Micron FQ4 FY2025: A Record Close, and a Guide That Crosses 50% Gross Margin

Results vs. Consensus

Quality of Beat

• Revenue: DRAM-led and pricing-rich. DRAM revenue jumped 27% sequentially to $9.0B (79% of the total), with bit shipments up low-teens and, crucially, prices up low-double-digits on tight supply and favorable mix. After several quarters of volume-led beats with prices down, this was the first quarter where price was a clear positive contributor, which is the signal that the cycle has turned decisively.
• Margins: The 45.7% gross margin beat the guide by 370 bps, up 670 bps sequentially, driven by mix, DRAM pricing, and cost execution. Operating margin of 35% (up 820 bps) shows the leverage flowing straight through. This is a high-quality, broad-based margin print, not a one-item benefit.
• EPS: The $0.26 beat was helped slightly by a 12% tax rate (below guide on discrete items), but the operating engine did the work: operating income of $4.0B, up 12 points of margin year-on-year. The one blemish is free cash flow of just $803M, as capex stepped up to $4.9B for the DRAM build.

Segment Performance

Revenue by Technology

Revenue by Business Unit (new segment structure, first reporting)

The new segment map is the most important disclosure of the quarter

For the first time, Micron broke out profitability by the reorganized business units, and the result reframes how to value the company. Cloud Memory, which houses HBM and cloud DRAM, generated a 59% gross margin at 40% of revenue. Core Data Center added another 14% at a 41% margin. Together the data-center franchise was 56% of fiscal-2025 revenue at a 52% gross margin. The point the disclosure makes is unambiguous: Micron is now, in the majority, a data-center memory company with software-like segment margins on its highest-value lines.

"In fiscal 2025, Micron's data center business reached a record 56% of total company revenue with gross margins of 52%." — Sanjay Mehrotra, CEO

Assessment: A 59% Cloud Memory gross margin is the single most powerful number in the release. It quantifies what the bull case has argued qualitatively: the HBM-and-cloud-DRAM core deserves a structurally higher multiple than legacy memory, and the rest of the portfolio (Mobile & Client at 36%, Auto & Embedded at 31%) is also re-rating fast. The new segments make the mix-shift story legible to the market for the first time.

Mobile, Client, Auto, Embedded: the broad lift

Every business unit grew sequentially with sharply higher margins. Mobile & Client rose 16% with gross margin up 12 points to 36%; Auto & Embedded rose 27% with margin up 540 bps to 31%, as automotive and industrial demand strengthened beyond Micron's initial forecast. The breadth confirms this is an industry-wide tightening, not a data-center-only phenomenon.

Assessment: When even the lowest-margin segments are expanding margins double-digits, the blended-margin tailwind has real durability. The cyclical recovery has fully broadened, which reduces the risk that one weak end-market pulls the blended number down before HBM and cloud DRAM can offset it.

Key Topics & Management Commentary

Overall Management Tone: The most confident posture of the four quarters we have covered, and earned. Management opened by reminding the market it had promised in early 2024 to be a primary AI beneficiary and then delivered a 49%-revenue-growth, 538%-EPS-growth year. The forward framing leaned hard into a tight 2026 DRAM environment and an industry-leading HBM4. The only hedging was the routine refusal to guide beyond fiscal Q1, and a candid acknowledgment that capex is stepping up materially.

1. Gross Margin Crosses 50%

Fiscal-Q1 gross margin was guided to 51.5%, up roughly 580 bps sequentially, driven by tight DRAM supply and pricing, improving NAND, favorable mix, and cost execution. Management expects margins to rise again from fiscal Q1 into fiscal Q2, citing durable supply-demand factors.

"We expect gross margin to improve sequentially first to second quarter. And it's on this tight DRAM supply and the associated pricing along with NAND business continuing to improve... we expect margins to be healthy in both HBM and non-HBM in 2026." — Mark Murphy, CFO

Assessment: Guiding margin over 50% and then signaling a further sequential rise is the strongest possible statement about pricing power. The explicit "healthy in both HBM and non-HBM" framing matters: this is not just an HBM mix story, the entire DRAM book is earning well. The margin bear case from our initiation is, for now, defunct.

2. HBM Reaches DRAM-Parity Share, HBM4 Leads the Industry

HBM revenue reached nearly $2B in the quarter (an ~$8B annualized run rate), and Micron's HBM share is on track to match its overall DRAM share in calendar Q3, delivering on a multi-quarter target. On HBM4, Micron shipped samples with bandwidth exceeding 2.8 TB/s and pin speeds over 11 Gbps, which management asserts outperforms all competing HBM4 products, with first production shipments expected in calendar Q2 2026.

"We have recently shipped customer samples of our HBM4 with industry-leading bandwidth exceeding 2.8 terabytes per second and pin speeds over 11 gigabits per second. We believe Micron's HBM4 outperforms all competing HBM4 products." — Sanjay Mehrotra, CEO

Assessment: Reaching DRAM-parity HBM share on schedule, then claiming outright performance leadership on HBM4, completes Micron's transformation from HBM laggard to co-leader. The in-house advanced CMOS base die is a genuine differentiator. This is the durable competitive position the entire structural thesis rests on, and it is strengthening.

3. 2026 HBM Increasingly Locked In

The HBM customer base expanded to six, and Micron now has pricing agreements with almost all customers for the vast majority of its HBM3E calendar-2026 supply. It is in active discussions on HBM4 specifications and volumes and expects to conclude agreements to sell out the remainder of total 2026 HBM supply in the coming months.

"We have pricing agreements with almost all customers for a vast majority of our HBM3E supply in calendar 2026... we expect to conclude agreements to sell out the remainder of our total HBM calendar 2026 supply in the coming months." — Sanjay Mehrotra, CEO

Assessment: Locking in the vast majority of 2026 HBM supply with agreed pricing converts a large slice of next year's revenue into something close to contracted backlog. That visibility is exactly what justifies a higher, more stable multiple than memory has historically earned. The remaining 2026 HBM4 pricing is the last big unknown, and it is being closed.

4. The DRAM Supply Story Is Structurally Tight

Management laid out a multi-factor case for 2026 DRAM tightness: low supplier inventories, constrained node migration (because the industry is extending D4/LP4 support), longer lead times, and higher global costs for new wafer capacity. The HBM trade ratio compounds it by consuming leading-edge wafers. Management also raised the calendar-2025 server unit growth outlook to roughly 10%, citing AI agents driving traditional-server workloads.

"We anticipate further DRAM supply tightness in the industry and continued strengthening in NAND market conditions... low supplier inventories, constrained node migration... longer lead times, and higher costs globally for new wafer capacity, all to limit the pace of supply growth for DRAM in 2026." — Sanjay Mehrotra, CEO

Assessment: This is the crux of why the margin can stay elevated. Supply additions are structurally constrained while demand broadens across AI servers, traditional servers, AI PCs, and AI phones. A tight 2026 DRAM market is the backbone of the maintained Outperform; the risk is that the same high prices eventually pull forward supply, which is the late-cycle worry.

5. 1-Gamma Yields in Record Time

The 1-gamma DRAM node reached mature yields 50% faster than the prior generation, Micron is first to ship 1-gamma, and it booked first revenue from a major hyperscale customer on 1-gamma server DRAM. 1-gamma will carry the majority of DRAM supply growth in 2026, freeing 1-beta capacity to feed HBM.

"Our one gamma DRAM node reached mature yields in record time, 50% faster than in the prior generation. We are the first in the industry to ship one gamma DRAM." — Sanjay Mehrotra, CEO

Assessment: Process leadership compounds the supply-tightness advantage: 1-gamma improves cost and bit density on standard DRAM while preserving 1-beta for HBM. The record-fast yield ramp is the operational proof that Micron's technology execution, not just the cycle, is driving the margin story.

6. Capex Steps Up Hard

Fiscal-2025 capex was $13.8B net; fiscal 2026 is framed at roughly $18B net (about $4.5B per quarter as a baseline), driven by DRAM front-end equipment and greenfield construction. Fiscal-Q4 free cash flow was just $803M as capex hit $4.9B, though management projects significantly higher annual free cash flow in fiscal 2026.

"We expect fiscal 2026 CapEx to be higher than fiscal 2025 levels. DRAM front-end equipment and fab construction will drive higher capital spending in fiscal 2026." — Sanjay Mehrotra, CEO

Assessment: This is the one pillar moving against the bull case. A roughly $4B step-up in capex is the cost of the supply tightness that supports pricing, and it compresses near-term free cash flow even as earnings soar. The bet is that the pricing environment more than pays for the build. It is the right bet at a generational inflection, but it raises the stakes if the cycle turns before the capacity monetizes.

7. NAND Quietly Turns the Corner

NAND prices rose high-single-digits sequentially on favorable mix, full-year NAND revenue hit a record $8.5B, and management sees improving conditions into 2026, helped by HDD supply shortages pushing demand toward NAND. Micron also ceased future mobile managed-NAND development to focus on higher-ROI opportunities.

Assessment: NAND has gone from the margin anchor of our initiation to a contributor with rising prices and a healthier supply backdrop. Exiting low-ROI mobile managed NAND is a sharpening of the portfolio toward data-center and high-value storage. The NAND bear pillar is now largely contained.

8. The U.S. Champion Positioning

Management leaned into Micron's status as the only U.S.-based memory manufacturer, tied to the $200B domestic investment plan and CHIPS grant disbursements, and highlighted internal AI productivity gains of 30-40% in select use cases such as code generation. A first EUV tool was installed in the Japan fab in record time.

Assessment: The U.S.-manufacturer angle is strategically valuable in a policy environment favoring domestic semiconductor capacity, and the internal AI-productivity gains are a quiet cost-and-speed advantage. Neither moves the near-term model much, but both reinforce the durability of the competitive position.

Guidance & Outlook

The fiscal-Q1 guide raises every line that matters: record $12.5B revenue, a 51.5% gross margin that crosses the symbolic 50% threshold, and record $3.75 EPS, even after absorbing a higher 16.5% tax rate. Management framed margin as rising again into fiscal Q2 on tight DRAM supply and improving NAND.

Implied trajectory: Annualizing the $3.75 fiscal-Q1 EPS yields a run rate around $15, and with management guiding margin higher still into fiscal Q2 and DRAM tight through 2026, fiscal-2026 EPS is set to dwarf fiscal-2025's $8.29.

Street at: The guide is comfortably above consensus on all three lines. The market is again chasing estimates higher rather than the company guiding down to a beatable number.

Watch items: Two offsets to the earnings ramp: the tax rate steps to 16.5% (Singapore global minimum tax), and capex rises to roughly $18B, holding back near-term free cash flow. Fiscal 2026 is also a 53-week year, with the extra week landing in fiscal-Q4 OpEx.

Analyst Q&A Highlights

Is 51.5% the new baseline, and how much room is left?

The dominant line of questioning probed whether the 51.5% margin is a peak or a step on a higher path, and how much further the 59% Cloud Memory margin can climb. Management declined out-quarter guidance but committed to margins rising from fiscal Q1 into fiscal Q2 on tight DRAM supply, improving NAND, mix, and cost.

Q: "How do you think about the puts and takes of gross margins as you go through the rest of the year, the 51.5, is this kind of the baseline? And as long as sales grow, can you expand off of this level?... when I look at your cloud data center business, gross margin's 59%... how much more room is there?"
— Vivek Arya, Bank of America Securities

A: "We expect gross margin to improve sequentially first to second quarter... on this tight DRAM supply and the associated pricing along with NAND business continuing to improve... we expect margins to be healthy in both HBM and non-HBM in 2026."
— Mark Murphy, CFO

Assessment: Management would not call a ceiling and pointed to further upside, which is the strongest available signal short of explicit guidance. The "healthy in both HBM and non-HBM" line is the key tell: this is a whole-portfolio margin story, not an HBM-only one.

The longer-term HBM TAM

A question sought an update to the long-term HBM TAM given the explosion in accelerator spending. Management reaffirmed a roughly $100B HBM TAM by 2030 and reiterated HBM bit growth outpacing overall DRAM, framing memory as central to trillions of dollars of coming AI infrastructure spend.

Q: "You had previously guided us to a $100 billion HBM TAM by 2028... it's obvious that the compute TAM is much bigger than what you probably would have seen at that time. Do you give an update to that number?"
— Timothy Arcuri, UBS

A: "By 2030, we expect HBM TAM to reach $100 billion. And we had also said that HBM bit CAGR will grow faster than the DRAM CAGR... the value proposition of HBM continues to increase... memory is very much at the heart of this AI revolution."
— Sanjay Mehrotra, CEO

Assessment: A $100B HBM TAM by 2030, with bit growth outpacing DRAM and value per bit rising through HBM4 and HBM4E, gives the structural story a multi-year runway. Management held the TAM number rather than chasing it higher, which reads as disciplined rather than promotional.

HBM3E-to-HBM4 crossover and 2026 pricing direction

A question pressed on the timing of the HBM3E-to-HBM4 crossover and the direction of settled 2026 HBM3E pricing. Management put first HBM4 production in calendar Q2 2026 ramping through the year, said share grows in 2026, and declined to comment on pricing direction beyond noting a healthy demand-supply environment.

Q: "How do you see the transition from HBM3E to four? When do you expect the crossover next year?... you mentioned that the pricing for 3E is settled for 26. What is the direction of that pricing versus what you're getting now? Is it higher or lower?"
— Vivek Arya, Bank of America Securities

A: "First production shipment in CQ2 of 26 time frame and production will ramp during the course of 2026... in 2026 versus 2025, we see our share growing... We are not commenting on the pricing of HBM3E... supply is tight. We expect a healthy demand supply environment in 2026."
— Sanjay Mehrotra, CEO

Assessment: The refusal to give a pricing direction is the most important non-answer on the call. With supply tight and demand locked, the silence leans constructive, but 2026 HBM pricing remains the single largest swing factor for next year's estimates, and it is not yet fully resolved.

Sustainability of the DRAM demand inflection into a seasonally soft quarter

A question asked whether the recent DRAM demand inflection led by inference hyperscalers is durable, and whether the seasonally slower February quarter would hold up better than normal. Management pointed to broadening AI demand across data center and edge, rising traditional-server demand, and tight supply as reasons for a healthy 2026.

Q: "It certainly feels like in the last month or two, there's been an inflection in DRAM demand led by inference hyperscalers... the sustainability of that... You've talked about tightness expected into fiscal 26... should we see kind of normal seasonality?"
— C.J. Muse, Cantor Fitzgerald

A: "AI trends are strong... not just in training, but in inference as well... we are also seeing traditional server demand increase... this is across data center, across AI-enabled smartphones, and AI-enabled PCs... we look forward to a healthy demand supply environment in calendar year 2026."
— Sanjay Mehrotra, CEO

Assessment: Management is implicitly signaling better-than-seasonal trends into the February quarter, supported by tight supply. The breadth of the demand (training plus inference plus edge plus traditional servers) is what makes the durability claim credible rather than hopeful.

The fiscal-2026 capex step-up

A question sought detail on the roughly $18B net fiscal-2026 capex and its equipment-versus-construction split. Management kept the breakdown high level (vast majority DRAM, with construction, node-transition tools, and greenfield installs) and clarified the net-versus-gross convention, noting fiscal-2025 was $15.8B gross less $2B incentives.

Q: "It appears net CapEx implied $18 billion versus $13.8 billion last year... Is there a way to kind of partition how much on equipment versus clean room? And can you share the implied gross CapEx for fiscal 26?"
— C.J. Muse, Cantor Fitzgerald

A: "Our spend in '26 will be the vast majority for DRAM, and we've got construction and facilities related to that, some tools for node transitions and beginning to install for the new greenfield... we ended up at 13.8 net, and we were at 15.8 gross with $2 billion of government incentives in '25."
— Mark Murphy, CFO

Assessment: The capex step-up is real and front-loaded into DRAM capacity that monetizes over multiple years. It is the cost of the supply tightness that supports pricing, but it is also the line most likely to disappoint free-cash-flow-focused investors if the cycle turns. The biggest single watch item from here.

HBM4 base-die design and performance leadership

A detailed question asked whether achieving industry-leading HBM4 bandwidth required a base-die redesign and whether power efficiency leadership held at the higher speeds. Management credited the in-house advanced CMOS base die (manufactured by Micron) and DRAM design for exceeding customer requirements on both bandwidth and power.

Q: "Some of your HBM4 customers are looking for as much as 25% more bandwidth versus the plain vanilla JEDEC standard... Did the team have to redesign the base logic die to achieve these results?... even with the higher speeds, is your power consumption still superior to your competitive solutions?"
— Harlan Sur, JPMorgan

A: "Our innovative design, our memory architecture, our advanced CMOS in the DRAM, as well as advanced CMOS in the base die, and that advanced CMOS base die is manufactured here by Micron giving us a competitive advantage... enabled us to achieve customers' increasingly higher requirements, bandwidth at 2.8 terabytes per second and speed at more than 11 gigabits per second."
— Sanjay Mehrotra, CEO

Assessment: Owning the advanced CMOS base die in-house is a structural edge over competitors that outsource it, supporting both performance and margin. That Micron exceeded customer specs while claiming power leadership de-risks the most important 2026 product and underpins the share-gain narrative.

What They're NOT Saying

1. The direction of 2026 HBM pricing: Management repeatedly declined to say whether settled HBM3E 2026 pricing is up or down versus current levels. With supply tight the silence leans positive, but it leaves the most important 2026 estimate variable unquantified.
2. Fiscal-Q2 and full-year FY26 guidance: Only a qualitative "margins up into fiscal Q2" was offered. The durability of 50%-plus margins beyond one quarter is asserted, not guided.
3. Gross fiscal-2026 capex: Management gave the ~$18B net figure but would not disclose the gross number or the equipment-versus-construction split, leaving the free-cash-flow bridge fuzzy.
4. What happens to pricing if supply catches up: The bull case rests on tight supply; management did not address the late-cycle risk that today's high prices pull forward the very capacity that ends the tightness.
5. NAND beyond "improving": NAND commentary stayed qualitative; no explicit margin or pricing trajectory was given for the segment that was the prior drag.

Market Reaction

• Pre-print setup: MU closed at $166.41 on September 23, up a remarkable 97.7% year-to-date and 41.4% over the trailing 30 days, near a 52-week closing high of $168.89. Expectations were extremely elevated heading into the print.
• After-hours move: Shares ticked up roughly 1% in the initial after-hours reaction (to about $165.61) on the beat-and-raise.
• Next-day session: The stock faded to close -2.8% at $161.71 on September 24, on 57.1M shares (2.7x the 30-day average), with an intraday low near $158.30. The S&P 500 was -0.3%.

For the third consecutive quarter a strong print drew a muted-to-negative reaction, which is a positioning story rather than a results story. With the stock up nearly 98% year-to-date, the bar was extraordinary, and even a blowout that guided gross margin over 50% could not produce a follow-through rally. We read a 2.8% fade on a guide this strong as digestion of an extended chart, not a verdict on the fundamentals. The pattern does, however, carry a message: from here the stock needs the cycle to keep surprising, because the easy expectations gap has closed.

Street Perspective

Debate: Peak margin or a higher plateau?

Bull view: A 51.5% guide rising into fiscal Q2, locked-in 2026 HBM, structurally constrained DRAM supply, and a 59% Cloud Memory margin argue the business has reset to a durably higher margin plateau, not a one-quarter peak.

Bear view: Memory margins always mean-revert; 51.5% is near prior-cycle peaks, capex is surging to add the very supply that ends the tightness, and the right move is to sell when margins peak, not chase them.

Our take: The trade-ratio mechanic and locked-in HBM genuinely lengthen this cycle versus prior ones, so we do not think the peak is imminent. But we respect the late-cycle math: the rating stays Outperform on still-rising forward earnings, and our downgrade trigger is the first concrete evidence of supply catching demand or HBM pricing softening.

Debate: How much should the new segments re-rate the multiple?

Bull view: A 59% Cloud Memory gross margin at 40% of revenue proves the data-center core deserves a semiconductor-grade multiple, and the sum-of-the-parts is worth more than the blended cyclical multiple the market applies.

Bear view: Segment margins are themselves cyclical; a 59% margin today can be a 35% margin in a downturn, so applying a structural multiple to a peak-cycle segment margin is the classic memory value trap.

Our take: The new disclosure justifies some multiple expansion because it reveals genuinely high-value, increasingly contracted revenue in Cloud Memory. We would not capitalize the 59% margin as permanent, but the visibility from locked-in 2026 HBM means the segment margin is more durable than a typical memory peak.

Debate: Does the capex surge undermine the cash story?

Bull view: Roughly $18B of fiscal-2026 capex is funding capacity into a sold-out, supply-constrained market; management projects significantly higher annual free cash flow in fiscal 2026 even after the step-up, so the build is self-financing.

Bear view: Fiscal-Q4 free cash flow was only $803M on $4.9B of capex; a $4B step-up at the top of the cycle is exactly when memory companies overbuild, and the free-cash-flow payoff assumes the pricing holds.

Our take: This is the most legitimate bear point now. The capex is the right call given the demand visibility, and the cash should still grow, but it raises the downside if the cycle turns before the capacity monetizes. We are watching free-cash-flow conversion closely as the swing variable on the rating.

Model Update & Valuation Framework

Valuation framework: At the $161.71 post-print close, the fiscal-Q1 guide of $3.75 non-GAAP EPS annualizes toward $15, and with management guiding margin higher into fiscal Q2 and DRAM tight through 2026, fiscal-2026 EPS looks set to run well above fiscal-2025's $8.29. That puts the stock around 10x forward earnings, which is undemanding for a company at record profitability with a data-center core earning a 59% gross margin and 2026 HBM largely locked in. The bull case is that the cycle has further to run and consensus FY26 numbers are still too low; the bear case is that a roughly 10x multiple is precisely what the market pays for peak-cycle memory earnings before they roll, and that capex plus tax cap the free-cash-flow upside.

Where we land: The fundamentals delivered on every commitment and the forward earnings power is still re-rating faster than the stock, which keeps us at Outperform despite the year-to-date run. We are, however, increasingly attentive to the late-cycle signals: the rising capex, the peak-adjacent margins, and the muted reaction to a blowout. We would move to Hold on concrete evidence that supply is catching demand, that HBM 2026 pricing is softening, or that free-cash-flow conversion deteriorates as capex outruns the cash.

Thesis Scorecard Post-Earnings

Overall: Thesis confirmed and strengthened on five of six pillars. The margin and NAND bears are resolved or contained; the capital-intensity pillar has escalated and is now the central risk, alongside the cyclical question of how long peak-adjacent margins persist.

Action: Maintain Outperform. Stay long on still-rising forward earnings and a sub-cycle-peak valuation, while tightening the watch on capex, free-cash-flow conversion, and the first signs of supply catching demand.

Bottom Line

Micron closed fiscal 2025 with the kind of quarter that retroactively justifies the upgrade: a record $11.3B, gross margin to 45.7%, EPS of $3.03, and a fiscal-Q1 guide that crosses 50% gross margin and lands EPS at a record $3.75. The full year grew revenue 49% to $37.4B and EPS 538% to $8.29. Management delivered on every commitment we were tracking, HBM reached DRAM-parity share on schedule, inventory returned to target, and the new segment disclosure put a number on the bull case: a 59% Cloud Memory gross margin.

We are maintaining Outperform. The honest tensions are that margins are near cyclical highs, the stock is up nearly 98% year-to-date, the reaction to a blowout was a shrug, and capex is stepping up to roughly $18B. But the rating is a twelve-month forward view, and on that horizon the forward earnings power is still re-rating faster than the price, DRAM supply is guided structurally tight deep into 2026, and the stock trades around 10x forward. That combination keeps the risk/reward favorable. We stay long with eyes open: the two things that would move us to Hold are evidence that supply is catching demand or that the capex surge is outrunning the cash. Neither has happened yet, and until it does, the cycle remains Micron's to ride.