Even if the AI bubble bursts tomorrow, memory chip prices will not return to their previous levels. Humanoid robots and advanced autonomous driving are becoming the next explosion points for storage demand, following AI data centers.
Micron's CEO stated during the conference call that each humanoid robot will require over 200GB of DRAM and several TBs of NAND flash memory. This capacity demand is comparable to Level 4 and above autonomous vehicles, representing an order of magnitude increase over current Level 2+ and Level 3 semi-autonomous cars.
He expects that by the end of this decade, physical AI will become a major driver of memory and storage demand.
Morgan Stanley predicts that humanoid robot shipments will range between 250,000 and 1.2 million units in 2030, reaching approximately 930 million units annually around 2050.
Based on a calculation of 1.2 million units, humanoid robots alone would require 240 million GB of DRAM, not including the multi-TB level NAND demand. This is just the starting phase, with an even steeper growth curve ahead.
Micron's CFO also stated that the first quarter of fiscal year 2027 will mark the bottom for annual gross margins, after which gross margins will rebound each quarter, with revenue growing quarter-over-quarter throughout fiscal year 2027.
Although HBM gross margins are lower than conventional DRAM in the current cycle, Micron has locked in most of its HBM supply for fiscal year 2027 at higher prices, which will narrow the gross margin gap with traditional DRAM.
Micron has accelerated the adoption of 1-gamma products in its mobile and client businesses, with nearly half of the revenue in this segment coming from this new technology in the fourth fiscal quarter.
Autonomous driving is considered the first major application scenario for physical AI, with subsequent expansion into humanoid robots and other intelligent autonomous systems. Several physical AI customers are already conducting sample testing on Micron's next-generation products.
For consumers, there is likely only one thing to do: accept the reality that memory and storage prices will remain high for the long term.