Humanoid Robots Are About to Devour DRAM — 20x Demand by 2030

The Next Memory Tsunami Is Coming From Robots, Not Data Centers

Humanoid robots are about to become the single fastest-growing customer for DRAM in the semiconductor industry — and the scale of that demand is hard to overstate.

According to a new projection from Counterpoint Research, global DRAM demand for humanoid robot “brains” will surge more than 20-fold by 2030, climbing from roughly 8,300 terabits (Tbit) this year to approximately 172,000 Tbit. That translates to around 21,500 terabytes (TB) — a number that, while it sounds abstract, is equivalent to filling thousands of server racks with nothing but robot-grade memory.

What makes this projection particularly significant is that it arrives at a moment when the memory industry is already wrestling with the demands of AI data centers. High-bandwidth memory (HBM), used extensively in GPU-based AI training and inference, has been the primary growth engine for Samsung Electronics and SK Hynix. But as those companies look beyond the current cycle, humanoid robots represent a second wave — one that could sustain their revenue even as AI data center demand normalizes.

The Numbers Behind the Boom

The per-robot DRAM requirement is itself doubling. Counterpoint Research estimates the average humanoid robot will carry 19 gigabytes (GB) of DRAM today, rising to 39 GB by 2030. That jump isn’t just about raw capacity — it reflects a fundamental shift in how these machines process information.

Early humanoid robots primarily relied on simpler sensor inputs and reactive programming. The latest generation, powered by increasingly complex AI models, needs to ingest and cross-reference video, audio, LiDAR, and tactile data simultaneously. This is where world models come in — AI systems that simulate physical environments to predict outcomes before a robot moves. World models are computationally expensive and memory-hungry, demanding real-time access to large datasets that DRAM is uniquely positioned to provide.

Micron’s CEO Sanjay Mehrotra drew a direct comparison to autonomous vehicles last year, noting that Level 4 and above self-driving cars typically require more than 200 GB of memory and several terabytes of storage. Humanoid robots, he suggested, will land in that same ballpark.

That’s not a trivial benchmark. An entire data center cluster can operate within 200 GB of DRAM. A single robot matching that capacity changes the economics of memory manufacturing overnight.

Who Wins, Who Loses

The immediate winners are clear: Samsung Electronics and SK Hynix. Both companies dominate the global DRAM market and have been aggressively expanding their HBM production lines. The robotics demand adds another layer of volume that their fabs are already configured to handle. Unlike custom silicon, DRAM is a commodity product — high volume, relatively standardized — which means these Korean giants can scale production without the design overhead of logic chips.

Nvidia and other GPU manufacturers also benefit indirectly. Counterpoint projects that compute semiconductor demand alongside DRAM, meaning chips like Nvidia’s offerings will see increased deployment in robotics platforms. The entire AI hardware stack — memory, compute, storage — gets a boost.

But there are losers in this scenario. Chinese memory manufacturers, still years behind Samsung and SK Hynix in process technology, face a future where the companies they compete with have diversified beyond the AI data center boom into a stable, long-term robotics demand curve. If humanoid robots become the next consumer of DRAM at scale, the gap between Korean and Chinese memory firms could widen precisely when China most needs to close it.

Tier-2 memory players may also struggle. The DRAM market has already consolidated around three major producers — Samsung, SK Hynix, and Micron. New demand categories tend to reinforce incumbents with the capital to expand fab capacity quickly, not newcomers.

What Happens Next

The 2030 timeline is ambitious. Humanoid robot production is still in its infancy. Tesla’s Optimus, Figure AI’s robots, and units from Boston Dynamics and Agility Robotics are being deployed in limited numbers — mostly in controlled industrial environments. The jump from hundreds of units today to the thousands, perhaps tens of thousands, needed to absorb 172,000 Tbit of DRAM annually requires both technological maturation and dramatic cost reductions.

There’s also the question of memory architecture. If future robots shift toward more specialized memory types — such as next-generation HBM variants or even emerging non-volatile alternatives — the DRAM-specific demand could plateau below Counterpoint’s projection. The 20-fold figure assumes DRAM remains the primary memory solution, which is plausible but not guaranteed.

For investors and industry watchers, the key takeaway is that the semiconductor narrative is shifting. The conversation has moved from “AI will consume all the chips” to “AI robotics will consume all the chips,” and the memory sector — long seen as cyclical and vulnerable to oversupply — may find itself with a structural demand driver that lasts well beyond the current AI hype cycle.

The question isn’t whether robots will need memory. It’s whether the memory industry is preparing for the sheer volume that’s coming.

Beyond the Headlines: Second-Order Effects and Supply Chain Implications

The implications of this demand surge extend far beyond the obvious. First, consider the infrastructure investment required. Expanding DRAM production capacity to meet robotics demand will require billions in new fab construction — capital that Samsung and SK Hynix are already committing, but which could strain supply chains for specialized equipment and materials.

Second, the labor market may feel the pressure. While robots are designed to replace human labor, the factories that build them — and the memory chips they contain — will need a larger, more skilled workforce. This creates a paradox: the very technology designed to solve labor shortages may initially exacerbate them during the build-out phase.

Third, geopolitical dynamics could shift. South Korea’s dominance in memory production already gives it significant economic leverage. A 20-fold increase in demand for robot-specific DRAM could further entrench that position, potentially drawing regulatory scrutiny in countries concerned about semiconductor concentration.

Finally, there’s the environmental angle. DRAM manufacturing is energy-intensive, and scaling production to meet robotics demand will increase the semiconductor industry’s carbon footprint unless companies accelerate their transition to renewable energy sources.

These second-order effects suggest that the humanoid robot DRAM boom won’t just transform chipmakers — it will ripple through entire supply chains, labor markets, and geopolitical landscapes.

The Path to 172,000 Tbit: Production Scenarios and Market Dynamics

To understand what 172,000 Tbit of annual DRAM demand looks like, consider that a single modern data center might consume 500-1,000 TB of memory. The robotics projection implies demand equivalent to 20-40 fully-loaded data centers — every year, starting in 2030.

This level of demand would require Samsung and SK Hynix to essentially double their current DRAM output, or more, assuming AI data center demand doesn’t collapse. Given that these companies are already investing heavily in HBM for AI training, the robotics demand adds a parallel production challenge that could test their manufacturing flexibility.

Market dynamics will also evolve. Currently, DRAM pricing is cyclical, driven by periods of shortage and oversupply. A stable, growing demand source like robotics could dampen those cycles, making memory investment less volatile but also reducing the upside potential during boom periods.

For now, the industry is watching. The projections are bold, the technology is nascent, but the direction of travel seems clear: if humanoid robots reach even a fraction of their promised potential, the memory industry will never be the same.