Story

Claude View

The Full Story

Micron Technology's narrative over the past three fiscal years (FY2023-FY2025) is a story of cyclical devastation followed by one of the most dramatic recoveries in semiconductor history, propelled by a structural demand shift toward AI. Management's credibility improved markedly: they promised supply discipline during the downturn, executed a painful 15% headcount reduction, and then rode the AI-driven HBM supercycle to record profitability. The one thread that did not resolve cleanly is China – the CAC ban of May 2023 cost Micron a low-double-digit percentage of worldwide revenue, and the company ultimately stopped supplying server chips to Chinese data centers in late 2025. What changed most was the company's identity: Micron entered FY2023 as a broad-based memory commodity supplier and exited FY2025 as an AI infrastructure play, reorganizing its entire business unit structure around that thesis.

The Narrative Arc

The arc has three distinct chapters. First, peak-cycle exuberance in FY2022 when memory pricing was strong and Micron earned $8.7 billion in net income. Second, the most severe downturn in the company's recent history during FY2023: revenue fell 49%, gross margins turned negative for the first time in a decade, and management was forced into a 15% workforce reduction, executive salary cuts, and bonus suspensions. Third, a recovery that accelerated far beyond the normal cyclical bounce because AI infrastructure spending created unprecedented demand for high-bandwidth memory.

The speed of the recovery is the key data point. Revenue went from $15.5 billion (FY2023) to $37.4 billion (FY2025) in just two years – a 140% increase. Gross margins swung from negative 9% to positive 40%. This was not simply mean-reversion; it was a structural demand shift that gave Micron pricing power it had never had in commodity DRAM.

What Management Emphasized – and Then Stopped Emphasizing

What rose to prominence. AI and HBM demand went from a footnote in FY2023 (HBM3E "sampling") to the defining narrative by FY2025 (HBM representing the majority of CMBU shipments). CHIPS Act funding and US manufacturing expansion grew steadily from aspiration to concrete agreements – $6.4 billion in federal grants secured. The data center focus became so dominant that Micron reorganized its entire business structure in Q4 FY2025, splitting its old CNBU and SBU segments into CMBU (cloud memory, HBM for all) and CDBU (core data center).

What faded. Supply discipline was the survival mantra of FY2023 – wafer starts cut, fabs underutilized, equipment redeployed from older nodes. By FY2025, capacity was fully utilized and the language shifted entirely to expansion. China risk, which consumed enormous management attention in FY2023 (with detailed discussion of the CAC decision's uncertain scope), was reduced to a single paragraph by FY2025 after Micron effectively conceded the Chinese server market. NAND leadership language also dimmed relative to DRAM, reflecting the margin disparity between the two product lines.

The quiet pivot. The FY2025 segment reorganization was the clearest signal of narrative realignment. The old structure (CNBU, MBU, EBU, SBU) was replaced with one organized around AI adjacency: CMBU for hyperscale cloud and HBM, CDBU for mid-tier data center, MCBU for mobile and client, and AEBU for automotive and embedded. This was not cosmetic – it reflected a deliberate supply allocation decision, as management admitted they were constraining MCBU supply to feed higher-value data center segments.

Risk Evolution

Risks that diminished. Oversupply and inventory risk, the existential threat of FY2023 (when $1.83 billion of inventory was written down), effectively disappeared by FY2025 as AI demand outstripped industry supply. Management described the environment as "AI-driven demand outpacing industry supply."

Risks that intensified. Three new risk clusters emerged. First, capacity expansion execution risk became the dominant concern. Micron committed to building new fabs in Idaho and New York, modernizing facilities in Japan, Singapore, and Taiwan, and constructing an assembly facility in India – all simultaneously. Capex surged from $7.7 billion in FY2023 to $15.9 billion in FY2025, with $4.5 billion per quarter projected for FY2026. The FY2025 10-K added an entirely new risk factor about the challenges of building fabs in regions where "fab building has been uncommon in recent years." Second, AI dependency risk appeared in the risk factors for the first time in FY2024 as "uncertainties and outcomes associated with the use and evolution of AI." By FY2025, the company's revenue concentration in AI-driven segments was substantial. Third, tax risk from OECD Pillar Two became material – Singapore enacted Pillar Two legislation effective for Micron in 2026, which will significantly increase tax expense given the company's $1.05 billion Singapore tax incentive benefit in FY2025.

Persistent risks. Taiwan concentration (majority of DRAM output), geopolitical China risk, and competition from Samsung and SK hynix remained consistent threats throughout.

How They Handled Bad News

The FY2023 downturn and the China CAC ban were the two major bad news events, and management handled them differently.

The downturn response was transparent and decisive. When revenue collapsed 49% and margins turned negative, management did not hedge or delay. They cut 15% of headcount, suspended executive bonuses, reduced executive salaries, and cut wafer starts "significantly below peak capacity levels." The 2023 Restructure Plan delivered $130 million in quarterly cost savings – a promise that proved accurate. The 10-K language was direct about the severity, describing conditions that "deteriorated sharply" and acknowledging that "average selling prices for our products have been below our manufacturing costs."

The China response was more evasive. When the CAC banned Micron products for "critical information infrastructure operators" in May 2023, management acknowledged the revenue impact (low-double-digit percentage of worldwide revenue at risk) but emphasized uncertainty about the ban's scope – "there is no list of the companies that have been designated as critical information infrastructure operators." The stated goal of retaining worldwide market share despite the ban was quietly abandoned. By October 2025, reports emerged that Micron planned to stop supplying server chips to Chinese data centers entirely. Yet the FY2025 10-K reduced China discussion to a single paragraph and still used the present-tense language "we have been working to mitigate that impact" without acknowledging the effective market exit.

The Google TurboQuant scare (March 2026). External reporting reveals a more recent test of management credibility. In late March 2026, Google unveiled an AI algorithm called TurboQuant that reportedly reduces memory usage by 6x, triggering a sharp selloff in memory stocks. Micron's stock dropped nearly 10% in a single session. Analysts at BofA quickly called the panic a "buying opportunity," and the stock recovered within weeks. This episode illustrates the market's sensitivity to any narrative that threatens the AI-driven demand thesis underpinning Micron's valuation.

Guidance Track Record

Credibility: 7 out of 10. Management earned strong marks for executing on cost restructuring, delivering HBM products on schedule, and accurately calling the demand recovery. The two deductions come from construction timeline slippage (Boise fab pushed from CY2025 to CY2027; Clay NY still awaiting approvals) and the quiet abandonment of the China market share retention goal. Capex guidance has consistently underestimated actual spending, which could be a positive (more aggressive investment) or a warning (capital discipline eroding). Overall, Sanjay Mehrotra's team navigated the worst downturn in company history and emerged with a stronger competitive position, which weighs heavily in their favor.

What the Story Is Now

The current story has four pillars:

1. AI infrastructure play, not commodity memory. Micron repositioned itself as a structural beneficiary of the AI buildout. CMBU (cloud/HBM) went from $1.9 billion in FY2023 to $13.5 billion in FY2025 – a 7x increase. HBM3E 12-high became the majority of HBM shipments, and HBM4 samples were delivered to customers. The company's 2026 HBM production capacity is reported as sold out under long-term agreements. The "memflation" narrative – memory prices rising structurally due to AI demand outstripping supply – is the bull case.

2. Massive capex cycle with government support. Micron secured $6.4 billion in CHIPS Act grants plus a 35% investment tax credit, plus up to $5.5 billion from New York state. It is building new fabs in Idaho and New York, modernizing in Japan, expanding advanced packaging in Singapore, and building a test facility in India. This is a multi-decade capital commitment that will define shareholder returns for years. The risk is execution: semiconductor fab construction in the US "has been uncommon in recent years" per the company's own risk disclosure.

3. Profitability structurally higher but cyclicality not dead. FY2025 margins (40% gross, 26% operating) are strong but not unprecedented – FY2022 was 45% gross and 32% operating. The current cycle benefits from HBM premium pricing, but analysts note that competition from Samsung and SK hynix is intensifying. The Google TurboQuant scare in March 2026 demonstrated that any technology threatening memory demand can still trigger a violent selloff.

4. Valuation reflects the supercycle. With a stock price near $465, market cap around $525 billion, and trailing P/E near 22x, Micron is priced for continued supercycle conditions. The forward P/E of approximately 5x based on FY2026 estimates suggests the market expects earnings to more than quadruple. If the memory supercycle extends through late 2027 as analysts project, this valuation is reasonable. If supply catches up or AI spending decelerates, the downside is severe given the fixed-cost nature of semiconductor manufacturing.

What to believe: Management's execution on technology (HBM3E/HBM4 delivery, node transitions) and cost discipline during the downturn. The AI demand tailwind is real and structural, not cyclical.

What to discount: The assumption that HBM pricing power persists indefinitely. SK hynix is reportedly pursuing a US listing and expanding HBM capacity aggressively. Samsung's profits jumped 700% in early 2026 on memory demand. The competitive moat in memory is narrower than in logic semiconductors. Also discount the timeline guidance for new US fabs – Boise has already slipped two years and Clay NY has not started construction.