Micron’s beat was small; the mix reset is the real earnings power

The print matters because the surprise was not the top line; it was the conversion. What was priced in was a record quarter, visible AI memory demand, and enough DRAM tightness to support a revenue number near consensus. What actually surprised was that Micron Technology turned a revenue beat of only +0.9% into an EPS beat of +5.9%, with non-GAAP EPS at $3.03 versus the Street’s $2.86. That gap is the tell. Investors were not paid for discovering that AI memory demand exists; they were paid for discovering that the incremental dollar is landing at a much better margin than the old memory cycle would imply. The thesis from this print is therefore not “Micron beat.” It is that the company’s earnings algorithm has moved faster than revenue revisions, and the next debate should be how much of Q4’s 44.7% gross margin is cyclical scarcity versus a mix-driven floor.

That distinction is essential because the headline revenue result was almost exactly what the market had modeled. Actual revenue was $11,315.0 million against the Street’s $11,217.0 million, a +0.9% surprise that should not by itself re-rate a memory stock. Yet EPS of $3.03 beat by +5.9%, which means the quarter’s information content sits below the revenue line. Management’s own account reinforces the same point without needing to stretch the numbers: Mark Murphy said Q4 revenue was “$11.3 billion, up 22% sequentially and up 46% year over year,” while non-GAAP diluted EPS reached “$3.03, with 59% sequential growth and a 157% increase versus the year-ago quarter.” The reason to quote that wording is the asymmetry it exposes: revenue grew rapidly, but earnings grew much faster, and the company is inviting investors to focus on operating leverage rather than unit recovery alone.

The financial trajectory supports that reading because this is no longer a simple rebound from trough losses. Revenue has moved from the post-downturn base of $3.75 billion to $11.31 billion in the Q4 print, while gross margin has swung from -17.8% to 44.7%. That scale of margin recovery would usually make a semiconductor PM worry about peak-cycle extrapolation, but the composition here is different from prior DRAM upcycles. Sanjay Mehrotra put the mix change in unusually concrete terms: combined revenue from HBM, high-capacity DIMMs, and LP server DRAM reached “$10 billion, more than a fivefold increase compared to the prior fiscal year.” The phrase matters because it identifies the products doing the margin work; this is not a generic price-cycle claim hiding behind industry language.

The capacity and mix story explains why the guide is more important than the Q4 beat. Management guided fiscal Q1 revenue to $12.5 billion, plus or minus $300 million, and gross margin to 51.5%, plus or minus 100 basis points. Those guideposts are the cleanest evidence that Q4’s 44.7% gross margin is not being treated internally as a peak print. If the company were merely harvesting a short squeeze in commodity DRAM, the margin guide would be more vulnerable to volume normalization; instead, management is guiding a step-up while also taking operating expenses to approximately $1.34 billion. The operating-expense guide matters because it absorbs some of the scaling benefit, yet the implied gross-margin direction still points higher, which is why the earnings power debate should move from “can revenue beat” to “how durable is the 50%-plus gross-margin regime.”

That durability rests on data center mix, and here the numbers are too large to dismiss as narrative. In fiscal 2025, the data center business reached 56% of total company revenue with gross margins of 52%, compared with total fiscal 2025 gross margins of 41%. That spread is the margin-reset argument in one comparison. Cloud memory business unit revenue was $4.5 billion in Q4 and represented 40% of company revenue, while core data center added $1.6 billion and represented 14%. The market may be missing that those two buckets together are no longer an adjacency; they are the earnings center of the company. Mobile client at $3.8 billion and automotive and embedded at $1.4 billion still matter, but they are not the reason Q4 EPS rose faster than revenue.

The customer read-through is therefore specific rather than broadly “AI positive.” For NVIDIA, the signal is that Micron’s HBM and server DRAM ramp has already created nearly $2 billion of Q4 HPM revenue, implying an annualized run rate of nearly $8 billion, which supports continued memory-content intensity in accelerator systems. For AMD, the relevant point is that Micron’s DDR5 DRAM and HBM3E exposure is scaling inside the same data-center pool that reached 56% of Micron revenue. For Apple and Intel, the second-order read is less favorable: mobile client and PC-adjacent demand is being monetized inside a company now allocating investor attention and likely scarce leading-edge bits toward higher-margin data center uses. Nanya Technology, linked to Micron through DRAM technology licensing, is a different case: the print validates DRAM economics, but Micron’s mix-led 52% data center gross margin sets a bar that commodity-license exposure alone does not automatically clear.

The supplier read-through is equally concrete because Micron is spending through the cycle rather than merely talking about future capacity. The company invested $13.8 billion in CapEx in fiscal 2025, and Q4 capital expenditures were $4.9 billion. That keeps the order funnel relevant for ASML in EUV and DUV lithography, Applied Materials in etch and deposition, Lam Research in DRAM and NAND process intensity, and KLA in inspection and metrology. It also supports memory-test demand at Advantest, because HBM and high-capacity server modules raise test complexity rather than just wafer volume. The packaging read-through is narrower but still material: Hanmi Semiconductor is tied to TC bonders, with about 50 units ordered, and that line item fits the HBM mix shift better than a generic wafer-capacity story.

The cash-flow profile tempers the most aggressive bull case, but it does not break the thesis. Q4 operating cash flow was $5.7 billion, yet free cash flow was only $803 million because capital expenditures were $4.9 billion. For the full fiscal year, free cash flow was $3.7 billion, representing 10% of revenue. That is not a capital-light AI beneficiary profile, and PMs should not value it like one. The variant perception is more precise: Micron is still capital intensive, but the return on that capital is being pulled higher by data center mix. Debt reduction of $900 million in Q4 shows management is not ignoring the balance sheet, but the investment case cannot rest on buybacks or cash return; it rests on whether the next $13.8 billion of annual CapEx earns data-center-like margins rather than commodity-memory margins.

The peer context makes the same point from another angle. In the latest reported memory peer set, MU shows revenue of $41.46 billion, gross margin of 84.6%, and revenue YoY of +345.7%. Those figures are not the same reporting basis as the Q4 FY2025 print, so they should not be mixed with the Street beat, but they do frame where investors are likely to argue about normalized profitability. Compared with SNDK at $5.95 billion of revenue and 78.4% gross margin, MU’s latest peer-table profile reads less like a lagging memory beta and more like a scale beneficiary of the AI memory shortage. Compared with WDC at $3.34 billion of revenue and 50.2% gross margin, the gap underlines that not all storage and memory exposure is being rewarded equally by the current demand stack. The market’s mistake would be to flatten those differences into a single “memory is up” trade.

The call delivery complicates the interpretation, but it complicates it in a useful way. The current tone history shows sentiment at 0.55 and guidance_tone at 0.50 for Q4 FY2025, with qa_sentiment at 0.54. That is constructive, but not euphoric. More interestingly, prepared_sentiment was only 0.02, while qa_evasiveness was 34.5. The numbers say management’s scripted language was restrained even as Q&A held up, which fits a company trying not to overpromise into a supply-constrained AI memory ramp. This is why the guide carries more weight than adjectives: the transcript tone is not screaming upside, but the gross-margin guide is.

That delivery pattern should keep investors from overpaying for a straight-line extrapolation. The tone series later shows guidance_tone rising to 0.72 while sentiment falls to 0.46, and uncertainty rises to 42.6. Those conflicting numbers matter because they describe a management team that can sound more confident about formal guidance while the overall transcript becomes less clean. For this event, the investable read is that Micron gave enough numerical commitment to support the margin-reset thesis, but not enough tonal comfort to remove cycle risk. In other words, the call does not justify a blank-check multiple; it justifies raising the probability that gross margin sustains above the old-cycle range if HBM and server DRAM stay supply constrained.

The NAND line is the main offset to the bull case, and it should not be hand-waved away. Q4 DRAM revenue was $9 billion, up 69% year over year, and represented 79% of total revenue. NAND revenue was $2.3 billion, down 5% year over year, and represented 20% of total revenue. That split is favorable for margins today, but it also means the thesis is increasingly dependent on DRAM and, within DRAM, data center and HBM. Fiscal 2025 NAND revenue still reached $8.5 billion, up 18% year over year, so this is not a broken business. It is simply not the part of Micron setting the multiple. A memory PM should treat NAND recovery as optionality, not as the core reason to own the stock after this print.

The operating model also shows why the EPS beat was not just a tax fluke, though taxes helped. Operating income was $4 billion in Q4, with operating margin of 35%, and operating expenses were $1.2 billion. Fiscal Q4 taxes were $471 million on an effective tax rate of 12%, which management said was lower than guidance due to discrete items. That means some EPS upside was below the operating line, but the larger point is that operating margin expanded meaningfully while revenue beat by only +0.9%. If investors fade the entire EPS surprise as tax-aided, they miss the more important evidence that gross margin and operating leverage are already doing the heavy lifting.

The trade from here is to own the margin reset but police the numbers that would prove it is merely cyclical. The first watch item is fiscal Q1 guidance delivery: revenue needs to land inside the $12.5 billion, plus or minus $300 million range, and gross margin needs to hold around 51.5%, plus or minus 100 basis points. The second is whether HPM revenue can sustain the nearly $2 billion Q4 level rather than slipping back toward a niche-product profile. The third is capital intensity: Q4 CapEx of $4.9 billion was manageable because operating cash flow was $5.7 billion, but that spread must not compress if Micron keeps spending at fiscal 2025’s $13.8 billion pace. The thesis breaks if revenue grows but gross margin fails to approach the guided 51.5% band, because that would mean the market was right to treat Q4 as a price-cycle peak. It is confirmed if Q1 delivers the guided revenue range with gross margin at or above the midpoint, because then the Q4 beat was not the event; it was the first visible quarter of a different earnings base.