Key Highlights

  • Micron Technology (Nasdaq: MU) on the Nasdaq Global Select Market reported in its Q2 FY2026 10-Q that AI-driven memory Demand is growing faster than the industry's ability to increase Supply — a statement that describes an inelastic demand environment with no direct precedent in the company's history.
  • DRAM average selling prices rose in the mid-110% range year-on-year in Q2 FY2026, while bit shipments also rose in the mid-40% range — simultaneous Volume and price growth that distinguishes this cycle from prior pricing recoveries.
  • The data centre segment (CMBU + CDBU) generated approximately $15.46 billion of the $23.86 billion Q2 FY2026 Revenue, representing roughly 65% of total revenue from a segment that accounted for less than 20% of revenue three years prior.
  • Cloud servers supporting AI and data-centric workloads require significantly increasing quantities of DRAM, including HBM, and NAND as the task of turning data into insight becomes increasingly memory-centric — per the FY2025 10-K.
  • Unlike consumer electronics cycles, AI infrastructure Investment is driven by competitive necessity among hyperscalers, reducing the price elasticity of demand that historically triggered supply-demand corrections in the memory market.

 

Every analytical model has boundary conditions — assumptions about the world that hold within a certain range of inputs but break down when inputs fall outside that range. The standard framework for analysing semiconductor memory companies was built on a specific understanding of demand: cyclical, consumer-led, price-elastic, and ultimately tied to the replacement schedules of personal computers and smartphones. That framework worked reasonably well for thirty years. It is no longer sufficient.

The Geometry of the Old Demand Curve

In the conventional memory cycle, demand followed a recognisable pattern. Consumer electronics manufacturers — PC assemblers, smartphone OEMs — would increase orders during periods of product launches and reduce orders during inventory digestion periods. Enterprise IT buyers would time server procurement to budget cycles and product refresh schedules. In all cases, the underlying demand was discretionary at the Margin: a company could delay a server purchase for one quarter, or a smartphone manufacturer could reduce DRAM specifications to manage costs. This marginal elasticity acted as a natural governor on pricing. When DRAM prices rose sharply, buyers reduced specifications, extended inventory lifetimes, or substituted where possible. Demand softened, supply caught up, and prices corrected.

The textbook evidence for this pattern is visible in Micron's own historical data. The company's gross margins in FY2023 fell to approximately 22% — barely covering costs — before recovering to 38% in H1 FY2025 and reaching 68% for the six months ended February 26, 2026. This is the anatomy of a conventional cycle. But the steepness and speed of the current recovery suggests something beyond convention.

The New Demand Architecture

The FY2025 10-K provides the clearest articulation of what has changed: cloud servers supporting AI and data-centric workloads require significantly increasing quantities of DRAM, including HBM, and NAND as the task of turning data into insight becomes increasingly memory-centric. This single sentence contains within it a description of a fundamentally different demand function. The key word is increasingly. AI model sizes are growing. Inference workloads are expanding. The number of queries processed per day across all AI applications is rising. Each of these trends independently increases memory demand. Combined, they create a demand curve that does not flatten as prices rise, because the entities doing the buying — Microsoft, Google, Amazon, Meta — are not making discretionary purchasing decisions.

When a hyperscale cloud company builds AI infrastructure, it is making a strategic investment to remain competitive in a market where falling behind means losing customers, losing developers, and potentially losing the AI race entirely. The decision to build the next AI Training cluster is not made by a procurement manager optimising against a quarterly budget. It is made at the level of the chief executive as a multi-year Capital commitment with existential stakes. Memory pricing is, at most, a second-order consideration. This is the definition of inelastic demand.

The Data That Proves the Hypothesis

The financial evidence for this structural shift is in Micron's segment revenue data. In Q2 FY2026, the Cloud Memory Business Unit generated $7.75 billion in revenue and the Core Data Centre Business Unit generated $5.69 billion, for a combined data centre revenue of approximately $13.44 billion out of total revenue of $23.86 billion. Compare this to Q2 FY2025, when CMBU and CDBU generated $2.95 billion and $2.24 billion respectively — a combined $5.19 billion. Data centre revenue grew by $8.25 billion year-on-year — faster than the growth of any other segment and faster than the overall revenue growth rate. Within data centre, the CMBU growth reflects the HBM ramp, which is the highest-value, highest-margin product in Micron's portfolio.

The pricing data corroborates the volume data. The Q2 FY2026 filing attributes DRAM revenue growth of 207% year-on-year to a mid-110% increase in average selling prices and a mid-40% increase in bit shipments. Both dimensions rising simultaneously is the defining characteristic of a market where demand is genuinely outrunning supply. In a conventional upcycle recovery, prices rise because supply has been cut; volumes often remain flat or decline as customers manage inventory. When both prices and volumes rise, the supply constraint is real and broad-based.

How Long Can This Last

The honest answer, which Micron's filings support but cannot definitively answer, is that the duration of the current supply-demand imbalance depends on the race between two variables: the rate at which AI infrastructure investment grows and the rate at which new memory Manufacturing capacity comes online. New fab construction takes three to five years from decision to first production. Micron's Idaho fab, for which groundbreaking occurred in September 2022, will produce its first DRAM wafers in mid-calendar 2027. The New York fab, which broke ground in January 2026, will provide supply in 2030 and beyond. AI investment, driven by the competitive dynamics of the largest technology companies in the world, is unlikely to pause for three to five years while new capacity is built.

Disclaimer: This article is for informational and educational purposes only and does not constitute financial, investment, or any other form of professional advice. All data and figures are sourced exclusively from Micron Technology Inc.'s (NASDAQ: MU) Form 10-Q for the quarter ended February 26, 2026 and Form 10-K for the fiscal year ended August 28, 2025, as filed with the US Securities and Exchange Commission.