Micron Technology: HBM4 and the AI memory supercycle redefine what 'cyclical' means

Stevie AI on Micron Technology, Inc. (MU-USA | microntechno)

3/26/2026

Summary

Micron Technology is the world's third-largest manufacturer of DRAM and NAND flash memory, supplying the foundational compute infrastructure that underpins AI training, inference, and storage at hyperscale. The critical structural insight is this: memory is no longer a fungible commodity traded on bit-volume economics alone. High Bandwidth Memory (HBM) has transformed DRAM into a co-engineered, application-specific product embedded directly onto AI accelerator packages — a product that commands multi-year design-win commitments, meaningfully higher ASPs, and structurally superior margins. Micron has closed the competitive gap with SK Hynix in HBM market share (achieving parity in Q3 calendar 2025) and is ramping HBM4 with industry-leading specifications — greater than 11 Gb/s throughput and 30% lower power consumption versus prior generation — positioning it as a preferred supplier to NVIDIA, AMD, and Google TPU programs for the next several years. Micron's financial transformation over the past two fiscal years validates the thesis. FY2024 was a trough year by any measure: revenue of $25.1B, net income of just $0.8B, and EPS of $0.70, reflecting a severe memory down-cycle with collapsed DRAM and NAND pricing. The recovery was swift and powerful. FY2025 delivered revenue of $37.4B — a 49% top-line rebound — and net income surged to $8.5B with EPS of $7.59, a more than ten-fold increase year-over-year. This is not simply mean reversion; it reflects a fundamental mix-shift toward HBM products and data center NAND that carries structurally higher margins than commodity DRAM or client NAND. Management's Q2 FY2026 guidance of $18.7B revenue and 68% non-GAAP gross margins — both records — signals that momentum is accelerating, not plateauing. We apply a 17x forward P/E multiple to derive our price targets, reflecting Micron's evolving position from a pure-play cyclical memory manufacturer toward a structurally higher-quality AI infrastructure supplier. The 17x multiple sits at a modest discount to the broader semiconductor peer group (20-25x) to account for residual cyclicality, capital intensity, and competitive concentration risk — but represents a material premium to Micron's own historical trough multiples of 8-12x, justified by the HBM-driven floor under earnings and the structural demand tailwind from AI workload growth. Applying 17x to our FY2026 EPS estimate of $25.90 yields a price target of $440, representing approximately 23% upside from the current price of $357.31. On FY2027 EPS of $30.44, the implied target rises to $517. Even through the anticipated FY2028 inventory digestion — where EPS moderates to $22.07 — the stock screens at only 16x trough earnings, providing meaningful downside support versus history.

Thesis

**1. HBM Has Structurally Re-Priced Micron's Earnings Power** The most important development in the global memory industry over the past three years is the emergence of High Bandwidth Memory as the critical performance bottleneck in AI accelerator design. HBM is stacked directly on the same package as the GPU or TPU die via silicon interposer, delivering terabytes per second of memory bandwidth that DDR5 or LPDDR5 cannot approach. Every NVIDIA H100, H200, B200, and next-generation Rubin GPU ships with HBM — and the bandwidth requirements per accelerator are growing with each generation. This means Micron is not simply selling commodity bits; it is supplying a precision-engineered, qualification-intensive component with a 12-18 month design-win cycle, a limited supplier base (effectively three global players: Samsung, SK Hynix, and Micron), and pricing that is approximately 4-5x the ASP of standard DDR5 per bit. Micron's HBM3E product already secured at Micron-parity market share as of Q3 calendar 2025, a remarkable achievement given that SK Hynix had a significant head start as the sole HBM3E supplier to NVIDIA's H100 platform. Micron's HBM4 is on track to ramp in Q2 calendar 2026, ahead of Samsung, which continues to struggle with HBM yield issues. This sequencing matters enormously: customers qualifying Micron's HBM4 in the first half of calendar 2026 will build supply chain dependency and institutional familiarity that creates switching friction for subsequent program generations. The combination of superior specifications, competitive market share, and a rival (Samsung) still working through execution challenges is a rare window of competitive advantage for Micron. The financial implication is a step-change in corporate gross margins. Our FY2026 revenue forecast of $70.0B — essentially doubling FY2025's $37.4B — is underpinned by HBM volume ramp, DRAM ASP expansion driven by constrained HBM supply diverting DRAM wafers from standard products, and data center NAND recovery. At the guided 68% non-GAAP gross margin for Q2 FY2026 alone, the blended corporate margin profile is moving decisively away from the 20-30% range that characterized Micron through prior cycles and toward the 50-60%+ territory that defines best-in-class analog and logic semiconductor businesses. **2. Data Center NAND Recovery Provides a Second, Underappreciated Earnings Driver** The investment community's focus on HBM has overshadowed a significant concurrent recovery in Micron's NAND business, which accounts for approximately 30-35% of total revenue. Enterprise SSDs powering AI storage infrastructure — the systems that store training datasets, model weights, and inference caches — have seen pricing normalize after a prolonged oversupply correction in calendar 2023-2024. Micron's Cloud Memory Business Unit (CMBU, 39% of Q1 FY2026 revenue) and Core Data Center Business Unit (CDBU, 17%) together represent over half of total revenue, and both segments are benefiting from hyperscaler capital expenditure programs that show no signs of deceleration. The structural argument for enterprise NAND demand is straightforward: large language models require vast storage for training data and checkpoint management, inference serving requires low-latency SSD access for model weight retrieval at scale, and retrieval-augmented generation (RAG) architectures are creating new storage-intensive workloads that did not exist two years ago. Micron's competitive position in enterprise NAND is not as dominant as in HBM — Western Digital and Kioxia are credible competitors — but Micron benefits from its manufacturing scale, in-house controller development, and the ability to offer integrated DRAM/NAND solutions to hyperscaler customers that value supply chain simplification. We estimate the NAND recovery contributes approximately $5-8B of incremental revenue to our FY2026 forecast versus a normalized FY2025 baseline, with gross margins in enterprise NAND recovering toward the 40-50% range as pricing inflects and Micron's 232-layer and next-generation node transitions improve per-bit manufacturing cost. This NAND tailwind is structurally different from prior cycle recoveries because the demand floor is now supported by AI infrastructure spending, which has proven remarkably inelastic to macroeconomic conditions given hyperscaler earnings power and strategic prioritization of AI buildout. **3. Operating Leverage on a $20B+ CapEx Base Creates Explosive Earnings Torque** Micron's earnings model is fundamentally a fixed-cost operating leverage story at scale. The company is committing approximately $20-23B annually in capital expenditure through FY2028 to build and equip fabs in Boise (Idaho Fabs 1 and 2), Clay (New York Fab 1), Hiroshima (Japan HBM expansion), and Singapore. This level of capital intensity — approximately 25-30% of expected FY2026 revenue — is severe in absolute terms and compresses near-term free cash flow. However, it creates the conditions for extraordinary incremental margin expansion as utilization rates on newly commissioned equipment increase from initial qualification levels toward full production throughput. The economics of semiconductor manufacturing at scale are unambiguous: once a fab's fixed depreciation, facilities, and indirect labor costs are absorbed, each additional wafer start contributes gross margin dollars at rates well above the corporate average. Our FY2026 net income estimate of $29.3B on $70.0B revenue implies a 41.9% net margin — a figure that would have been unthinkable for Micron during any prior cycle peak and reflects both the HBM ASP premium and the operating leverage dynamic described above. The incremental EPS growth from FY2025's $7.59 to FY2026's estimated $25.90 — a $18.31 per share increase — is the most visible manifestation of this leverage. CHIPS Act incentive receipts provide a meaningful offset to the CapEx burden. We estimate approximately $2-4B annually through FY2028 in government incentive receipts related to Micron's domestic fab investments, reducing the effective net CapEx burden and preserving balance sheet flexibility. The combination of incentive receipts and strong EBITDA generation means Micron's net debt position, while elevated at approximately $5.0B in FY2026 as construction peaks, should decline toward $2.0B by FY2027 before the next CapEx-intensive phase of New York fab ramp pushes it modestly higher in FY2028. **4. The Market Continues to Undervalue the Structural Earnings Floor AI Has Created** Micron currently trades at approximately 13.8x our FY2026 EPS estimate of $25.90 — a multiple that implicitly prices in a severe cyclical correction that we believe significantly underestimates the structural demand floor AI infrastructure spending has established. Historically, Micron has traded at 8-12x earnings at cycle peaks precisely because investors correctly anticipated that commodity memory pricing would collapse and those earnings were not repeatable. The embedded assumption in that multiple framework is that memory demand is primarily driven by consumer electronics refresh cycles and PC/smartphone upgrade rates — industries with well-understood saturation dynamics and inventory cyclicality. The AI workload demand curve has fundamentally different characteristics. Hyperscaler AI CapEx for calendar 2025-2026 is tracking at $300B+ across Microsoft, Google, Amazon, and Meta — capital programs driven by competitive necessity and revenue-generating AI products rather than cyclical consumer behavior. HBM supply is constrained by the complexity of through-silicon-via (TSV) stacking and advanced packaging, meaning demand satisfaction requires years of capacity investment rather than months of production ramp. These dynamics create a more durable pricing environment than any prior DRAM cycle and justify a re-rating of Micron's normalized earnings multiple toward the 15-20x range that reflects a company with structural demand visibility rather than pure cycle exposure. At 17x our FY2026 EPS of $25.90, the stock should trade at $440 — 23% above current levels. At 17x FY2027 EPS of $30.44, the implied target is $517. We acknowledge the FY2028 inventory digestion cycle moderates EPS to $22.07, but even at that trough the stock at current prices represents less than 16x trough earnings — a level that historically has marked excellent entry points for Micron even in structurally less favorable demand environments. **5. HBM4 Customer Ramp and Q2 FY2026 Earnings Are Near-Term Catalysts with Asymmetric Upside** Micron has three identifiable near-term catalysts that could drive a significant re-rating. First, the Q2 FY2026 earnings report (reported March 18, 2026) will test whether management's record guidance for $18.7B revenue and 68% gross margins was accurately calibrated or conservative. Given that Q1 FY2026 results already showed strong momentum and that hyperscaler AI CapEx programs have not decelerated, we believe the probability of an upside surprise or reiterated full-year record guidance is high. Any beat-and-raise outcome should compress the market's residual cyclical discount. Second, HBM4 customer product ramps in Q2 calendar 2026 (approximately May-June 2026) represent a critical design-win validation event. Successful qualification and volume ramp into NVIDIA's next-generation platform would confirm Micron's competitive parity with SK Hynix in the most strategically important memory product segment and de-risk the FY2026-2027 HBM revenue contribution in our model. Third, incremental CHIPS Act funding announcements or disbursement confirmations related to the New York and Idaho fab programs would provide tangible evidence of the government subsidy offset to capital intensity and improve the market's assessment of Micron's long-run cost structure. Taken together, these catalysts represent a sequenced series of de-risking events over the next 6-12 months that should progressively compress the discount at which Micron trades relative to its earnings power. We note that the consensus is still anchored to historical Micron multiple frameworks and has been consistently too slow to revise price targets upward as the HBM structural thesis has materialized — a pattern that creates opportunity for investors who act ahead of consensus re-rating. **6. Capital Allocation Discipline and Buyback Optionality Add Shareholder Value Upside** Micron's balance sheet is entering a period of meaningful improvement. Net debt of approximately $5.0B in FY2026 declines to $2.0B in FY2027 as FCF of $8.5B and $9.0B in those years, respectively, provides substantial debt reduction capacity. With the share count of approximately 1.13-1.15B shares (management's Q2 FY2026 guidance basis), even modest buyback activity has meaningful EPS accretion potential given the stock's current valuation. Management has signaled that resumed buybacks are a priority once FCF inflects strongly positive, and the FY2026-2027 FCF profile supports that commitment. The buyback optionality is not trivial at current prices. If Micron deploys $2-3B annually in buybacks at current levels, the share count reduction of 5-8M shares per year adds approximately $0.10-0.15 to annual EPS — modest in absolute terms but directionally supportive of the EPS growth trajectory and signaling management confidence in the business outlook. The FY2028 FCF turning negative (estimated at -$2.0B) as New York fab CapEx peaks will temporarily suspend buyback activity, but this is a predictable and already-disclosed capital commitment rather than an unexpected deterioration, and the FY2029 recovery to $6.5B FCF and $31.6B net income demonstrates the earnings power durability through and beyond the construction phase.

Risks

**1. Memory Cycle Inventory Correction: FY2028 Digestion Could Be Deeper Than Modeled** Our FY2028 forecast models a moderate revenue decline from $78.0B to $71.5B and an EPS compression from $30.44 to $22.07 — a significant but manageable correction consistent with historical mid-cycle inventory digestion patterns. The risk is that this correction proves more severe in magnitude or duration. If hyperscaler AI CapEx programs slow (driven by ROI skepticism, regulatory action, or competition from more capital-efficient model architectures), DRAM and NAND demand could fall sharply below our baseline, ASPs could collapse across both segments simultaneously, and Micron's highly operationally leveraged cost structure could amplify the earnings impact. In prior downturns — FY2019 and FY2023 — Micron posted operating losses and negative free cash flow for multiple consecutive quarters. A deeper-than-expected FY2028 correction with net debt already elevated at $7.1B could constrain financial flexibility at precisely the wrong moment. **2. Extreme Capital Intensity and Concurrent Fab Execution Risk** Micron is simultaneously constructing and ramping production facilities in Idaho (Fabs 1 and 2), New York (Fab 1), Japan, Singapore, and India — a geographic and logistical complexity without precedent in the company's history. At $20-23B annually, CapEx represents approximately 25-30% of FY2026 expected revenue, and any material delay in fab qualification, equipment installation, or yield ramp directly erodes the FCF bridge we model for FY2026-2027. Clean room construction timelines have been lengthening industry-wide due to HVAC, ultrapure water system, and tool availability constraints. A 6-12 month delay in Idaho Fab 2 or New York Fab 1 ramp could simultaneously increase net debt, reduce future capacity available for HBM4/5 volume, and undermine management's credibility on capital allocation — all of which would compress the multiple investors are willing to pay. **3. Samsung HBM Recovery Would Intensify Competitive Pricing Pressure** Samsung's HBM3E yield challenges have created a two-player effective market (SK Hynix and Micron) for leading-edge HBM supply in calendar 2025-2026, which has been a key driver of the pricing discipline and ASP expansion embedded in our FY2026-2027 forecasts. If Samsung resolves its HBM manufacturing and quality issues — through process changes, packaging redesign, or thermal management improvements — HBM supply could increase materially in calendar 2026-2027, compressing ASPs and eroding the margin premium Micron currently enjoys. Samsung has virtually unlimited balance sheet resources to sustain below-cost pricing if it chooses to prioritize market share recovery, a strategy it has employed repeatedly in prior memory downturns. A Samsung HBM recovery combined with Micron and SK Hynix capacity ramps could create an oversupply scenario that arrives earlier and more severely than our FY2028 digestion baseline. **4. Customer Concentration and AI Hyperscaler Dependency** Micron's HBM revenue is highly concentrated among a small number of AI accelerator customers — primarily NVIDIA, and to a lesser extent AMD and Google. NVIDIA alone likely represents a disproportionate share of Micron's HBM demand in FY2026. This concentration creates vulnerability to any development that reduces NVIDIA's GPU shipment volumes: export control tightening targeting AI chip sales to China, a demand pause from cloud customers absorbing prior GPU purchases, or a shift in training architecture toward models requiring less HBM bandwidth per accelerator. Any of these scenarios could cause a rapid and sharp HBM revenue shortfall with limited near-term ability to redirect that high-specification capacity to alternative customers. The concentration risk is further compounded by NVIDIA's negotiating leverage — as a dominant customer, it has structural pricing power in long-term supply agreements that may limit Micron's ability to fully capture ASP upside in periods of tight supply. **5. Geopolitical and Export Control Escalation** Micron generates approximately 10-15% of its revenue from Chinese customers and has previously been subject to a Chinese government cybersecurity review that resulted in a sales ban for certain critical infrastructure customers in China (2023). Escalating U.S.-China trade tensions, additional semiconductor export controls targeting advanced memory (HBM is already subject to export restrictions to China), or retaliatory Chinese government actions against Micron's China business represent a persistent and unpredictable risk to revenue and supply chain continuity. Conversely, U.S. export control tightening could also restrict Micron's ability to source certain equipment or materials for its Japan and Singapore fabs, adding cost and timeline risk to the international capacity expansion program. CHIPS Act funding, while supportive, comes with domestic sourcing requirements and operational restrictions that add compliance complexity. **6. Valuation Sensitivity to Multiple Compression in a Risk-Off Environment** Our $440 FY2026 price target is premised on a 17x forward P/E multiple — a modest premium to Micron's historical cycle-peak multiples but a significant discount to the broader semiconductor sector. In a market environment characterized by rising interest rates, risk-off sentiment, or a broader technology sector de-rating, investors may be unwilling to sustain even a 17x multiple on what they continue to perceive as a cyclical earnings stream. If the market reverts to applying a 12x multiple — still above Micron's trough historical average — the FY2026 price target falls to $311, below current levels, and the investment thesis requires holding through what would likely be a painful mark-to-market period. The stock's high beta to risk sentiment and the narrative sensitivity of AI infrastructure investment themes (where a single high-profile AI demand disappointment from a hyperscaler can trigger sector-wide multiple compression) means the valuation risk is asymmetric in a negative macro scenario.

📈 Price Targets