Micron CEO: AI is just beginning, but memory is already insufficient. This year, DRAM and NAND demand may consume over half the market.

```

The demand for AI-driven DRAM and NAND is expanding at a pace that surpasses the industry's supply capabilities. Based on current trends, this demand is expected to account for over 50% of the total addressable market (TAM) of the memory industry this year—and according to analysts, the supply bottleneck is unlikely to be substantially alleviated before 2028.

Micron Technology CEO Sanjay Mehrotra recently told CNBC that the AI industry is still in the "early innings." He pointed out that as inference scales up, token demand will continue to rise, and the speed of token generation relies on faster and higher-capacity memory—"memory has become a strategic asset for customers." He emphasized that the core issue at present is not demand or pricing, but the fact that supply is extremely tight and cannot be rapidly expanded.

Micron’s just-announced Q2 results for fiscal 2026 support this view: the company set historical highs for revenue, gross margin, earnings per share, and free cash flow, and Q3 is expected to break records again.

Supply Gap: No Turning Point Before 2028

Micron noted that demand for both traditional servers and AI servers remains strong, but is simultaneously constrained by tight supplies of DRAM and NAND. The Financial Times cited analysts, saying that considering the construction timeline for new wafer fabs, supply shortages are unlikely to ease before 2028.

Korean media outlet Global Economic gave more specific forecasts: as next-generation AI GPUs like NVIDIA Vera Rubin begin to ramp up 12-layer HBM4 production, more cautious estimates suggest that by 2027, the industry may only be able to meet about 60% of DRAM demand. This supply-demand dynamic will reinforce memory manufacturers' pricing power, but on the flip side, if mass production progress falls short, manufacturers' market positions could also come under pressure.

The push from the demand side is still accelerating. Agentic AI workloads are driving CPU memory support specs to expand to 400GB—about four times current levels. Next-generation AI GPUs such as NVIDIA Vera Rubin and AMD MI400 will be equipped with HBM4, significantly boosting bandwidth and capacity ceilings. LPDDR, thanks to its power efficiency, is becoming the preferred solution for large-scale AI deployments.

Micron’s Product Response: HBM4 Delivered, HBM4E Mass Production Next Year

Facing tight supply and demand, Micron is advancing next-generation plans along multiple product lines. The company is now supplying 36GB (12-Hi) HBM4 DRAM for the NVIDIA Vera Rubin platform, and its HBM3 process is expected to reach mature yields. Next-generation HBM4E memory is scheduled to begin mass production ramp-up next year.

In the LPDDR sector, Micron recently launched its 256GB SOCAMM2 solution based on LPDDR5X, offering up to 2TB of capacity. For DDR5, the company is supplying for NVIDIA Groq 3 LPX, with Groq LPU single chips supporting up to 12TB capacity.

The consumer market is also feeling the ripple effects of supply shortages. Micron expects PC and smartphone shipments to see a low double-digit decline, mainly driven by limited supply and rising prices. A noteworthy trend signal: 32GB is becoming the standard memory configuration for PCs running local Agentic AI workflows.

Risk warning and disclaimer clauseThe market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situation. Investing based on this is at your own risk. ```