Humanoid Robot Wave Could Drive 20x DRAM Demand Growth by 2030
nashnova research
Counterpoint forecasts humanoid-robot DRAM demand will leap from 8,300 terabits in 2025 to 172,000 terabits by 2030 — roughly a 20× increase; Goldman Sachs has simultaneously raised its 2030 shipment forecast to 890,000 units, signalling a second demand shock for the memory industry after AI data centres.
Where does the 20× number come from?
Counterpoint Research projects humanoid-robot DRAM demand will surge from roughly 8,300 terabits in 2025 to 172,000 terabits by 2030 — an approximately 20-fold increase.
Average DRAM per robot is expected to double over the same period, from about 19 GB to 39 GB. This means → the explosive total comes less from each robot getting bigger memory and more from the sheer number of robots shipping.
The forecast, first reported by South Korea's *Chosun Ilbo* citing the Counterpoint report, positions humanoid robots as the next major memory-demand driver after AI data centres.
Why does a single robot need so much memory?
A humanoid robot must process images, video and audio simultaneously, while relying on a "world model" — an AI system that predicts how the surrounding environment will change — to make real-time decisions.
In plain terms = unlike a phone that runs one app at a time, a humanoid robot "sees, hears and thinks" all at once, each stream consuming data in real time. That pushes memory and processing demands far beyond consumer electronics.
Micron CEO Sanjay Mehrotra has noted that L4-and-above autonomous vehicles typically require more than 200 GB of memory and several TB of storage, adding that "humanoid robots may need similar levels." This reflects how major memory makers have already folded the robotics market into their strategic outlook.
How high did Goldman Sachs raise its shipment forecast?
Goldman Sachs analyst Eric Sheridan sharply revised his projections in late August: 2026 base-case shipments rose from 51,000 to 75,000 units; 2030 from 256,000 to 890,000 units; 2035 from 1.4 million to 6.5 million units.
Sheridan wrote to clients: "The global market is expected to reach 6.5 million units by 2035, representing a roughly $138 billion market opportunity."
This means → if this shipment curve materialises, humanoid robots will shift from a lab-stage concept to a hundred-billion-dollar hardware category, with memory as only the first supply-chain link they pull on.
What does this mean for the memory industry?
Today, hyperscale AI data centres are already bidding up prices for HBM — high-bandwidth memory, a stacked DRAM architecture purpose-built for AI acceleration — squeezing procurement costs for consumer-electronics firms like Apple.
If humanoid-robot shipments track Goldman's trajectory, incremental memory demand will extend from data centres to physical AI endpoints. In plain terms = the big memory consumers expand from "servers in the cloud" to "robots on the ground."
Whether the DRAM supply-demand balance can sustain this data-centre + robotics twin engine will be the critical validation point for the market over the coming years.
市场有风险，内容仅供研究参考，不构成投资建议。