Counterpoint Research Projects Humanoid Robot DRAM Demand Will Grow 20x by 2030 as Memory Per Robot Doubles to 39GB
Counterpoint Research expects DRAM demand from humanoid robots to surge 20x by 2030, from around 8,300 terabits in 2026 to 172,000 terabits. The firm also expects memory per robot to more than double, from 19GB today to 39GB by 2030.
Context
As humanoid robots adopt onboard vision language action models and world models under strict battery limits, physical AI is emerging as a new growth segment for low power LPDDR memory and high density NAND flash.
Global humanoid robot shipments surpassed 22,000 units in the first half of 2026, up roughly 300 percent year over year, and are projected to exceed 50,000 units for the full year. Memory makers Samsung Electronics, SK hynix, and Micron have increasingly targeted physical AI platforms such as Nvidia's Jetson Thor as their next growth market after data center accelerators.
Deep dive
Market research firm Counterpoint Research forecasts that global DRAM demand for humanoid robot computing units will expand more than 20x over the next four years, rising from approximately 8,300 terabits in 2026 to 172,000 terabits in 2030. The surge is driven by both rising annual robot shipments and a projected doubling in average onboard DRAM capacity per humanoid from 19 gigabytes today to 39 gigabytes by 2030.
Humanoid robots must continuously process multimodal streams from cameras, LiDAR, audio arrays, and tactile sensors while running vision language action policies and predictive world models on the device. Because humanoids operate on internal batteries, Counterpoint notes that robot builders prioritize power efficient low power DRAM (LPDDR) and high capacity NAND flash over the power hungry High Bandwidth Memory (HBM) stacks used in data center servers.
Semiconductor analyst Zephyr pointed out on X that 172,000 terabits equals 21.5 petabytes of DRAM, which implies roughly 550,000 humanoid robots produced annually at 39 gigabytes per unit. While that volume represents only about 0.04 percent of total 2026 global DRAM output of roughly 48 exabytes, Samsung Electronics and SK hynix have already begun bundling custom LPDDR memory, image sensors, and packaging for robotics platforms including Nvidia's Jetson Thor.
What to expect next
While humanoid robotics will account for only a tiny fraction of global DRAM bit volume through 2030 compared with hyperscale data centers, edge AI devices require specialized low power memory packages with high reliability and wide temperature tolerances. As autonomous vehicles and industrial humanoids converge on 40GB to 200GB onboard memory footprints, automotive and robotics LPDDR will become a lucrative margin tier for memory suppliers.