Micron's CEO says AI memory demand is just in its 'first innings' and memory supply is already insufficient

The demand for DRAM and NAND from AI has expanded exponentially, and the memory supply crunch we're currently facing is unlikely to ease anytime soon. At least that's what Sanjay Mehrotra, CEO of memory and storage maker Micron, believes, and in his view, the demand isn't going away and will only get stronger.

In an interview with CNBC, Mehrotra described the current phase of AI-driven demand as just the "first innings," suggesting that we are still early in the cycle. As AI companies scale up compute, faster and higher-density memory will become essential to keep up with growing workloads.

He added that as inference and token demand rise, so will the need for both higher-capacity and higher-performance memory. AI GPUs rely heavily on HBM, while AI CPUs use DRAM, and both are currently in short supply. Mehrotra emphasized that the issue goes beyond pricing, as memory production cannot be ramped up quickly enough to meet demand.

2

On the hardware side, upcoming platforms such as NVIDIA's Vera Rubin and AMD's Instinct MI400 are expected to adopt HBM4, pushing both bandwidth and capacity to new levels. According to the South Korean outlet Global Economic, as shipments of 12-layer HBM4 for next-generation AI GPUs ramp up in earnest, more cautious forecasts suggest the industry may meet only around 60% of DRAM demand by 2027.

Numerous companies are already being hit by this situation, with the fallout including delayed launches, price hikes, and stopgap fixes such as reviving older GPUs with more VRAM or enabling mixed DDR5 setups without XMP. That being said, if we follow the latest trajectory, data suggests AI demand for DRAM and NAND is expected to exceed 50% of the total industry TAM this year. In line with this trend, the Financial Times, citing analysts, adds that the supply crunch is unlikely to ease before 2028, given the time required to build new fabrication plants.