History

The Story Management Has Been Telling

In four fiscal years Micron's narrative has flipped from "memory is a cyclical commodity we are surviving" to "memory is a strategic AI asset we are rationing." Two events did most of the work: the FY2023 collapse (revenue cut in half, an $1.83B inventory write-down, a 15% headcount cut) and the HBM ramp that began as a "several hundred million dollar" probe in FY2024 and became a roughly $8B run-rate by Q4 FY2025. Management's early-cycle credibility is high — every concrete FY2025 promise (HBM share parity by 2H CY25, multi-billion HBM revenue, record FY2025 results) was hit or beaten — but the late-cycle narrative is now stretched: TAM forecasts have been raised five times in eighteen months, capex has been raised three times in twelve months, and the company is openly saying it can supply only "50% to two-thirds" of key customer demand. Whether memory truly stays "structural" or reverts to the cycle the FY2023 risk factors still describe is the question the current story rests on.

1. The Narrative Arc

The shape: a bad cycle (FY2022 peak → FY2023 trough), a textbook recovery (FY2024), a single product (HBM) that turned the recovery into a re-rating (FY2025), and a supply-shortage narrative now pulling forward four years of TAM into eighteen months (FY2026). The pivot point everyone remembers is the September 2024 call where Sanjay Mehrotra said "we are entering fiscal 2025 with the strongest competitive positioning in Micron's history" — that line, repeated almost verbatim every quarter since, has so far been validated by results.

2. What Management Emphasized — and Then Stopped Emphasizing

Talking-point intensity, scripted remarks (0 = not mentioned, 5 = headline theme).

Three patterns matter more than the rest:

• The 3D XPoint silence. Through FY2020 it was a flagship product line and the reason Micron operated the Lehi fab. In Q2 FY2021 the program was killed and the fab put up for sale; Lehi sold to TI in early FY2022 for ~$900M. After that, 3D XPoint was never voluntarily mentioned again — only referenced in 10-Ks as historical context. The pivot was clean and explained as a focus on data center memory, not as a write-off, but it killed years of "memory–storage hierarchy" investor framing.
• Mobile managed NAND followed the same script in 2025. In Q4 FY2025 Micron disclosed it had "ceased future mobile managed NAND product development … to focus our resources and investments on higher ROI opportunities." Two months later the language was gone from the script. Same pattern: a category that received product-launch slides for years was deprioritised in two sentences, and the reframe was capital allocation, not failure.
• The new vocabulary of FY2026. "SCA" (strategic customer agreement, multi-year, "specific commitments"), "physical AI / robotics," "agentic AI on PCs," and "trillions of dollars of AI infrastructure" are all FY2026 vintage. Each one extends the narrative further out in time and harder to falsify. Investors should track which of these still have content in two years.

3. Risk Evolution

Risk-factor emphasis in 10-K Item 1A (0 = absent, 5 = featured).

Risks that disappeared: COVID-19 (gone by FY2024), 3D XPoint commercialisation (after Lehi sale). Risks that shrank but never went away: China CAC ban — quantified in FY2023 as putting "low-double-digit percentage of worldwide revenue" at risk; by FY2025 it is referenced in past tense, but the 10-K still lists it. Risks that emerged or grew: tariffs, AI demand forecast accuracy ("Although AI is a relatively new demand driver for our products, it is evolving rapidly … we may incur costs in anticipation of demand that ultimately does not materialize"), capex execution across simultaneous greenfield builds in Idaho, New York, Singapore, Japan, and Taiwan, and CHIPS Act compliance conditions on share repurchases.

The most important new sentence in the FY2025 10-K is the one acknowledging that the company is now exposed to AI-demand forecasting error in both directions: "If such demand does materialize, but is lower than expected, we may not be able to reduce our costs in response." That is a textbook capex-cycle risk being committed to print at the same moment management is publicly raising capex.

4. How They Handled Bad News

The cleanest test was FY2023. Revenue had peaked at $30.8B in FY2022; by FY2023 it was $15.5B with a negative gross margin. Management's response, in order of how it was communicated: (1) cut wafer starts and capex hard, (2) write down $1.83B of inventory in one year, (3) cut headcount 15% and suspend bonuses, (4) disclose the China CAC ban and quantify the at-risk revenue band the same quarter it became known. None of this was hidden. The MD&A used the words "deteriorated sharply" and the call narrative repeatedly framed the situation as a cycle to be managed, not a structural problem to be denied.

The second test came smaller, in early FY2025. After a record Q1, management warned in the same call that bit shipments would step down in Q2 — "Our fiscal Q2 bit shipment outlook is weaker than we previously expected" — and walked the gross margin guide from a 300bps Q1 expansion to a sequential decline. NAND capex was cut, wafer starts reduced "by a mid-teens percentage." This was a real walk-back of the "FY25 will be linear improvement" implication carried out of Q4 FY2024, and management did it without a euphemism. The Q2 print proved the warning was right.

What management has not done well: the NAND segment never gets the same forensic treatment as DRAM. NAND has been "the part of the business that is still getting its legs back under it" or "where we are taking supply actions" for nine consecutive quarters. Until Q4 FY2025, when QLC mix and AI-driven SSD demand finally inflected, NAND margins consistently surprised to the downside without a granular root-cause explanation beyond "industry conditions."

5. Guidance Track Record

The pattern: every quarter for six straight quarters, revenue, gross margin, and EPS came in above guidance midpoint. The early prints (Q1–Q2 FY2025) were closely managed beats, around the high end of the range. The late prints (Q4 FY2025 onward) were misses to the upside large enough to call into question whether management is sandbagging on purpose or genuinely cannot forecast a shortage market — by Q2 FY2026 the actual gross margin (75%) was 700bps above the midpoint guide, and EPS came in $3.78 above the midpoint.

The bigger promises

Credibility score: 9 / 10. Every concrete, near-dated promise from the start of the AI cycle has been delivered or beaten. Bad news has been disclosed in the same call it became known (Q1 FY2025 demand walk-back, mobile managed NAND wind-down, NAND underutilisation). The single point off the maximum is for the FY2026 vintage of forecasts — TAM upgraded to "$100B by 2028" (two years earlier), capex raised three times in twelve months, and the share number Micron used to give every quarter ("HBM share commensurate with DRAM share") is now declined ("we will manage the mix between HBM and non-HBM"). All three are reasonable in a tight market, but each one removes a quantifiable yardstick the next downcycle will be measured against.

6. What the Story Is Now

The current story has three parts and they hang together only if memory has actually become structural.

De-risked, on the evidence so far. HBM execution has gone from probationary to dominant. Micron led 1-beta DRAM, led 1-gamma DRAM, hit HBM-share parity on schedule, and is shipping HBM4 to NVIDIA Vera Rubin in volume during CY2026. The data center business mix went from 16% of revenue at the trough to 56% in FY2025; the cloud memory business unit alone printed 74% gross margin in Q2 FY2026. The balance sheet is in the best shape in company history — net cash of $6.5B at Q2 FY2026, weighted average debt maturity of 2034, two recent credit upgrades to triple-B. Customers are signing five-year supply agreements with "specific commitments." None of this resembles the FY2023 commodity-cycle story.

Still stretched. Capex has gone $8B → $14B → $20B → "above $25B" in twenty-four months and is going up again in FY2027 with construction CapEx alone rising more than $10B year on year. Three greenfield fab projects (Idaho 1, Idaho 2, New York) plus a fourth (Tongluo, Taiwan, just acquired) plus Singapore HBM packaging plus Japan Hiroshima cleanroom expansion are under simultaneous construction. The supply-tightness story depends on AI demand growing roughly as forecast for the four to six years it will take this capacity to come online. The 10-K explicitly flags the mismatch risk; the script does not.

What to discount. TAM forecasts that compound (HBM "two years earlier than prior outlook"), the implied multi-year price duration, and assertions that this cycle is "structural" while the same risk factors that described the last cycle remain in print. Also discount the SCA framing until more than one is signed and the terms are independently visible — Sanjay has been explicit that they are "confidential" and "different from prior LTAs," but how different is unverifiable today.

What to believe. Sanjay Mehrotra's team has done what they said they would do at every checkpoint that has come due. The technology roadmap (1α → 1β → 1γ → 1δ; G8 → G9 → next; HBM3E → HBM4 → HBM4E) has been delivered ahead of, or in line with, plan for four straight nodes in DRAM and three in NAND. The FY2025 record was promised in September 2024 and the FY2026 record is being delivered now. If the capex bet is wrong, this is not a management team that will hide it — they did not hide FY2023, they did not hide the Q2 FY2025 demand wobble, and they did not hide the mobile managed NAND wind-down. The risk is not in management's honesty. It is in whether the demand curve they are building for is the curve that arrives.