Robot DRAM demand is set to explode — Counterpoint projects a 20x jump by 2030, from ~8,300 terabits this year to ~172,000 terabits. Per-unit content only doubles (19GB → 39GB), so the real driver is sheer volume: millions more humanoid robots.
Why the memory surge? Humanoids process camera feeds and sensor data in real time, running increasingly heavy AI models. Battery constraints mean the pull will be low-power DRAM and high-capacity NAND — not server RDIMMs.
Micron's CEO noted that a Level 4+ autonomous vehicle already carries over 200GB of memory and several terabytes of storage. Humanoids will need similar loads. Counterpoint also sees rising demand for compute chips, including NVIDIA GPUs, as models scale.
HBM dominates data centers. This is a new end market for the same three suppliers — different product, different use case.
172,000 terabits by 2030 is a big multiple off a small base, but it's not yet a second HBM wave. The question: how many humanoid units need to ship before this actually moves $DRAM pricing?
Why the memory surge? Humanoids process camera feeds and sensor data in real time, running increasingly heavy AI models. Battery constraints mean the pull will be low-power DRAM and high-capacity NAND — not server RDIMMs.
Micron's CEO noted that a Level 4+ autonomous vehicle already carries over 200GB of memory and several terabytes of storage. Humanoids will need similar loads. Counterpoint also sees rising demand for compute chips, including NVIDIA GPUs, as models scale.
HBM dominates data centers. This is a new end market for the same three suppliers — different product, different use case.
172,000 terabits by 2030 is a big multiple off a small base, but it's not yet a second HBM wave. The question: how many humanoid units need to ship before this actually moves $DRAM pricing?