Records at the Peak, Capital Pouring In: Micron's "Escape the Cycle" Thesis Enters Its Falsifiable Window

On September 4, Micron (MU) closed at $1,003.22 with a market capitalization of roughly $1.15T, up more than 200% year to date and sitting in all-time-high territory; the immediate catalyst for the day's 4%-plus gain was the re-rating of the compute chain triggered by a next-generation model release. Behind the tape sit a rare set of results and an even rarer guide: FQ3-26 (fiscal quarter ended 2026-05-28) revenue of $41.46B, +346% year over year; non-GAAP gross margin of 84.9%, close to 10 points above the prior quarter's 75.0%; and FQ4 guidance of $50.0B±1.0B in revenue at roughly 86% gross margin. This is the strongest single quarter we have seen since we began covering the memory chain. In that same quarter, Micron's non-GAAP operating margin (81%) exceeded both Nvidia's (66%) and TSMC's (60%) — for the first time in history, the margin of this commodity memory business stands above those of the monopoly designer and the foundry downstream of it.

Our core judgment is this: the near-term demand pulse and allocation-driven price increases will most likely keep coming through, and there is little suspense about the $50B in FQ4. But the current price already assigns roughly 70% credence to the proposition that "HBM and contractualization have let memory escape the cycle" (the pricing decomposition follows below), while the variable that actually renders the verdict is the supply response in 2027-2028 — and evidence on that side is accumulating quickly. From here, the proposition enters its falsifiable window. This note lays out the market's pricing, the supply-side evidence, and a set of time-stamped verification criteria.

Revenue +74% Quarter over Quarter, About 95% of It from Price, With Bits Barely Moving

Start with the structure of the price cycle. The FQ3 decomposition: DRAM revenue of $31.3B, with bit shipments up only low single digits sequentially and ASP up just over 60% sequentially; NAND revenue of $9.9B, with bits up mid single digits and ASP up in the mid-80s percent sequentially. Revenue rose 74% sequentially, almost entirely on price. This is a pure price cycle under allocation, structurally identical to 2017-18 but more extreme in magnitude. Price is the spring; volume is the foundation.

The marginal momentum in price has to be stated alongside it. Our early-September note on the A-share memory chain already confirmed that four independent spot-price measures decelerated in the same direction in August (DDR4 spot, month over month: +2.5% in June, +16.9% in July, +5.4% in August), while Micron's reported results and guidance reflect contract and allocation prices locked in two to six months earlier. The reported numbers will most likely look even better for another quarter; the curve that the marginal price-setter watches has already turned first. The weekly observation item from here is the week-over-week change in DDR4/DDR5 spot: two consecutive negative weeks upgrade deceleration to a top; two consecutive weeks of re-widening give the price-increase narrative another lease on life.

Sixteen Take-or-Pay Agreements: Contractualization Changes the Shape of the Cycle, Not Its Existence

The hardest incremental point in the bulls' hands is the reconstruction of the business model: 16 strategic customer agreements (SCAs), take-or-pay in structure, running CY2026-2030, covering roughly 20% of DRAM bit volume and one third of NAND volume over the period; 14 of them carry cumulative revenue of roughly $100B at contract minimum prices; customer cash deposits and financing commitments total $22B; and per management, the floor prices on the largest few contracts correspond to gross margins "above the quarterly peak of any historical cycle." Once the full plan is in place, roughly half of company revenue becomes contractualized. Floor prices plus deposits genuinely raise the bottom of the downside, and that structure did not exist in the last cycle.

The other side of the coin deserves attention. The price on roughly 40% of revenue is fixed or capped near CQ2-26 market levels: if prices keep rising in 2027-28, that upside elasticity has already been sold forward. The $22B of deposits against more than $100B of five-year commitments is roughly 20% coverage; and the take-or-pay history of optical communications in 2001 and iron ore in 2012 says that once spot falls deep below the floor, contracts are either renegotiated or defaulted on. The contracts redistribute price risk; they have not eliminated a single ton of capacity. The first test point is the FY26 10-K (expected from late September): if the gross margin corresponding to SCA floor prices is disclosed for the first time and is no lower than 65%, that is a credit to the structural-improvement case; below 60%, and the verbal characterization has not been delivered on.

Capital Pouring In on Four Parallel Tracks, With Output Concentrated in a 2H27-2028 Arrival Window

The first reading in capital-cycle analysis is where the money is going. Micron itself: net capex goes from $13.8B in FY25 to roughly $27B in FY26, with FY27 guided to "above FQ4's $10B every quarter" — an annualized run rate above $40B, a tripling across two fiscal years, with more than half of the increment going to cleanroom construction. SK hynix: $38.3B of HBM-dedicated investment scheduled out to 2031. Samsung: P4's 1c conversion exceeding 200k wpm, with P5 pulled forward by six months and volume production in 2028. Per SEMI, Korean fab investment goes $27B (2025) → $35B (2026, +29.6%) → $38B (2027E); industry capex for 2027-28 combined exceeds $250B.

New entrants are at the door as well. CXMT's capacity expands from 180k wpm (2025) to 300k (2026, equivalent to roughly 13% of global DRAM wafer capacity), with 500k planned (2028); its A-share IPO raised RMB 66.6B, the largest of the year; HBM3 is planned at 60k wpm with overall yield of roughly 25%, and HBM3E has already entered risk production. A wave of equity financing at the top of a cycle has always been the loudest bell on the supply side.

In our view, the 2.5-to-3-year physical lag from groundbreaking to output at a greenfield site explains two things at once: why the shortage can still run another two to six quarters, and why 2028 has become the inflection year that industry participants and the sell side name in unison. The current position looks more like mid-2017, some distance from the top; but every month of record profit places one more order for capacity arriving in 2028. Management's behavior is more honest than its rhetoric: maintaining the language of "supply discipline" on one side while tripling capex on the other. Within the capital-cycle framework, that is precisely the self-evidence that current returns are unsustainable.

What the Market Has Bought Into: Roughly 20% Compounding × Roughly 30% Margin × A Decade Without Drawdown

Solving back from the early-September closing price (reverse DCF: WACC 9.5%, terminal growth 3%, ten-year window, with growth in years 6-10 decaying linearly to the terminal value), the current price implies a 21.57% revenue CAGR over the next five years and a 28.99% steady-state free cash flow margin (on a GAAP FCF basis; roughly 33.1% on an adjusted FCF basis). The solved values are sensitive to WACC and to the price date chosen, so reading them as roughly 20% and roughly 30% is the more honest framing. Set that against two anchors: Micron's historical mid-cycle FCF margin of roughly 10%, and the current peak of 44% (FQ3, adjusted basis). What the market is pricing is an intermediate state — revenue compounding gently from a post-surge plateau, with margin parked permanently at roughly three times the historical mid-cycle level. Extrapolate $50B a quarter directly, or mean-revert to the historical midpoint, and neither solution reconciles with the current price.

The "roughly 70% credence" calibration comes from exactly here: if "escaping the cycle" is entirely false, the steady-state margin returns to the roughly 10% midpoint; if it is entirely true, it stays at the current 44% peak. The steady-state margin solved back from the current price (roughly 33.1% on the adjusted basis) sits about two thirds to three quarters of the way along that range.

The structure on the consensus side deserves its own sentence: of 48 covering institutions, 44 rate it Buy and none rate it Sell, with a target-price band of $361-2,200, a 6x spread. Zero bears coexisting with a 6x spread says the sell side is caught in the gap between the two narrative frames of "keeps growing" and "cycle crashes," with no one pricing the path in between. This does not contradict the "intermediate state" of the previous paragraph: the current price, as the probability-weighted mean of the two-sided narratives, lands mechanically in the middle; but not a single firm is underwriting plateauing as its base case. An expected value that lands in the middle, and someone actively pricing the middle, are two different things. The cheap multiple is also an illusion: FY27E consensus EPS of roughly $155 implies a 6.5x forward P/E, but at Micron's June 2018 top the trailing P/E was roughly 5x, and the stock fell 55% over the following six months. Cyclicals are cheapest at the top; that is an old lesson.

Scenario Distribution: The Roughly 45%-Probability Plateau Path That No One Is Pricing

We are writing our scenario assessment down in advance here, in three buckets, so it can be reconciled when the time comes. A full mean-reversion scenario (supply arrival compounded by a pullback in hyperscaler demand) carries roughly 30% probability. AMZN's Q2-26 FCF of -$7.6B and META's FCF down more than 90% year over year are, for now, just the other face of high-intensity capex investment; for that to become demand destruction requires one further step of transmission — FCF pressure forcing 2027 capex growth to be revised down from the consensus +50% (2026E consensus is +98%). The verification-calendar item "capex consensus growth revised down to below +20%" is watching exactly that step. The plateau scenario — in 2027, with all three HBM suppliers in place, price competition begins, commodity prices top out and retreat, but demand does not collapse and the SCA floors hold, with margin converging from above 80% to a 50-60% plateau — carries roughly 45% probability; in that scenario the current price is roughly fair and upside is limited. Shortage extending past 2028 with "escaping the cycle" fully validated carries roughly 25% probability.

The fair thing to say to the bulls: producer inventories stand at 2-3 weeks for DRAM and 4-5 weeks for NAND, far below the 7-9 weeks normal for the late cycle; industry bit growth in 2026 still runs below demand; HBM4 is already shipping in high volume to a leading customer platform with cumulative revenue above $1B, making Micron the only company in the industry to quantitatively disclose HBM4 revenue; and net cash of $24.4B leaves the balance sheet at its strongest in company history. The clock for caution runs in 2027-28, not next quarter; until then, the reported numbers will keep looking good.

Proposition-Level Falsification Criteria: "Escaping the Cycle" Holds Only If All Three Hold

First, price: in 2027, with Samsung, SK hynix and Micron all ramping and CXMT's HBM3E entering, HBM4/4E contract prices decline no more than 15% year over year. Second, margin: all three companies' HBM gross margins stay above 60% through full-year 2027. Third, linkage: if commodity DRAM goes into oversupply in 2028, HBM is not dragged into linkage by the cross-subsidy mechanism of a shared capacity pool; the observation proxy is the quarterly ratio of HBM to commodity DRAM contract prices, and two consecutive quarters of narrowing during commodity-oversupply quarters counts as a return of linkage and the failure of this criterion. If any one fails, the proposition is downgraded to "a cycle with narrower amplitude." Our assessment is that the probability of all three surviving together into 2028 is roughly 25%, against market pricing that corresponds to roughly 70% credence. That gap of forty-odd percentage points is the core expectations gap of this note; the direction favors the cautious, but the clock for realization runs in 2027.

Verification Calendar

To close: Micron's reported results sit at a record peak, the market has already paid roughly 70% of the price for "escaping the cycle," while capital is pouring into the supply side on four parallel tracks, with output arriving in concentrated fashion in 2H27-2028. The three proposition-level criteria (HBM contract prices year over year, the 60% line on HBM gross margin, and commodity linkage) are the failure sensors for this story; our position is already written down at 25%, and the criteria will render the verdict automatically when their time comes.

The next observation point is the FQ4 results in late September and the FY26 10-K that follows: watch the slope of guidance, and watch whether the gross margin at SCA floor prices is disclosed. The 4Q contract prices settling in October-November deliver the final ruling on the price side. Under the same capital-cycle framework, our deep read on SanDisk's NAND is the sister piece to this note, and the two should be read against each other.

This note represents the research views of LunarTulip Research, based on analysis of publicly available information, and does not constitute investment advice. Markets carry risk; decisions must be made independently.