Humanoid robot boom set to drive 20-fold surge in DRAM demand by 2030, opening new door for Samsung, SK hynix

AI-powered humanoid robots are emerging as a major new source of demand for memory chips. Following the rapid growth of the high-bandwidth memory (HBM) market driven by AI data center expansion, analysts now expect the wider rollout of robots to significantly boost demand for low-power DRAM and NAND flash as well.

Global DRAM demand from humanoid robot computing units is forecast to surge from roughly 8,300 terabits (Tbit) this year to about 172,000 Tbit by 2030, according to market research firm Counterpoint Research. That represents a more than 20-fold increase in just four years.

The amount of memory packed into each robot is also set to grow. Counterpoint projects that the average DRAM capacity per humanoid will more than double, rising from 19 gigabytes (GB) this year to 39 GB by 2030.

Humanoid shipments surging — and so is memory per unit

The primary driver of rising memory demand is the simultaneous increase in both the number of robots being deployed and the memory capacity installed in each unit.

Global humanoid robot shipments exceeded 22,000 units in the first half of this year, up roughly 300 percent from the same period last year, according to Counterpoint. Full-year shipments are expected to surpass 50,000 units, with the global total projected to reach around 256,000 units by 2030.

Humanoid robots continuously gather environmental data through cameras, LiDAR sensors and tactile sensors. They must then recognize people and objects, calculate movement paths, and make real-time decisions about joint and hand movements — all based on that incoming data stream.

Multimodal AI systems that simultaneously process images, video and audio are being rapidly adopted, along with vision-language-action (VLA) models that directly link AI perception to physical movement. The addition of "world models" — which predict changes in the surrounding environment — is further increasing the volume of data that must be processed at once.

Higher computing chip performance alone cannot solve this challenge. Expanding DRAM capacity and bandwidth is equally essential to quickly load and temporarily store AI models and sensor data.

Servers need HBM; robots need low-power DRAM — and battery life matters too

The memory requirements for humanoid robots differ significantly from those of AI data centers.

Data centers are rapidly adopting HBM for the high bandwidth needed to run large-scale AI workloads. Robots run on batteries, making power consumption a far more critical constraint. Counterpoint expects current humanoid robots to rely primarily on LPDDR-series low-power DRAM and high-capacity NAND, rather than HBM.

As robots take on more autonomous decision-making, the share of "edge AI" — on-device processing rather than sending all data to external servers — is likely to grow. That would require robots to carry even more onboard memory and storage.

Micron also views robotics as a next-generation memory market. In its earnings presentation last June, the company projected that the memory in a single humanoid robot would be roughly 10 times greater than that in an average Level 2+ self-driving car. It cited the need to handle real-time perception, reasoning and control simultaneously as the key driver of demand for high-bandwidth, low-power memory and storage.

Micron CEO Sanjay Mehrotra has said he expects the computing platform of humanoid robots to eventually reach a level comparable to high-performance Level 4 autonomous vehicles. Level 4 and above self-driving cars typically require more than 200 GB of memory and storage measured in terabytes, and humanoids may demand similar specifications over the long term.

The next big thing after HBM? Samsung and SK hynix eye the opportunity

The expansion of the humanoid market could open a significant new opportunity for domestic memory chipmakers, including Samsung Electronics and SK hynix.

Growing investment in AI data centers has already shifted the center of gravity in the memory market toward HBM. In the second quarter of this year, SK hynix held a 50 percent share of the global HBM market, while Samsung Electronics accounted for 33 percent, according to Counterpoint.

If the robotics market takes off in earnest, demand could broaden from HBM to standard DRAM, low-power DRAM and NAND. Should humanoid robots scale into a mass-market product category like smartphones or PCs, the industry would see rising memory content per unit and growing shipment volumes at the same time.

Memory chipmakers are already expanding product development for automotive and robotics applications, which they view as next-generation growth markets. Micron expects sustained long-term demand for low-power memory and high-capacity storage to support real-time AI processing in robots and self-driving vehicles.