InvestAlso reported elsewhere2 publishers2 min readPublished
Counterpoint's 20-fold robot DRAM forecast needs roughly ten times as many robots by 2030
Counterpoint Research projects DRAM for humanoid robots will grow about 20-fold, from 8,300 terabits this year to 172,000 terabits in 2030. Memory per robot only about doubles, so by our calculation unit volume carries the rest.
The Investor · Invest desk
Key figures: robot DRAM demand of 172,000 terabits in 2030, up from 8,300 terabits; DRAM per robot of 39 GB, up from 19 GB; implied robots in 2030 about 551,000; implied unit growth about 59% a year; Jetson Thor reportedly has 128 GB.
- 172,000 terabits Robot DRAM demand, 2030, claim 1 up from 8,300 terabits this year
- 39 GB DRAM per robot, 2030, claim 2 up from 19 GB this year
- 551,000 robots Implied robots in 2030, claim 19 our calculation; approximate
- 59 % a year Implied yearly unit growth, claim 21 compound, our calculation; approximate
- 128 GB Jetson Thor memory (reported), claim 13
| Figure | Value | Note | Claim |
|---|---|---|---|
| Robot DRAM demand, 2030 | 172,000 terabits | up from 8,300 terabits this year | 1 |
| DRAM per robot, 2030 | 39 GB | up from 19 GB this year | 2 |
| Implied robots in 2030 | 551,000 robots | our calculation; approximate | 19 |
| Implied yearly unit growth | 59 % a year | compound, our calculation; approximate | 21 |
| Jetson Thor memory (reported) | 128 GB | 13 |
What happened
- Counterpoint said the amount of data robots must store and process on their own is rising sharply, and it attributes the growth in their memory requirements to that.
- Samsung Electronics and SK hynix both showed next-generation low-power DRAM, including LPDDR6, at CES 2026.
- SK hynix and Nvidia are jointly developing low-power memory for Jetson Thor, a computing platform for robots.
Why it matters
- constraint The forecast holds only if robot units compound at about 59% a year for five years, so a slow 2026 or 2027 has to be made up in later years.
- contradiction Jetson Thor's reported 128 gigabytes is 6.7 times Counterpoint's 19-gigabyte average for this year, so either flagship platforms are a small share of units or the average assumes far smaller builds.
- exposure Analysts cited by the JoongAng Daily expect robots to use low-power DRAM and NAND, not HBM, and Micron is already sampling LPDDR6, so the Korean makers meet a competitor with samples out.
Counterpoint's two published figures fix a third. At 8 bits a byte, 8,300 terabits is about 1.04 million gigabytes, and at 19 gigabytes a robot that is about 54,600 robots this year [18][1][2]. The same conversion puts 172,000 terabits at 39 gigabytes a robot at about 551,000 robots in 2030 [19]. Capacity grows 20.7-fold and memory per robot 2.05-fold, so the unit count grows about 10.1-fold [15][16][17].
The unit conversion cancels out of that ratio, so the tenfold holds as long as both Counterpoint figures cover the same robots. The absolute counts also depend on the conversion. The JoongAng Daily report does not quote a Counterpoint shipment forecast, so these counts are our estimates. Lee Su-rim, an analyst at DS Investment & Securities, framed the dependence in the same terms. "Memory demand is directly proportional to the number of robots deployed, while the amount of LPDDR installed in each robot is also steadily increasing," said Lee [10].
The count can be wrong in either direction. If the 2030 average were 128 gigabytes, the reported Jetson Thor configuration, the same 172,000 terabits would need only about 168,000 robots [13][22]. If shipments disappoint instead, content per robot would have to climb well past Counterpoint's 39 gigabytes to fill the gap [2].
Micron's chief executive put vehicles ahead of humanoids. "Autonomous vehicles are the first major deployment of physical AI, which we believe will expand over time to humanoid robots and other intelligent autonomous systems," Mehrotra said on the Sept. 30 call [4]. "With the anticipated increase in both units and memory content, physical AI can become a significant driver of memory and storage demand by the end of this decade," he said [5].
Samsung's storage target is on a five-year horizon. The company has introduced a removable automotive SSD adapted for physical AI [14], and "Our goal is for removable SSDs to become a standard solution for robotic platforms within five years," a Samsung Electronics representative said [11].
We think the forecast is a bet on robot shipments, with memory content a secondary input, and the Korean makers' upside follows those shipments from a base we estimate near 54,600 robots [18]. Counterpoint's figures are in bits. The report as quoted does not include a price or a revenue estimate, so no comparison with HBM revenue is possible yet. The thesis fails if annual robot shipments are still in the tens of thousands in 2027.
What to watch
- Robot shipment figures for 2026 and 2027 against our estimate of about 54,600 robots this year and 59% annual growth.
- Whether Counterpoint or the chipmakers publish a dollar estimate or a split between DRAM and NAND per robot, which would allow a comparison with HBM revenue.
- Whether Samsung or SK hynix name robot customers for LPDDR6 beyond SK hynix's work with Nvidia on Jetson Thor.
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Counterpoint Research projects the global market for DRAM used in humanoid robots' computing systems to grow roughly 20-fold over five years, from 8,300 terabits this year to 172,000 terabits in 2030.
- [2]
Counterpoint projects DRAM capacity per robot will more than double from 19 gigabytes this year to 39 gigabytes by 2030.
- [3]
Counterpoint Research said the amount of data that robots must store and process independently is rising sharply, which is driving rapid growth in their memory requirements.
- [4]
"Autonomous vehicles are the first major deployment of physical AI, which we believe will expand over time to humanoid robots and other intelligent autonomous systems," Mehrotra said during the call.
- [5]
"With the anticipated increase in both units and memory content, physical AI can become a significant driver of memory and storage demand by the end of this decade," Mehrotra said.
- [6]
Micron has begun supplying customers with samples of LPDDR6, its next-generation low-power DRAM, to target the physical AI market.
- [7]
Samsung Electronics and SK hynix both showcased next-generation low-power DRAM, including LPDDR6, at CES 2026.
- [8]
SK hynix is working with Nvidia to develop low-power memory for the Jetson Thor robotics computing platform.
- [9]
Because humanoid robots run on batteries and are highly sensitive to power consumption, analysts believe they will primarily use low-power DRAM and high-capacity NAND flash rather than the HBM used in data center servers.
- [10]
"Memory demand is directly proportional to the number of robots deployed, while the amount of LPDDR installed in each robot is also steadily increasing," said Lee Su-rim, an analyst at DS Investment & Securities.
- [11]
"Our goal is for removable SSDs to become a standard solution for robotic platforms within five years," a Samsung Electronics representative said.
- [12]
Micron CEO Sanjay Mehrotra predicted that physical AI, including humanoid robots, would become a major growth driver for the memory industry by the late 2020s, during the June-to-August earnings announcement on Sept. 30.
- [13]
Jetson Thor is reportedly equipped with 128GB of LPDDR5X memory.
- [14]
Samsung has introduced a removable automotive solid-state drive (SSD) adapted for physical AI applications.
- [15]
Capacity growth is about 20.7-fold.
Derived - [16]
Memory per robot grows about 2.05-fold.
Derived - [17]
Implied robot unit growth is about 10.1-fold.
Derived - [18]
Implied robots this year: about 54,600 (8,300 terabits is 1,037,500 gigabytes).
Derived - [19]
Implied robots in 2030: about 551,000.
Derived - [20]
Jetson Thor's reported 128GB is about 6.7 times Counterpoint's 19GB per-robot average for this year.
Derived - [21]
A tenfold rise in robot units over five years is about 59% compound annual growth.
Derived - [22]
At a 2030 average of 128GB per robot, 172,000 terabits would need about 168,000 robots.
Derived
Sources
2 independent publishers whose own reporting we read for this story.
- en.sedaily.comRobots Emerge as Next Growth Driver for Samsung, SK hynix
1 article · October 10, 2026
- koreajoongangdaily.comHumanoid robots eyed as next driver of memory chip demand
1 article · October 12, 2026