Humanoid robots to drive 20-fold increase in DRAM demand by 2030

The semiconductor industry is bracing for a massive surge in memory demand driven by the rise of humanoid robots and 'physical AI'. According to Counterpoint Research, global DRAM demand for humanoid robot processing units is projected to grow 20-fold by 2030, rising from approximately 8,300 terabits this year to 172,000 terabits.

As AI models evolve toward 'end-to-end' architectures that require real-time processing of data from cameras, LiDAR, and tactile sensors, the memory requirements per robot are also increasing. The average DRAM capacity per unit is expected to double from 19GB to 39GB by 2030. Micron CEO Sanjay Mehrotra noted that humanoid robots will likely require memory and storage levels similar to Level 4 autonomous vehicles, exceeding 200GB of DRAM and several terabytes of SSD storage.

Major memory manufacturers including Samsung Electronics, SK hynix, and Micron are positioning themselves to lead this next market cycle following the HBM boom. Key focus areas include low-power DRAM (LPDDR) for battery efficiency and high-capacity NAND flash. However, Chinese manufacturer CXMT is emerging as a potential competitor by advancing its next-generation LPDDR6 production capabilities.