, and Micron Technology—have outlined expansion plans toward 2027, projected bit output growth still lags behind anticipated demand from the AI ecosystem. Current preliminary supply allocations of LPDRAM from these three vendors are estimated to meet only around 60% of NVIDIA’s requirements, with limited flexibility for rapid scaling.

To mitigate this imbalance, NVIDIA has reportedly adjusted the SOCAMM configuration in its Vera Rubin Superchip platform to support broader deployment of Vera CPU-based systems and to reduce future supply risk amid tightening upstream availability.

NVIDIA’s next-generation AI server lineup is expected to include both integrated Vera Rubin rack-scale systems and standalone Vera CPU rack solutions. The company is also actively promoting the standalone CPU configuration to expand its presence in inference-oriented AI workloads, which is becoming a key driver of LPDRAM demand growth.

LPDDR5X (LPDRAM), known for its high bandwidth efficiency and lower power consumption, is increasingly extending beyond smartphones into AI servers and automotive applications. In addition, more ASIC designers outside NVIDIA are evaluating its adoption, signaling a broader diversification of use cases.

TrendForce further forecasts that between 2028 and 2030, the global AI server ecosystem could become the largest single demand driver for LPDRAM, surpassing smartphones and reshaping long-term memory supply dynamics across the semiconductor industry.