3Q26: The Supply Constraint, Now Under Contract

Thesis status: Affirmed in mechanism. Underwritten by contract. Priced beyond the model.

Thesis Refresher

The primer asked whether Micron Technology ($MU) could convert pricing power built on a structural supply constraint into durable business-model change before the cycle turns. The mechanism is the HBM wafer trade ratio, which diverts DRAM capacity into high-bandwidth memory and holds commodity supply scarce. The two initiatives meant to institutionalize that advantage before pricing normalized were node leadership on 1-gamma and the Strategic Customer Agreements. The horizon runs FY2026 through FY2028.

This is the first update to that primer. It covers 3Q26, reported 24 June 2026, the first quarterly results since publication. Two developments carry the period. The supply constraint moved from observed pricing into signed multi-year contracts, and the share price re-rated from roughly $330 at the primer to $1,207 after the print. The re-rating warrants a reverse-DCF treatment the primer did not carry. That analysis lives in this memo.

3Q26 in Brief

	2Q26A	3Q26A	Change
Revenue	$23.9B	$41.5B	+74%
DRAM	$18.7B	$31.3B	+67%
NAND	$5.0B	$9.9B	+99%
Non-GAAP gross margin	74.9%	84.9%	+1,000 bps
Non-GAAP operating margin	69.0%	81.2%	+1,220 bps
Non-GAAP diluted EPS	$12.20	$25.11	+106%
Operating cash flow	$11.9B	$25.4B	+113%
Free cash flow	$6.9B	$18.3B	+165%
Net cash	$6.5B	$24.5B	+$18.0B

Source: Micron 3Q26 results release and earnings call, 24 June 2026, Marvin Labs

2Q26 figures derived from disclosed sequential growth rates. Margins and EPS non-GAAP.

Export

3Q26 is a pricing event, not a volume event. DRAM bit shipments grew 2.5% sequentially and NAND 5.0%, while DRAM ASPs rose in the low-60s percent and NAND ASPs in the mid-80s. The gross margin expanded on price, not on cost or volume. Management guided 4Q26 to approximately 86% gross margin with a meaningful moderation in the rate of price increases, which dates the surge to this window. Whether these economics survive the surge is the question the rest of the memo tests, and it is the question the Strategic Customer Agreements are built to answer.

The SCA Floor

The supply constraint is now written into contracts. At the primer, Micron had signed one Strategic Customer Agreement. At 3Q26 it disclosed sixteen, and put a number on what they lock in: roughly $100B of remaining performance obligations at floor prices, with $22B of customer deposits and financial commitments behind them.

Source: Micron 3Q26 earnings call and 10-Q commentary, 24 June 2026, via Marvin Labs. RPO stated as the minimally enforceable amount under ASC 606.

The floor is the load-bearing term. Management stated the floor prices are set to hold gross margin well above the company's peak quarterly margin in any past cycle. That prior peak sat in the low-60s percent. The remaining performance obligation is calculated on minimum committed volumes at those floor prices, the minimally enforceable amount under the accounting standard, which makes the $100B a contractual minimum rather than a forecast. The ceilings cut the other way, but they apply only to existing products at the current market price. New products carry separately negotiated premiums, so the structure caps commodity DRAM upside without capping the HBM and next-generation roadmap.

This reshapes the downside, not the upside. In prior memory downturns ASPs fell faster than cost and gross margin turned negative, as it did on a GAAP basis in FY2023. With roughly 40% of revenue moving to fixed or capped pricing once all planned agreements execute, and ~$100B of volume floored at margins above the prior cycle peak, a comparable downturn now meets a contractual floor under a large share of the book. This is the mechanism the primer named: converting pricing power built on a supply constraint into durable business-model change before the cycle turns. The disclosure is that change taking contractual form. The test it has not yet faced is a genuine demand downcycle.

The agreements also pre-fund the working capital of the boom. Customers committed $22B, $18B of it cash, with $400M received in 3Q26 and approximately $10B expected in 4Q26, returning to customers weighted toward 2029 and 2030. The $11.66B receivable build that pulled 3Q26 operating cash flow below net income is partly funded by those deposits. The customers underwriting the revenue are also financing the balance sheet that carries it, at a cost set years out.

What the Price Implies

This section characterizes what the current share price requires in cash-flow terms. It is not a valuation or a price target. The watch conditions test whether the business is navigating toward what the price embeds.

	Value
Share price (25 June 2026)	$1,207
Diluted shares	1,149M
Market capitalization	$1,387B
Net cash	$24.5B
Enterprise value	$1,362B
FY2026E free cash flow	$66B
WACC	13.4%
Terminal growth	2.5%

Source: MU reverse-DCF model, Structural Floor sheet, Marvin Labs

WACC is the cost of equity given the net-cash position: a 4.5% risk-free rate plus a Blume-adjusted beta of 1.78 against a 5.0% equity risk premium.

FY2026E free cash flow is an estimate.

Market capitalization rounds the price times share count.

Export

The base case carries the FY2026 surge and then lets EBIT margin compress to the 27% terminal level the primer assumed. Stressing that result the conventional way, by flexing the discount rate and terminal growth, leaves the compression in place and does not close the distance to the market. The assumption the SCA floor actually challenges is the compression itself. The grid below removes it, holding EBIT margin flat across the whole ramp after FY2026. This is the most generous honest framing of the bull case, and a compressing path sits below every cell in it.

WACC \ Flat EBIT margin	40%	45%	50%	55%	60%
10.4%	$723	$844	$964	$1,084	$1,204
11.4%	$642	$747	$853	$958	$1,064
12.4%	$577	$671	$764	$858	$952
13.4%	$524	$608	$692	$776	$861
14.4%	$480	$556	$633	$709	$785
15.4%	$443	$513	$582	$652	$721
16.4%	$412	$475	$539	$603	$666

Source: MU reverse-DCF model, Structural Floor sheet, Marvin Labs

Base WACC row (13.4%) in bold. Base case, with margin compressing to 27%, implies $379. Current price $1,207.

Export

At the base discount rate, no flat-margin assumption through 60% reaches the price. At a 2.5% terminal growth rate, no cell in the grid does. The market is cleared only by stacking the lowest defensible discount rate against the highest flat margin, and then lifting terminal growth beyond the rate above.

The gap is structural, not a discount-rate artifact. The price embeds a regime change: that the SCA floor has taken the cycle out of memory margins, and that the resulting earnings warrant the discount rate of a structural compounder rather than a semiconductor cyclical. The reverse-DCF frames that claim. It cannot underwrite it.

The base trajectory rests on the load-bearing conditions. The table maps where a failure truncates the cash flows the price assumes.

The price implies none of these bite within the horizon.

Watch Condition Assessment

Key. Conditions carry the primer's framing: C marks a confirming signal (the thesis is working), W a warning signal (the thesis is breaking). Tier is the thesis's dependence on the condition: a load-bearing failure breaks the thesis, an amplifying one sets the scale of the outcome. Status denotes thesis impact, not signal direction, so a warning that is not firing reads green. 🟢 Affirmed tracks at or ahead of what the thesis needs, 🟡 Developing is mixed or in progress, 🔴 Weakening tracks against it. Trend is movement since the primer: ↑ improved, → unchanged, ↓ deteriorated.

C1: Gross margin sustained above 75% during consumer unit declines

🟢 Affirmed ↑ The margin held above the bar in the exact quarter volume softened.

The condition tests for pricing power that survives weak unit demand, not pricing power in a broad upcycle. 3Q26 supplied that test. Mobile and client revenue rose on higher pricing while bit shipments in the segment fell, and the consolidated non-GAAP gross margin still reached 84.9%. Management guided 4Q26 to approximately 86% with a slower rate of price increase, which holds the structural-margin case for another quarter without relying on continued ASP acceleration.

C2: Additional Tier-1 SCAs beyond the 2Q26 agreement

🟢 Affirmed ↑ One agreement became sixteen, with a contractual floor disclosed.

This is the period's defining development and is treated in full under The SCA Floor. For the condition, the relevant facts are the count and the structure: sixteen agreements covering four very large customers, roughly $100B of remaining performance obligations at floor prices, and floor margins set above the prior cycle peak. The primer asked for evidence that the supply-constraint advantage was being institutionalized. The contract structure is that evidence.

Outside the filing, the demand side drew corroboration. An Anthropic agreement was reported as the stock set a new high two days before the print (Yahoo Finance, 23 June 2026). Management named no SCA counterparty, so the report cannot be tied to a specific agreement, and it is read here only as consistent with the hyperscaler demand the SCAs convert into contract.

C3: HBM4 yield ramp tracking faster than HBM3E

🟢 Affirmed ↑ HBM4 12-High is ramping at roughly twice the HBM3E pace, with over $1B already shipped.

The yield read is the cleanest of the quarter. Management stated the HBM4 12-High volume ramp is tracking twice as fast as HBM3E 12-High, and that mature yields will arrive significantly faster than the prior generation. Cumulative HBM4 revenue has already passed $1B. The execution risk the primer flagged on this node has not appeared in the first test.

The forward read is the bridge from yield to scaled revenue, and it carries the platform dependency the primer named. HBM4 is in high-volume shipments for what management calls the lead customer's platform, with qualification samples shipped to multiple end-customers. Management did not name NVIDIA ($NVDA) or Rubin. The disclosures that frame the bridge are below.

Source: Micron 3Q26 press release and earnings call, 24 June 2026, via Marvin Labs.

The bridge is supply-disciplined rather than volume-maximizing. Management stated it is holding HBM bit share close to its DRAM share precisely because of the trade ratio, since every incremental HBM bit consumes disproportionate wafer capacity and tightens the commodity supply that the margin thesis depends on. The concentration the primer warned about remains: a single lead-customer platform anchors the HBM4 ramp, and the company will not name it. The yield is affirmed. The dependency is unchanged.

An off-cycle report sharpened that dependency during the quarter. Korean media reported that Nvidia's Rubin platform is expected to cut production and that SK Hynix is slowing its HBM4 capacity expansion (Moomoo, 24 June 2026). The primer's framework treats unverified third-party reports on platform status as noise, and this one is uncorroborated. If it holds, it cuts both ways. A Rubin production cut would defer the HBM4 ramp the lead-customer platform anchors. A competitor slowing HBM4 capacity tightens the supply the trade-ratio thesis depends on. The report is logged as a watch item, not a thesis input.

W5: HBM trade ratio narrowing

🟢 Affirmed ↑ The ratio is widening with each generation, not narrowing.

The warning tests for any sign the wafer trade ratio is compressing, which would relieve the commodity supply scarcity underpinning the thesis. Management stated the opposite. The trade ratio increases with every new generation, through HBM4, HBM4E, and HBM5. This is the structural foundation of the supply constraint, and the disclosure reinforced it rather than eroding it.

C4: Data center bits exceeding 50% of industry TAM

🟡 Developing → The metric was not disclosed in the form the condition specifies.

Management did not quantify data center bits as a share of industry TAM. The directional evidence is strong: data center revenue exceeded $25B, a run-rate above $100B annualized, and industry data center DRAM and NAND bit shipments are expected to more than double versus two years ago. The condition as framed is not directly measurable from this disclosure, which is the basis for the proposed refinement in Thesis Standing.

C5: Enterprise SSD mix reaching 60%+ of NAND revenue

🟡 Developing ↑ Climbing fast, still below the threshold.

Data center SSD revenue exceeded $5B and more than doubled sequentially, reaching approximately 50.5% of NAND revenue. The trajectory is toward the condition. The level is not yet there. The 60% threshold remains the bar, and the quarter moved toward it without clearing it.

W1: Inventory days rising from the 123-day level

🟢 Affirmed ↑ Days fell rather than rose.

Days of inventory ended 3Q26 at 120, down from the 123-day baseline. The warning is not firing. The decline came against record shipments, which is consistent with demand outrunning supply rather than inventory accumulating.

W2: 1-gamma bit mix missing its mid-2026 majority target

🟡 Developing ↓ The majority claim softened.

At 2Q26 management guided 1-gamma to a majority of DRAM bits in the second half of calendar 2026. At 3Q26 the language moved to 1-gamma being on track to become the highest-volume node in company history, without reiterating the majority target or its timing. The node is ramping. The specific commitment the warning tracks was not restated, which is the kind of timeline softening the primer defined as a weakening signal.

W3: China exposure broadening beyond the baseline

🟡 Developing ↓ The disclosure that measured this condition was withdrawn.

Neither the 3Q26 press release nor the call quantified China-headquartered customer revenue. The last specific figure, approximately mid-teens percent of revenue, dates to 1Q25. Management referred only to trade and geopolitical developments being excluded from guidance. The condition cannot be read as broadening or contained, because the data that tested it is no longer disclosed. This is handled under the Framework Modification in Thesis Standing.

W4: FY27 capex rising without corresponding revenue

🟢 Affirmed → Capex is stepping up, and revenue is stepping up faster.

FY27 capex is guided to step up meaningfully, with quarterly spend above the 4Q26 level and an implied annual figure above $40B, more than half of the increase in construction. The warning fires only if that spend outruns the revenue to support it. FY2026 revenue is tracking to roughly $129B, supply-demand conditions are guided tight beyond calendar 2027, and the SCAs commit volume against the new capacity. Capex intensity rising toward 21% of revenue is the figure to watch, but the spend is matched by contracted demand rather than running ahead of it. Off-cycle coverage during the quarter framed the New York fab as central to U.S. AI memory capacity (Yahoo Finance, 24 June 2026), which is the construction the FY27 step-up funds.

Thesis Standing

The thesis is affirmed and harder than it was. Every load-bearing condition is green. The supply constraint the primer treated as observed pricing power is now written into sixteen contracts with a floor beneath roughly $100B of revenue. The primer asked whether Micron could institutionalize the advantage before the cycle turned. One quarter into the horizon, the contractual form of that institutionalization exists.

A new fault line runs alongside the affirmation, and it is not operational. The business is converting pricing power into durable structure on schedule. The price has moved past what that structure can support. The reverse-DCF located the gap precisely: the market clears only at a margin and discount rate that together imply memory has stopped being cyclical. The watch conditions now carry a second duty. They track the business, and they track whether the regime the price assumes is materializing. The first is going well. The second stays unproven until a downcycle tests the floor.

The single most important variable is unchanged from the primer. It is whether the SCA floor holds when demand softens. Everything marked green in this memo was recorded in an up-cycle. The floor's value is a downside property, and the downside has not arrived.

Two further adjustments are logged for the fiscal-year restatement, not adopted here. C4 is framed around data center bits as a share of industry TAM, a number management does not disclose and did not provide. The proposed refinement re-anchors it to the data center revenue run-rate, above $100B annualized, which management does report. And the SCA structure has become central enough to warrant its own condition. A candidate addition would track the share of revenue under floor or capped pricing, the metric that most directly measures the institutionalization the thesis turns on.

What to Watch

4Q26 results and the FY2026 10-K, late September 2026

The next quarterly read and the first full-year filing. The 10-K carries the $100B RPO 12-month maturity split the 3Q26 disclosure deferred, which turns the contracted floor from an aggregate into a schedule of near-term protected revenue. 4Q26 is guided to approximately 86% gross margin on a slower rate of price increase. The first quarter where the rate of price increase moderates is the first read on whether the level holds without continued acceleration.

The first sign of demand softening

The floor thesis is a downside property, and every green mark in this memo was recorded in an up-cycle. The test is not another record quarter. It is the first quarter where bit demand or pricing softens, when the SCA floor either holds gross margin above the prior cycle peak as management states, or it does not. That quarter has not arrived. It is the one that matters.

China disclosure (W3)

Under the reframed condition, watch for the reappearance of a China revenue figure, any Cyberspace Administration review development, or a shift in management's characterization of China demand. The number's absence is the current signal. Its return, at whatever level, is the next.

1-gamma majority and the HBM4 platform

Whether 1-gamma reaches the majority of DRAM bits the second-half calendar 2026 target implied, against the softened 3Q26 language (W2). On HBM, the markers are HBM4E volume production in calendar 2027, the Singapore packaging contribution from the first half of 2027, additional end-customer qualifications, any naming of the lead-customer platform, and corroboration or denial of the Rubin production report that surfaced around the print.

The contract is signed. The cycle has not yet asked it to pay out.

Source: Micron 3Q26 results release, earnings call, and 10-Q commentary, 24 June 2026, via Marvin Labs. Prior-period comparatives from Micron 2Q26 results and 1Q25 disclosure. Off-cycle items from the 24 June 2026 daily news digest: Anthropic agreement and New York fab coverage (Yahoo Finance), Rubin production and SK Hynix HBM4 capacity report (Moomoo).