A single reply from Elon Musk has pushed an unusual idea into the spotlight: humanoid robots could become one of the biggest buyers of memory chips. Musk responded on X on October 26 to an analyst at Citrini Research, saying the amount of memory humanoid robots will need will exceed everything else on Earth, according to a November 2. Shares of Micron Technology rose about 2.5% after the remark, partly recovering after falling about 6% on October 22 during another AI-related chip sell-off.

Analysts see a real runway. Counterpoint Research expects DRAM demand from humanoids to climb more than 20 times between 2026 and 2030, with the memory in each robot rising from about 19GB today to 39GB by the end of the decade. Musk thinks even that may be too low. In early October he said Tesla had halved the memory on its AI5 chip to 72GB and cut the AI6 chip by a third to 144GB, calling it the only way to get enough volume for Optimus production, and adding that the cuts should have a negligible effect on performance because memory bandwidth matters more than total capacity.

The robot-memory story rests on supply. Musk’s post listed the robot chip’s memory as LP5, a low-power type of DRAM that also goes into phones and laptops, Benzinga reported, and the broader market is already tight. JPMorgan has said the memory shortage is lasting longer than analysts expected, and Micron’s management believes DRAM and NAND markets will stay tight beyond 2027, which means robot builders must actually be able to secure the chips they need.

Production is moving from talk to reality. Tesla has started early production of its Optimus robot at its Fremont, California factory and targets initial customer shipments toward the end of 2027, with early buyers expected to use the robots for dangerous and repetitive industrial work. The line is capable of producing 1 million units a year, though Musk estimated it would take a year or longer to reach that pace. Toyota has told investors it may need about 400,000 robots, both humanoid and conventional, to modernize its factories, at a cost of as much as $6.4 billion a year starting in 2028.

Forecasts are rising to match. Goldman Sachs lifted its 2035 humanoid shipment estimate nearly fivefold to about 6.5 million units, a market of around $138 billion, and sees each robot carrying thousands of dollars in chip content. For perspective, only about 7,000 humanoids were sold worldwide in 2025, according to The Motley Fool, so the gap between today’s market and those projections is enormous.

Memory makers are already positioning. On Micron’s fiscal fourth-quarter call, CEO Sanjay Mehrotra called self-driving cars the first major deployment of physical AI and said humanoid robots should have similar memory and storage needs. He noted that Level 4 and higher autonomous vehicles typically need more than 200GB of memory and multiple terabytes of storage, and that several physical AI customers are already sampling Micron’s next-generation products.

There are caveats. New supply from Micron, Samsung and SK Hynix could arrive faster than expected, AI spending could slow, and memory could be used more efficiently, any of which would temper the demand story. Still, the discussion shows how the next phase of AI is widening from data centers to machines that walk, lift and work alongside people, and why memory, not just processors, is becoming a strategic component of that shift.