Business

Claude View

Micron is a commodity memory manufacturer whose economics are dictated by a single variable: the balance between global DRAM/NAND supply and demand. The business has been structurally transformed by AI-driven data center demand, which has shifted the revenue mix toward high-margin HBM and server memory and created a pricing environment unlike any prior cycle. The market is most likely underestimating how long Micron can sustain above-mid-cycle margins as HBM supply remains structurally constrained through at least calendar 2027.

How This Business Actually Works

Micron manufactures two commodity semiconductors – DRAM (~76% of FY2025 revenue) and NAND (~23%) – in capital-intensive fabs that run 24/7. The economic engine is simple: shrink the process node to produce more bits per wafer, sell those bits at whatever price the market dictates, and hope that industry supply discipline holds.

The cost structure is dominated by depreciation (FY2025: $8.4B on $22.5B COGS) and fab operating costs. R&D runs about 10% of revenue. The critical economic lever is average selling price – a 10% change in ASP flows almost entirely to the bottom line because variable costs per bit are low relative to the fixed cost base. This is a business with enormous operating leverage.

The AI transformation is visible in one number: CMBU (Cloud Memory) went from $1.9B in FY2023 to $13.5B in FY2025 with a 45% operating margin – a 257% revenue increase in a single year. This segment, centered on HBM and high-capacity server DRAM, is where Micron's value creation is concentrated. HBM is not a commodity in the traditional sense – it requires advanced 3D stacking with through-silicon vias, and there are only three qualified suppliers globally (Samsung, SK hynix, Micron). Micron's HBM3E 12-high product became the majority of HBM shipments by Q4 FY2025, and HBM4 samples have shipped to key customers.

The bargaining power dynamic has inverted in data center memory. Historically, memory buyers dictated price. In the current AI buildout, hyperscale customers are signing prepayment agreements and long-term contracts to secure HBM supply – a structural change from the traditional spot-driven commodity model.

MCBU (Mobile/Client) is still the largest segment at $11.9B, but it grew only 2% in FY2025 because Micron deliberately constrained DRAM supply to mobile in favor of higher-value data center markets. AEBU (Automotive/Embedded) grew 3% with steady margins – this is the most defensible segment but the least exciting. The real story is the massive pivot toward data center, which now represents 55% of revenue (CMBU + CDBU combined) versus 27% just two years ago.

The Playing Field

Memory is a global oligopoly with three producers controlling ~95% of DRAM output: Samsung (~40%), SK hynix (~35%), and Micron (~25%). In NAND, the field is wider – Samsung, SK hynix/Solidigm, Kioxia, Western Digital (SanDisk), and Micron each hold meaningful share, plus Chinese entrants YMTC and CXMT.

Micron's real competitors – Samsung Semiconductor and SK hynix – are not listed in the US. The table above captures adjacent semiconductor peers. What it reveals: Micron's margins are now firmly in the upper tier of the semiconductor industry, but its P/E of 22x is the cheapest in the group by a wide margin. The market still prices it as a cyclical commodity company despite the structural demand shift.

The moat question matters. NVIDIA has a genuine platform moat (CUDA ecosystem). Texas Instruments has an analog moat (product longevity, low-cost fabs). Micron's moat is narrower but real: it is one of three companies on earth that can manufacture leading-edge DRAM, and one of three qualified HBM suppliers. The barriers are enormous ($16B+ capex annually, decades of process know-how, 60,000+ patents), but the product is still fundamentally a commodity. The moat depends on maintaining technology parity with Samsung and SK hynix at each node transition – fall behind on a single generation (as Samsung did with HBM3E yields) and market share moves fast.

Micron occupies a unique position: the fastest revenue growth in the semiconductor peer set combined with margins that are rapidly converging with traditionally higher-margin analog and fabless businesses. The question is whether this growth rate is sustainable or cyclical.

Is This Business Cyclical?

Memory is the most cyclical corner of the semiconductor industry. The cycle hits through three channels simultaneously: price (ASPs can fall 50%+ peak-to-trough), utilization (fabs are either full or burning cash), and inventory (customers double-order in upcycles and destock violently in downturns).

The swing from FY2023 to Q2 FY2026 is staggering: a $5.8B annual loss became a $13.8B quarterly profit. Gross margins moved from -9% to 74%. The FY2023 downturn featured $1.8B in inventory write-downs and negative gross margins, followed by the most powerful upcycle in memory history driven by AI demand.

Three arguments suggest the upswing is more durable than past peaks. First, HBM demand is structurally supply-constrained – adding HBM capacity requires 3x the wafer area of standard DRAM. Second, hyperscale customers have prepaid for supply, reducing destocking risk. Third, all three major producers are investing in advanced packaging rather than adding commodity wafer capacity, limiting the supply response that typically kills margins.

The counter-argument: memory cycles have humbled every analyst who declared "this time is different." Google's TurboQuant algorithm (announced March 2026) demonstrated 6x reduction in AI memory requirements, briefly sending memory stocks into a bear market. Technology risk to demand is real, even if the current supply/demand balance favors producers.

The capex cycle is the other half of the story. Micron is spending $15.9B in FY2025 and guiding to ~$18B in FY2026 on new fabs in Idaho, New York, India, and Singapore. CHIPS Act grants of up to $6.4B and a 35% investment tax credit partially offset the cash burden, but free cash flow was negative in FY2023-2024 and only turned marginally positive in FY2025. The Q2 FY2026 results show the business finally generating massive cash, with a 30% dividend increase signaling management's confidence in durability. But capital intensity will remain elevated through at least 2028.

The Metrics That Actually Matter

DRAM ASP is the single most important variable. It determines whether Micron earns $8B or loses $6B. Everything else – gross margins, FCF, ROIC – is downstream. Tracking contract vs. spot pricing trends gives you the earliest signal on the cycle.

HBM revenue share is the new structural metric. HBM carries substantially higher ASPs and margins than conventional DRAM. As long as HBM is supply-constrained and Micron maintains technology parity (HBM3E today, HBM4 ramping calendar 2026), the mix shift provides margin support even if conventional DRAM prices soften.

Gross margin is the operating leverage gauge. Because ~60% of COGS is depreciation and fixed fab costs, each incremental revenue dollar above breakeven (~$20B annualized) falls through at high incremental margins. The move from -9% to 74% gross margin on a 2.4x revenue increase illustrates this leverage precisely.

The Q2 FY2026 quarter is extraordinary – revenue nearly doubled sequentially to $23.9B, and gross margins expanded 18 percentage points in a single quarter to 74.4%. Diluted EPS hit $12.07, compared to $0.30 in Q3 FY2024. The sustainability of this trajectory is the central valuation question.

What I'd Tell a Young Analyst

Study past memory cycles before building a model. The 2018-2019 downturn, the 2022-2023 collapse, and the 2024-2026 recovery all follow the same pattern: supply/demand imbalance drives extreme margin swings. Your job is to figure out where we are in the cycle, not to extrapolate the current quarter forward.

Watch DRAM contract price data monthly. When contract prices start declining quarter-over-quarter, the cycle is turning. Everything else – management commentary, analyst consensus, stock momentum – lags the pricing data.

Do not model Micron as a secular growth story. It is a cyclical business with a structural tailwind (AI/HBM). The tailwind does not eliminate cyclicality – it shifts the mid-cycle earnings level higher. The right valuation framework is normalized earnings at mid-cycle margins (35-40% gross margin historically), adjusted upward by 5-10 points for the HBM mix shift.

The biggest risk the market may be underpricing: Google's TurboQuant and similar AI efficiency algorithms could meaningfully reduce memory intensity per AI workload. If inference becomes the dominant use case (as it likely will) and inference memory requirements flatten, the demand trajectory could disappoint even as AI adoption accelerates.

The biggest opportunity: at a forward P/E of ~5x (based on current annualized earnings), the stock is pricing in a severe earnings decline. If HBM structural demand keeps margins elevated for 2-3 more years, the current valuation is deeply disconnected from reality. The 30% dividend increase signals management confidence in durability.