FY2026: A Record Year and a Narrower Set of Tests

Thesis status: Affirmed on up-cycle evidence. The decisive test, a falling price, has not arrived.

Key Takeaways

1. Micron Technology ($MU) closed FY2026 with a record year built on price. Revenue rose 256% to $133.2B and non-GAAP gross margin reached 81.1%, while in 4Q26 DRAM prices rose in the high teens against bit growth in the mid-single digits. The fourth quarter beat its revenue guide by 8.5%. It also ran fourteen weeks against thirteen, which lifts its sequential and annual growth rates, a calendar effect the results materials do not mention.
2. Management guides industry supply and demand "much tighter" in calendar 2027 and 2028 than in 2026 and states that it does "not have line of sight" to a return to balance. Its 1Q27 guide is $61.5B of revenue, with sequential growth through FY2027 at a more moderate rate of price increase. These are management's forecasts. Retail DDR5, a public price updated monthly, set a new high in September and is consistent with them so far.
3. Twenty-six strategic customer agreements (SCAs) now cover close to 35% of Micron's DRAM bit supply through 2030, up from about a fifth at 3Q26, backed by about $150B of remaining performance obligations and $32B of customer commitments. The floors in those contracts have yet to meet a falling market.
4. The price now asks for less because guidance raised the revenue base. At the 5 October close it requires a flat EBIT margin of 42-48% from FY2027 to FY2035, below the 49.3% of FY2018, the best full year before the current boom, and well above the FY2015-25 average of 17.5%. On the 3Q26 model, at almost the same price, the requirement was 74.4%. Most of the fall comes from FY2027 guidance, so the lower requirement depends on revenue holding near that level, and the width of the range is a capital spending input that waits on the 10-K.
5. Ten watch conditions close into five, each read on one disclosed metric and to be adopted in a re-issued primer after the FY2026 10-K. The central test, whether gross margin holds above the prior-cycle peak once prices fall, needs a quarter of falling prices that has not yet come.

This is the second update to the March primer, after Memo No. 1 on 3Q26. It covers the fourth quarter and fiscal year to 3 September 2026, from the 8-K, results presentation and prepared remarks of 30 September, the post-earnings analyst call, and the reverse DCF re-based to the new guidance. The FY2026 10-K, due at the end of October, will be carried in the re-issued primer, including the maturity schedule of the remaining performance obligations, revenue by customer headquarters and the capital spending detail behind the one open model input.

FY2026 Results

The fourth quarter extended the year's price-led growth. Revenue rose 31% sequentially to $54.2B and non-GAAP gross margin rose 210 bps to 87.0%, while operating cash flow reached 81% of revenue as the receivables build that held back 3Q26 cash conversion slowed from $11.7B to $5.3B. Non-GAAP operating margin rose by less, 110 bps to 82.3%, because operating expenses grew by $1.05B in the quarter (Exhibit 5).

Price still led the quarter, but its rate of increase fell sharply. DRAM average selling prices rose in the high teens on bit shipments up in the mid-single digits, and NAND prices rose about 30% on bits up about 10%, after DRAM prices had risen by more than 60% in each of 2Q26 and 3Q26. The slowdown matches the meaningful moderation in the rate of price increases that management guided at 3Q26. Management attributes the gross margin gain primarily to higher pricing and execution, partly offset by mix. Mobile and client bit shipments fell for a second consecutive quarter while the unit's revenue grew 14% on price and mix.

The business unit that carries HBM was the only one whose margin did not rise. Cloud Memory (CMBU) held at 83% while the other three units gained between 260 and 470 bps, and management attributes CMBU's flat margin to a higher HBM mix offsetting higher pricing. HBM revenue grew faster than company revenue in the quarter, but Micron discloses no HBM revenue or margin, so the effect cannot be sized. The two data centre units, CMBU and Core Data Center, together made up 63% of 4Q26 revenue.

Seven items flatter, depress or reclassify the quarter, and three of them cannot be sized from the disclosure. The fourth quarter ran fourteen weeks against thirteen in 3Q26, so revenue per week rose 21.5% against the 30.8% headline. Micron also raised FY2026 incentive compensation for all employees and capitalised the manufacturing share into inventory. That cost reaches the income statement as the inventory sells, which management gives as the reason 1Q27 gross margin is guided below 4Q26 and expected to be the floor for FY2027.

Outlook: Guidance and Industry Supply

Management guides FY2027 to another record year, with 1Q27 as the low point for gross margin. The 1Q27 revenue guide of $61.5B is 13% above 4Q26, or 22% higher per week once the extra week in 4Q26 is removed, and gross margin is guided 75 bps lower at 86.25% as the incentive compensation capitalised in 4Q26 passes through cost of sales. Beyond 1Q27 management guides sequential revenue growth in every quarter and higher gross margins, and states that more than 75% of the year's shipments are already committed. Read literally, that wording puts FY2027 revenue above $246B, four times the 1Q27 midpoint. The figure is derived from management's wording, which gives no annual number.

Management's case for how long the shortage lasts rests on supply growing more slowly while demand broadens. It expects industry DRAM bit shipments to grow in the low-20s percent in each of calendar 2027 and 2028, down from the mid-20s in 2026, and names three limits on supply: smaller bit gains from each new node, HBM taking a rising share of output at rising trade ratios, and cleanroom space, which it calls the principal constraint. On demand, management points to CPU-run agentic workloads as a second source of server memory demand alongside accelerators, and states that customers are building systems with less memory than they would choose because supply is short. Server units are guided to grow in the high teens in both 2026 and 2027.

These are forecasts from the seller, and Micron's own framing has shifted. In March the CFO described the profitability of that time as "simply not sustainable". In September he spoke of continued price increases "albeit at a more moderate pace" and conditions "supportive through 2028". Two independent checks arrive before the next results: the September-quarter results of SK hynix and Samsung, due at the end of October, and the retail DDR5 price. That price is a consumer cheapest-listing basket, distinct from Micron's contract prices, and it moves sharply from month to month, but it is a public read on the commodity market that updates monthly, ahead of Micron's quarterly disclosure. It set a new high of $13.41 per gigabyte in September, 4.7 times its August 2024 level, after a dip in the spring that reversed.

The Contract Book

The contract book grew and lengthened on every measure Micron discloses on a comparable basis. Ten new agreements took the total to 26, spread across every business unit, and management now states coverage in bits: a little under 35% of DRAM supply and a little over 35% of NAND through 2030, against about 20% and 33% at 3Q26. Some new agreements run into 2031 and two existing ones have been extended to 2031. Micron names no counterparties and did not say on the call whether any of the ten new agreements are with hyperscale customers.

The pricing terms set what the book protects: three-quarters of expected agreement revenue has a defined pricing framework, mostly floor and ceiling bands, and the remaining quarter reprices periodically at market. Management states that margins at floor prices would stay meaningfully above any prior cycle peak. At 3Q26 it framed the floor in gross margin terms, and Memo No. 1 put the prior-cycle peak in the low-60s percent. Earlier agreements were priced on second-quarter 2026 conditions and newer ones on today's higher prices, so bands set in different quarters reflect different market levels. RPO counts only committed volumes at minimum prices, and management expects revenue to well exceed it.

The deposits fund the build and carry an obligation. Customers have committed $32B, most of it in cash, and $12.7B sat on the balance sheet at year end after $12.3B arrived in 4Q26. Micron books the deposits in financing cash flow, so free cash flow excludes them, and returns them toward the latter half of each agreement's term once minimum purchases are met. The $68.3B of net cash therefore includes money owed back to customers. The reverse DCF carries the remaining receipts and the returns on an assumed schedule until the 10-K gives more detail.

HBM and the Capital Programme

HBM trails conventional DRAM on margin for a reason that resets in January. Micron prices HBM by calendar year and has contracted most of its 2027 HBM supply at significant increases, which management states will narrow the profitability gap with conventional DRAM when they take effect. If the 2026 contracts followed the same cycle, they were agreed in 2025, before the commodity price increases that began late that year (Exhibit 8). The flat CMBU margin in Exhibit 4 is consistent with that lag. The same annual reset would slow HBM's response to a falling market, though no falling market has yet tested it.

The next HBM generation changes the test the primer set. The primer treated the HBM4E ramp as the first full test of Micron's 1-gamma node and packaging at scale. HBM4E, including the custom NVHBM product co-designed with NVIDIA, uses a base die made on a foundry process while its memory dies stay on Micron's own node, so part of its yield and timing now depends on an outside manufacturer. Management gave no HBM4 yield or ramp figure this quarter and expects its HBM share to stay near its DRAM share.

Most of the increase in FY2027 capital spending over prior plans is construction, mostly for cleanroom space that opens from late 2028 (Exhibit 11). Net capital spending rises to above $50B from $27.4B in FY2026, and gross spending already ran at 3.2 times depreciation in FY2026. Management states that it will equip the new cleanrooms in line with demand at the time and has signed long-term supply agreements with equipment makers, so construction commits the space while tool spending stays adjustable. Depreciation from the programme reaches the income statement as each facility enters service, from Singapore packaging in early 2027 and Idaho in mid-2027, and management names start-up costs as a partial offset to margins.

Micron is spending into the up-cycle alongside SK hynix (Exhibit 12). Both have raised capital spending by more than 70% in each of their last two reported years, while Samsung discloses capital spending only for the whole group, where its share of revenue fell. Management's supply outlook assumes this spending reaches output slowly because cleanroom space takes years to build. How fast the industry's new space converts into bits is the evidence the proposed supply condition will track.

What the Price Implies

This section sets out what the share price requires in cash-flow terms and offers no valuation, price target or view on the shares. At the 5 October close of $1,063.96 the market values Micron's operations at about $1.15 trillion after $68.3B of net cash, and the reverse DCF solves for the flat GAAP EBIT margin from FY2027 to FY2035 that would justify that value with every other driver on the model's base path. The answer is 41.8%, or 47.9% if capital spending after FY2027 is scaled up with the re-based revenue, an input that waits on the 10-K. Both sit below the 49.3% of FY2018, the best full year before FY2026, and well above the 17.5% average of FY2015-25 (Exhibit 13).

The requirement fell by 32.4 points between the 3Q26 and 4Q26 models while the share price moved by three cents (Exhibit 14). Moving FY2027 revenue to the guided level accounts for 29.6 points of the fall, the larger cash balance for 2.6 points, and the remaining updates to FY2026 actuals, the deposit schedule and the share count for 0.6. Each step changes one input on the model and recalculates it, so the steps describe this model's arithmetic in this order, and a different order would change their sizes.

The lower requirement rests on that revenue base. The model grows revenue from the guided FY2027 level at 10% in FY2028, tapering to 3% from FY2033, so it treats FY2027 as a level to build on. If FY2027 proves to be a peak and revenue falls back, the requirement rises again. Holding FY2027 at the model's 83% margin and solving for FY2028 to FY2035 alone gives 38.6-45.1%. At the base path's own margins, which average 37.3% over FY2027-35, the price instead needs revenue to compound at 19.5-23.7% a year to FY2035, against 12.0% on the path.

The discount rate moves the requirement by about 4.6 points for each percentage point, so the open capex input is worth about as much as 1.3 points of WACC. Peer multiples are left out of this memo because Micron's trailing twelve months now run to September and the peers' to June, in a market rising fast enough that a quarter's lag distorts the comparison, and the re-issued primer will compare them on matching periods once the peers report.

Closing the Framework

The ten conditions set in March were built to test whether Micron could institutionalise a supply advantage, and the year answered several of them. Data centre SSDs exceeded 60% of NAND revenue and 1-gamma became the majority of DRAM bits, so those conditions passed outright. Data centre bit share was never disclosed in the form its condition asked for, and China exposure is now expected in single digits, so neither tests anything material any longer. The gross margin bar and the capex warning are unlikely to fail while prices are rising, which leaves the thesis's central question without a test.

The five conditions that replace them each turn on one metric Micron discloses every quarter (Exhibit 16). The first, whether gross margin holds above the prior-cycle peak once prices fall, is the question the original thesis was built around, and it stays dormant until the first quarter of falling DRAM prices. The fifth is designed to signal that quarter early. The set will be adopted with baseline reads in a re-issued primer after the FY2026 10-K, and the conditions ledger records the final reads on the original ten from this memo.

What to Watch

The next ten weeks bring three reads: the peers' September quarters, the FY2026 10-K with the re-issued primer, and 1Q27 in mid-December, the first quarter tested against the new conditions (Exhibit 17). Beyond them, the January HBM price reset would first show in CMBU margin at 2Q27, and Idaho's first wafers in mid-2027 add the depreciation and start-up costs that new capacity brings.

Sources: Micron 8-K, results presentation, prepared remarks and post-earnings analyst call, 30 September 2026. Micron 3Q26 10-Q. MU reverse DCF model, 3Q26 and 4Q26 files, recalculated by Marvin Labs. Epoch AI retail memory prices, snapshot of 1 October 2026. SK hynix and Samsung Electronics annual reports. Margins are non-GAAP as Micron reports them, except in the reverse DCF, which uses GAAP EBIT. Adjusted free cash flow is Micron's measure: operating cash flow less capital spending net of incentives.