Humanoid Robots Could Need 20 Times More Memory Chips By 2030

The post Humanoid Robots Could Need 20 Times More Memory Chips by 2030 appeared first on 24/7 Wall St..

Elon Musk needed two words to dismiss a skeptic on X on October 26, 2026: “Silly analysis.” The Tesla (NASDAQ:TSLA) chief executive posted the reply at 1:32 p.m. ET. It drew more than 1 million views and brought a memory forecast from earlier in the month back into discussion.

Counterpoint Research released that forecast on October 11, projecting that demand for DRAM from humanoid robots will rise about 20 times by 2030. DRAM is the working memory that holds data while a processor is using it.

Micron Technology (NASDAQ:MU) rose 2.98%, or $26.15, to $903.70 and traded at $913.25 after hours. Tesla gained 5.21% to $391.62 after Barclays (NYSE:BCS) upgraded it to Overweight with a $450 target, citing hopes for a SpaceX merger. For Micron shareholders, a very high growth rate on a small base can still barely register at a company that reported $54.23 billion of revenue in a single quarter.

What the Counterpoint Forecast Actually Measures

Counterpoint Research projects humanoid DRAM demand rising from about 8,300 terabits in 2026 to about 172,000 terabits in 2030, an increase of 20.7x. Counterpoint Research also expects average DRAM per robot to grow from 19 gigabytes to 39 gigabytes.

Most of the growth therefore comes from building many more robots, each carrying somewhat more memory. The forecast covers DRAM only and leaves out storage chips.

X user @zephyr_z9 replied that the forecast implies annual production of about 550,000 humanoids and total robot DRAM demand of 21.5 petabytes. @zephyr_z9 put that at about 0.04% of roughly 48 exabytes of DRAM produced in 2026.

Musk’s Own Chip Cuts Weaken His Post

Musk wrote that “Robot RAM needs will exceed everything else on Earth, as you need the robots to be on physical substates in order to shape atoms.” Musk added that “~100% of SI will be deep space compute,” which puts superintelligence memory off-planet.

On October 1, Musk said Tesla had cut memory on its AI5 chip in half, to 72GB, and on AI6 by a third, to 144GB. He called the cuts “the only way to get enough volume for Optimus production.” The company most eager to build humanoids is reducing memory per unit to reach volume, which shows that cost and supply limit how much memory each robot carries.

Micron’s Robot Upside Arrives Late in the Decade

Micron’s chief executive, Sanjay Mehrotra, said on the September 30 earnings call that “humanoid robots will also require similar amounts of memory and storage as autonomous vehicles.” Highly autonomous vehicles typically carry more than 200GB of memory, well above Counterpoint’s per-robot figure for 2030.

Micron has begun sampling LPDDR6, a low-power DRAM, to physical AI customers. Low power draw matters in battery-powered machines. SK hynix is developing low-power memory for a major robotics computing platform, so Micron will have to compete for these orders.

Micron’s Automotive and Embedded segment, the business closest to robotics, produced $6.82 billion of fiscal fourth-quarter revenue, a small share of the total that robot orders would take years to move.

What Micron Investors Need to Watch Next

Micron’s data center business supports the investment case, and robot demand is likely a long-dated addition on top of it. Micron forecast fiscal first-quarter 2027 revenue to $61.5 billion, and executives said more than 75% of shipments are committed for the year. The stock has fallen 16.49% over the past month.

The main risk is the memory cycle, because new clean rooms coming online in late ’28 and beyond could move pricing more than robot demand can offset.

Tesla offers less direct exposure to this theme because Optimus has not yet sold robots at commercial scale. Micron’s next report is estimated for December 16, though the company has not confirmed the date. If management names robot design wins at that report, the argument gains support. If it does not, Counterpoint Research’s 20-times robot DRAM forecast likely remains a 2030 story rather than a near-term revenue driver.

The post Humanoid Robots Could Need 20 Times More Memory Chips by 2030 appeared first on 24/7 Wall St..