Q4 and Full Year 2025 Earnings Release and Analyst Call

[SmoothHughes]

9/23/25

Press Release and Investor Presentation Commentary

Here are the points from the press release, outside the financial statements, that I found noteworthy:

• Revenue, gross margin, and EPS were all above the high end of their updated guidance range. The updated guidance was issued on August 11^th.

• They call better pricing “pricing execution.” I think their intent here is to make it sound more like they are not price takers. While there is a little truth to this, memory companies a narrow range of control around the prices they get. Mostly, they are price-takers with the prices they take determined by supply and demand. Where that range lands in the full spectrum of prices is out of their control.

• “AI-driven demand is accelerating, and industry DRAM supply is tight.”

• Conditions in the NAND industry are improving.

• If they hit the midpoint of their EPS guidance, it will tie their quarterly record.

• Their 1-gamma DRAM node reached mature yield 50% faster than in the prior generation. They are the first company to ship 1-gamma DRAM. The 1-gamma node has 30% higher bit density over 1-beta.

• They are ramping their G9 NAND production and are “scaling at a pace aligned with market demand.”

• They received a CHIPS grant disbursement in FQ4 following completion of key milestones in ID1. Their first wafer output is scheduled for the second half of calendar 2027. That is unchanged from what they said last quarter.

• They began design work on ID2, which will provide additional capacity beyond 2028. That means 2029 or later, but doesn’t sound as far in the future, nor does it seem as vague as it is. Calendar 2029 should be a good year for the DRAM industry, the heart of the next upturn.

• They need EUV to run 1-gamma DRAM. In FQ4 they installed their first EUV tool in their Hiroshima fab. They already have EUV in their production fab in Taiwan.

• The HBM assembly and test facility being built in Singapore will begin to contribute HBM supply in calendar 2027.

• Total data center server units in calendar 2025 are expected to grow approximately 10%, up from their prior expectation of mid-single digits growth.

• The growth outlook for traditional servers in calendar 2025 has strengthened significantly from flat to growth in the mid-single digit range. They believe this increase in demand is related to AI use cases.

• In fiscal 2025, their data center business was 56% of total company revenue with gross margins of 52%. They didn’t say as much, but this business includes HBM, non-HBM DRAM and SSDs.

• In fiscal Q4, their HBM revenue grew to nearly $2B, implying an annualized run rate of nearly $8B. Their HBM share will grow again, and be in line with their overall DRAM share in calendar Q3. Last quarter they said this milestone would be reached sometime in the second half of calendar 2025, thus they pulled in the range from Q3-Q4 to just Q3. This continues the trend in these statements over the last several quarters. This tells me the company has been both exceeding their internal HBM shipment plans, and they have been conservative in their forecasts to analysts.

• To review Micron’s HBM revenue: FQ4-24 was approximately $325M, $650M in FQ1-25, $1B in FQ2-25, $1.5B in FQ3-25, and nearly $2B in FQ4-25. For the last three quarters, HBM revenue has risen steadily by about $500M per quarter. Before that, the rate accelerated from $175M in a quarter to $350M in the next quarter, then the series of three quarters of $500M rise per quarter.

• Their HBM customer base now includes six customers. That is up from four customers last quarter.

• The company has pricing agreements with “almost all customers for a vast majority of our HBM3E supply in calendar 2026.” This sounds good but it doesn’t mean they have pricing guarantees. It just means they have a contract that includes language about pricing.

• They expect to sell out the remainder of their total HBM supply in calendar 2026 in the coming months. Similar to the pricing agreements above, “selling out” doesn’t necessarily mean take or pay. As I’ve said before, I don’t believe if the HBM market goes into oversupply that HBM customers won’t force Micron and other manufacturers to change their pricing and unit delivery terms more in their favor.

• HDD supply shortages are expected to improve NAND demand and thus the overall supply demand environment.

• They raised their expectation for PC unit shipment growth in calendar 2025 from low-single digit growth to mid-single digit growth. They have now round-tripped back to mid-single digit, which is what they forecasted two quarters ago.

• Their forecast for smartphone unit shipments is unchanged for calendar 2025 at low-single digit percent growth. This has been the same forecast for the last four quarterly calls. They believe AI-ready smartphones will increase DRAM content in mobile phones.

• Automotive and industrial demand strengthened throughout the quarter, exceeding their initial forecast.

• Overall customer inventories are healthy across all end markets. I would like to know what the HBM inventory levels are like at customers.

• They have raised their view of calendar 2025 industry DRAM bit demand growth to high-teens percent range. They also said high-teens last quarter. It was mid-teens two quarters ago (back in March).

• They expect industry NAND bit demand growth to also be higher than their previous view. Now they see it in the low-to-mid-teens percent range. Last quarter they said low-double-digit growth.

• Micron’s bit supply growth in calendar 2025 will be below bit demand growth for non-HBM DRAM and for NAND. The “non-HBM” part is key. The company is excluding HBM from their supply-demand picture. The company has been saying how insatiable HBM demand is, yet they won’t say their supply will fall short of industry demand. I don’t think they know they will oversupply the market. I think they just don’t know how much demand is really out there, so they don’t know how their supply plan compares to it.

• The company anticipates further DRAM supply tightness in calendar 2026 and continued strengthening in the NAND market. I bookmark these predictions and check back on them to see if memory industry executives’ predictions more than one quarter out are worth paying attention to.

• Their medium term view of industry bit demand growth is mid-teens CAGR for both DRAM and NAND. This is the same as last quarter and the quarter before.

• The majority of their DRAM supply growth in fiscal 2026 will come from node migration to 1-gamma. I think they worded this statement this way because they don’t want to say they are adding wafers. It doesn’t directly say that, but it could mean that nearly half of their DRAM supply growth is from new wafers. More likely, some is from new wafers but well more than half is from node migration, to 1-gamma and moving forward on older nodes.

• Their 1-beta capacity will support HBM growth in 2026. They didn’t say fiscal or calendar here, which implies they mean calendar 2026.

• DRAM: 79% of total revenue (was 76% last quarter.) Revenue increased 27% Q/Q (was a 15% sequential increase last quarter), bit shipments increased low-teens percent Q/Q, and ASPs increased in the low-double-digit percent range Q/Q. That is simply massive sequential growth in their DRAM revenue, driven by strong shipment growth and impressive price increases. DRAM is nearly four-fifths of the company.

• NAND: 20% of total revenue. Revenue increased 5% Q/Q, bit shipments declined mid-single-digits percent sequentially in the quarter, and ASPs increased high-single-digits percent range in the quarter. Last quarter, NAND seemed oversupplied, but not by much. Now it seems to be close to balanced. I say this because pricing rose 9%, a healthy increase, yet bit shipments were down 5%. I think that drop in bit shipments is a hangover from the 25% sequential growth in bit shipments we saw from the company last quarter. Micron is shipping only 20% more NAND bits than they did two years ago. We learned in the analyst calls that the rise in NAND pricing this quarter was from mix, not like-for-like. Thus, while it is improving, the NAND market for Micron isn’t healthy.

• Micron reorganized their business units last quarter. To give some continuity, they provide selected financials for this quarter, last quarter, and one year ago. Hyperscalers, including all HBM sales, are in CMBU. Their revenue has grown 3x in the last year, to $4.54B, with 59% gross margin.

• The other three business units are CDBU, MCBU and AEBU. The first is core data, which is non-HBM and non-hyperscalers data center. That BU saw revenues decline by more than 20% in the last year, with GM of 41%. Mobile and client (that’s PCs) grew about 20% in the last year and has GM of 36%. Auto and embedded has revenue up somewhat with GM of 31%.

• The summary of the business units is there has been some long-awaited improvement in PCs and mobile. Auto and embedded are muddling along. Traditional data center customers are reducing memory purchases, probably because that workload is shifting to HBM and hyperscalers. Most of the action is in HBM and high-end DRAM.

• The higher guidance for Q1 will come from higher prices, lower cost, and better mix.

• Fiscal 2026 will be a 53-week fiscal year with the fourth fiscal quarter being the one with the extra week.

• Their fiscal Q1 and FY-26 tax rates will be around 16.5%. They said “high-teens” last quarter, because of Singapore adopting a global minimum tax. They must have made some progress over the last three months on how they’ll manage taxes next year.

• The company will spend $4.5B in net CapEx in FQ1-26. They said this is a “reasonable baseline” for the rest of the year’s spending. I have never seen Micron not give a full year CapEx guidance. They did so indirectly (it’s $18B, net CapEx). I think they don’t want to spook the market by printing such a large CapEx number. If we assume a similar level of government incentives in FY26 as was received in FY25, their gross CapEx in this fiscal year will be $20B. In the five fiscal years prior to FY24, Micron averaged $440M per year in government incentives. That was mostly pre-CHIPS, making last year an outlier, one which may be repeated this year.

• Guidance does not include impact from any new tariffs.

• Non-GAAP guidance for FQ1-26: revenue $12.2B to $12.8B, gross margin between 50.5% and 52.5%, operating expenses of $1.34B, +/-20M, diluted EPS of $3.60 to $3.90.

Financial Statements

Statements of Operations

The midpoint of Micron’s revenue guidance for the fourth fiscal quarter was $10.7B Management pre-announced a few weeks ago a higher range centered around $11.2B. Actual revenue in the quarter came in at $11.32B. The market has improved steadily through the June, July, August period, leading to $600M more in revenue than management foresaw just three months ago. That is the memory business. That increase in revenue is a 21.7% sequential gain. Because so much improvement happened that a pre-announcement was needed, we can conclude that most of the revenue gains were from improved pricing rather than lower cost or better mix. Only pricing can change that fast. COGS only rose by 8.1% sequentially, leading to an expansion of gross margin from 37.5% to 44.7% quarter-over-quarter. That is closing in on the peak gross margin during the last upturn. If they hit the midpoint of guidance, GM will crest 50% for the first time since the first fiscal quarter of 2019 (late calendar 2018).

Speaking of records, the company spent more than one billion dollars on R&D in a single quarter for the first time in their history. They took a $38M restructuring and asset impairment charge in the quarter. The company is not disciplined enough to have their 10-K out at the same time as their quarterly announcement, so I can’t see what is behind this one-time charge. Sequentially, operating expenses grew by $61M (a little) while gross margin rose by $1.546B (a lot). Hence, operating income rose from $2.1B to $3.6B, and operating margin grew by almost 50% (900 bps), to 32.3%. In the financing line, interest income, interest expense, and other about canceled out. Income taxes were 11.8%, lower than the 13% the CFO guided for in the last call. They may have moved some earnings and costs around in the quarter, or perhaps the pricing rises happened in a way that led to lower income taxes. The bottom line was $3.20B of net income on 1,131 million shares, or $2.83 per share in earnings. That is the highest EPS Micron has had since their quarterly record, way back in FQ4-18, seven years and two upturns ago. That was the peak of the data center super cycle. I estimate that the full memory cycle has a duration of about three-point-five years. That makes the last peak two cycles ago, if indeed this quarter was the peak. But we know it won’t be, because of Micron’s guidance.

The company guided continued improvement in the first fiscal quarter of 2026. That is the quarter from September through November, inclusive. In that quarter, the midpoint of guidance is $12.50B of revenue with a gross margin of 50.5%. That would be a $1.2B increase in revenue, less than the $2.0B of revenue growth this quarter. But COGS would actually need to decline by 1.2% to achieve a gross margin of 50.5% on that level of sales. The two most likely ways COGS could decline while revenue goes up a lot is better pricing and lower cost. I think that is most of what is in the guide is better pricing, helped by some cost reduction. The company will sell the same number of bits for more money. The trend in semiconductor cost is always down, so the same number of bits will cost 2-3% less to make this quarter than they costed last quarter. Within cost reduction – which over the long-term is driven by process node migration – is yield improvement. HBM is a complex product with a lot of die in a single package. Hence, HBM yields have been lower than those of other products and always will be. But the gap has been wider for the past three years than it will be in a mature state. Micron is still in the steep part of the HBM learning curve. For this reason, I suspect they are seeing significant yield improvements in HBM flow through to their finished products now, leading to meaningful cost reductions. The company’s Q1-26 EPS guidance is for $3.57 per share. I don’t believe it is a coincidence that this would tie the all-time company record for quarterly earnings, set way back in FQ3-18, also during the data center super cycle. My prediction is they will beat that number, so Sanjay can claim the best-ever quarter for the company. I also think Mehrotra will announce his retirement within the next six months. He wants to go out on top, and I think this cycle will represent a peak which won’t be matched for at least three more years. He is 67 years old now and doesn’t have much more to prove. He should exit while the exiting is good and leave the next downturn for someone else to navigate.

Cash Flows

The company generated $5.73B in cash from operations. This total could almost completely be accounted for between $3.2B of net income and $2.2B of depreciation and amortization. Management has been saying their production will be under bit demand, so they would make up the difference by drawing down inventory. The 10-K isn’t out yet so I can’t see how much FG inventory changed, but total inventory was down less than 5% this quarter from last quarter. That is meaningful, but I thought it would be more. As is often the case for memory companies, most of the cash generated from operations in the quarter went back out in capital expenditures. This quarter, Micron spent a record $5.66B in CapEx. This was much more than was suggested at the end of last quarter, because they often quote net CapEx instead of gross.

For those of us monitoring capacity investment growth, it doesn’t matter how much a government grants Micron to offset their capital investments. What matters is how much equipment they buy. Micron spent $15.9B in CapEx during FY-25. That is up from $8.4B in FY-24, and well up from the $7.7B of two years ago. In that year, FY-23, the depths of the last downturn had been reached and spending on production capacity, in the form of node migrations, was unsustainably low. But FY-25 spending was nearly 2x the healthy level of FY-24. And the company guided for $18B in net CapEx spending in the current fiscal year (FY-26). Gross will be between $500M and $2.0B more. It is noteworthy they didn’t explicitly say $18B. Instead they said $4.5B in the first quarter and that that level is a good baseline for the rest of the year. I think they don’t want to scare everyone by saying “$18 billion.” Going back to the start of the last downturn in FQ3-22, Micron’s ratio of CapEx to revenue has been 40.8%. The industry targets around 35% of revenue for their average CapEx. At that target level, the $18B of CapEx that Micron is targeting is appropriate for total company revenue of $12.9B per quarter. To grow into that CapEx, Micron needs to grow revenue in Q2 by 3% over their Q1 guide and stay at or above that level. Given today’s growth rates, that is a given. The point is, the level of investment Micron is planning to do assumes the DRAM industry has entered a new regime, one where any downturn is quite soft and where revenue is stable to up for at least two to three more years.

The company shifted somewhat higher into available-for-sale securities. They also paid off one billion dollars in debt. That is a sign of confidence from management in how they see the strength of their business environment. To summarize, most the cash flow the company made from operations in this quarter went back out in a record level of capital expenditures. They also paid off $1.015B worth of their debt. The net effect was that their cash position declined by $523M in the quarter. In a record quarter, the company had negative free cash flow. Now, it was a record quarter for CapEx as well, and about $700M of that CapEx was paid for by government grants. They also put a lot into receivables in the quarter. Still, one would hope for more cash flow from a company reporting record results and trading at a market capitalization of $180B. A rough estimate of free cash flow for next quarter is they will generate $1.75B. That annualizes to $6.10 per share, which is a 3.8% free cash flow yield. At $160 per share, the stock market believes Micron’s FCF is going to grow a lot from here, perhaps 2x or more. Because pricing can swing so far and so fast, the lever on cash flow is long in memory. That means a 2x or more increase in FCF is possible for Micron, in just two or three quarters. But if the market turns down, that free cash flow turns negative quickly.

The Balance Sheet

As I said above in the cash flow section, Micron’s cash and equivalents declined by half a billion dollars in the quarter. They now stand at $9.64B. Add in the $665M of short-term investments and Micron has more than $10B in cash and cash-like assets. While that is a record, the company also has a near-record level of long-term debt, a total of $14.0B. The high level of CapEx being spent has PP&E up to $46.6B as of the end of the fourth quarter. That is up about $9B, or almost 25%, in the last year. Though their debt is at record levels, the company’s revenue is also at record highs, so they have plenty of capacity to support the debt. Their debt also has maturities many years out in the future. They haven’t released their 10-K yet so I can’t comment on finished goods inventory, but total inventory is down about 4.3%. If the three inventory categories are similarly proportioned to last quarter, FG won’t be near the record lows seen back at the last peak. It is contrary to what many commentators believe is a sign of continued strength in memory markets, but inventory levels at producers are lowest at the peak of the cycle. Micron’s total equity value stands at $54.2B. At $160 per share, MU is trading at 3.45x book value. That is the highest level this measure has seen going back to at least the year 2000. That peak value of 3.15x, reached in August of 2000, was caused by the dot.com investment boom. The equity market believes this time is different for Micron.

Conference Call

Numbers given during the conference call are non-GAAP. As I provide summary and commentary on the investor presentation slides above, this section will only include new information not included in the slides, plus bullet points on the Q&A with analysts.

Sanjay Mehrotra (President and CEO

· In a deviation from last quarter, Mehrotra inserted a thank you to the global employees of Micron for their performance to make this quarter happen.

· Besides that change, Mehrotra again spent the first seventeen minutes of the call reading the slides to the audience, with little additional commentary. It has been seventeen minutes in the last several calls. At least Sanjay is consistent in wasting our time. I think the management team does this because they don’t want more time answering questions.

Mark Murphy (CFO)

· There was sequential growth in consumer-oriented markets.

· DRAM bit demand was driven by strong demand in all end markets. Better DRAM pricing resulted from better pricing and better mix.

· FY-25 all-in DRAM costs were down by low-single-digits percent. This includes the increased cost per bit associated with more HBM in their product mix.

· The pricing increase in NAND was driven by favorable mix. It is noteworthy that better pricing isn’t behind the higher NAND ASPs in the quarter.

· FY-25 all in NAND cost reductions were around low-teens percentage.

· The business unit that sells all HBM said their gross margins were improved because of cost reductions, while their BU that serves “legacy” data centers (ones that have much less or no AI workload) saw their gross margin improved because of higher pricing and favorable mix. I interpret this to mean that HBM ASPs are coming in as predicted, which is my baseline expectation given all the talk from Micron executives over the last two years about how HBM volume and pricing was in long-term agreements. I think their gross margin growth in HBM is because their yields are getting better, plus they are seeing cost savings from economies of scale. Note the GM increase in the HBM business unit was only 100 bps. In the non-AI data center BU, they saw a GM expansion of 300 bps. That market has improved in terms of pricing.

· They saw better pricing also in their mobile, automotive, and embedded business units.

· The company gross margin increased 670 bps because of (in order of how the CFO listed them, which is likely also order of importance) favorable product mix, improved DRAM pricing, and cost reductions. As mentioned above, the NAND market isn’t seeing better like-for-like pricing at Micron. They saw higher ASPs per bit because of better product mix.

· Their lower-than-guided tax rate came from favorable treatment of “certain discrete items.”

· Adjusted free cash flow (which adds in proceeds from government incentives as well as cash from selling used equipment) for FY25 was $3.7B. That is a 2.0% FCF yield at $160 per share. The Market is assuming much more FCF in the future from Micron.

· They ended the quarter with 124 days of inventory. Inventory was down 15 days sequentially, driven by strong sequential bit shipments in DRAM. DRAM inventory levels are below targets, and NAND inventory “improved.” Their NAND quarter wasn’t as good as it first appeared from the bit shipments and ASP change. It’s not bad, it was just flat to down. They are still carrying NAND inventory above target levels.

· The $900M of debt reduction was $700M in reducing term loans and $200M in buying back some of their paper.

· The company sees higher sequential free cash flow in FY26 over FY25. The market would have to turn down before the third quarter for this not to come true.

Mehrotra and Murphy spent more than half the call on prepared remarks. This is way too much. Sanjay loves to hear himself talk, which demonstrates a puzzling lack of self-awareness as he is monotone and uninspiring to listen to. Write an investor letter and leave more time for Q&A.

Analyst Q&A

· In FQ1-26, their revenue growth will be more heavily from DRAM. The split will come from all three areas, but it sounds like pricing is most important. DRAM is stronger than NAND. There are “strong supply and demand factors” at play in the market.

· They expect tight DRAM supply in 2026 for both HBM and non-HBM DRAM. I’m noting this to check back on my rule that memory executives don’t know much more than the rest of us beyond one quarter out.

· They expect gross margin to rise sequentially from Q1 to Q2 of FY26. They believe the market improvements they are seeing now are “durable.” Customer inventory levels are “healthy.”

· The $18B CapEx number they gave for FY26 was a net number. This is an important fact. They declined to provide a gross number for FY26. For reference, which the CFO offered up, they got $2B in government incentives in FY25. It is reasonable to assume that they will get a similar level of incentive dollars in FY26, which is $5B a quarter.

· The analysts asked few useful questions during this call, hence the short section here.

Post Earnings Q&A call with Mark Murphy (CFO), Sumit Sadana (CBO) and Manish Bhatia (VP of Global Operations)

· The NAND bit decline in FQ4 was mix-driven noise, according to the CBO. Hyperscalers will need more SSDs, in part because of shortages in HDD. They expect NAND conditions to “start improving.” Does that mean that NAND conditions are still not very good? DRAM is tight today and getting tighter. NAND is behind this.

· Management confirmed that the $18B in FY26 CapEx is a net number. That is up from net $13B in FY25 to $18B in FY26. They declined to say more than that the increase is driven by DRAM and construction.

· They expect to have higher share in HBM in calendar 2026 than they do in 2025. Sadana reiterated that they would match their overall DRAM share in Q3 of calendar 2025. This means they will be growing their share of HBM beyond their overall DRAM share, so they’ll be taking it from Hynix and/or Samsung.

· Right now, and in the foreseeable future, the non-HBM DRAM portfolio is less profitable than their HBM products.

· Management believes the share-shifting among the HBM market will happen between Hynix and Samsung. They don’t think their share will be impacted.

· DDR4 is a low-single-digit portion of their business. Pricing and margins have become significantly better because of the shortage.

· DRAM margins are now higher than mid-2022. I think he means FY22 since that would have been the last peak. The NAND business can “continue to improve.” NAND is still below mid-2022 levels. The CFO says meaningful incremental supply additions are constrained. I don’t believe that. They have cleanroom space available now. They just need to add the equipment; thus new DRAM wafers can come out within a few months. Maybe Murphy’s definition of incremental is shorter than mine.

· The 1-gamma DRAM node will be the primary source of new supply in FY26. 1-gamma plus 1-beta production are the “significant majority” of their bit production today.

· They got their 12-high HBM yields up much faster than 8-high. Since 8-high was a newer product, that might have been a low bar.

· Prior to generative HBM, when large new segments came online (mobile, data centers) they peaked at roughly a third of the total market. Sadana believes this limit does not apply to generative AI data center demand. The management team seems to believe that AI means memory has entered a new regime with unprecedented growth and margins. They didn’t say it, but this implies they believe this time is different. Micron’s management team see nothing but blue sky ahead from here.

· The decision to exit managed NAND was made because the pricing expectations over time did not generate adequate ROI. I think they (and all the other manufacturers) are stuck in the NAND business, so they are exiting the least appealing parts of the NAND business. This is a smart change to make, one that Micron has not been willing to make in the past.

ASP and Bit Shipments

Recall last quarter I said this: “I’ll characterize the DRAM market in the third quarter as mixed while NAND remained weak.” This quarter, it is obvious that both markets have continued to improve. Blended DRAM prices rose by 11% sequentially and bit shipments increased 13%. When you can raise prices by double digits while also shipping double-digits-more bits, there is strong demand. In the current situation, customers are saying “ship me more DRAM and I’m willing to pay a higher price to get it.” One level below, we can see that this strength is coming from hyperscalers first, and somewhat from improving demand in the PC and mobile segments. But mostly it is hyperscalers with HBM the outsized source of this strength. Pulling back to the level of the DRAM cycle, since bottoming out two years ago (in the third CALENDAR quarter of 2023,) Micron’s DRAM prices have increased by about 110%. That is a CAGR of 45% over those eight quarters. It should be noted that this massive increase came off of an historically bad downturn, one that saw prices drop by more that 50%. But in the prior upturn – the COVID upturn – DRAM ASPs at Micron only rose 30% from trough to peak. Over that same two year span, Micron’s bit supply has increased at a 36% CAGR. Recall, management expects DRAM bit demand growth to be in the mid-teens-percent range over the medium term. The last two years have been so far above that.

After three quarters of softness, Micron’s NAND business has found its footing. I believed we were in for a longer downturn than this two quarters ago. Historically, I have not seen downward pricing trends reverse in less than a year. I was wrong about this one. NAND prices dropped for three quarters and were up 9% at Micron in this period. However, that 9% pricing increase was all from better mix. Management didn’t say it explicitly, but like-for-like NAND pricing appears to have been flat quarter-over-quarter. If we go back to the last cyclical low two years ago, NAND ASPs are up around 62%. Micron’s bit supply is up even more. The fourth calendar quarter of 2022 was the bottom for NAND bit shipments last cycle. Since that point eleven quarters ago, Micron’s NAND bit supply is up almost 150%. That is a 39% CAGR for two-and-a-half years. Considering the supply growth from Micron over that time frame, it is surprising the NAND market has held up as well as it has. To summarize for NAND at Micron, prices up strongly with bit shipments down some. The drop in shipments is to be expected given the massive increase in shipments seen from Micron, Samsung, and Hynix in the second calendar quarter.

The DRAM cycle is clearer over the last three years than is the NAND cycle. Excepting a single quarter where Micron saw their blended DRAM ASPs drop about 1%, their pricing has gone up for eight consecutive quarters. Over those same eight quarters, NAND saw a soft period over three of them, from which pricing is just now recovering. We are now two years into a really strong DRAM upturn. Both prices and bit shipments are up more than 100% off the bottom. Surprisingly over that same period, while NAND prices are up only 60%, NAND bit shipments from Micron have risen almost 150% off their bottom two-point-five years ago. The trough was deep, but that is a lot of supply. In the last ninety days, the NAND market has gone from somewhat weak to at the cusp of balanced.

Summary

I’ll start with the NAND market, since DRAM is getting most of the attention in memory today. It was promising for the NAND market when Micron announced their ASPs were up 9% sequentially. Then we learned this was all driven by better mix. Like-for-like, prices were flat sequentially. The NAND market is showing improvement, but it is behind DRAM in its upward trend. I was surprised when I looked back over this cycle and observed that Micron’s NAND bit shipments have grown at a 39% CAGR for the last ten quarters. Over that time pricing is up 150%.

Micron’s strong earnings and guidance are mostly from improved DRAM performance. The PC and mobile segments have come back to life, but they are still around the same level Micron thought they would be at the beginning of calendar 2025. HBM has grown rapidly over the last two years, but that was not the big driver of recent performance. The surprising strength seen in FQ4 was from better pricing in the non-HBM data center market. That is where pricing has improved and will continue to do so in the fourth quarter. A quarter ago, management was cautiously optimistic, saying the market will stay “constructive for the rest of 2025.” In this call, they are bullish and believe the market will be strong throughout 2026. Since the DRAM cycle bottomed two years ago (in the third calendar quarter of 2023,) Micron’s DRAM prices have increased by about 110%. That is a CAGR of 45% over those eight quarters. Over that same two year span, Micron’s bit supply has increased at a 36% CAGR. And supply is going to keep increasing. Micron spent more $15.9B in gross capital expenditures. They announced they will spend at least $18B this fiscal year, and probably more like $19B or $20B. Most of that spending will be on DRAM. This level of capital expenditures assumes revenue will continue to grow substantially from here. The vast majority of Micron’s DRAM production is on their 1-beta and 1-gamma nodes. Both are at mature yield. 1-gamma is a 30% bit density increase over 1-beta, which means converting wafers between those nodes creates a large supply increase. That is what a lot of the $18B+ in CapEx will go to this year. Management didn’t actually say $18B. They don’t want to spook investors.

HBM continues to be the center of the story for Micron and the DRAM industry writ large. The Big Three are all racing to grow their supply as fast as they can. Micron said they will match their overall DRAM market share (which is in the 22 – 23% range) sometime in the third calendar quarter of 2025. They had almost $2.0B of HBM sales in FQ4, out of $9.0B in total DRAM. Their market share of $9.0B is a little above $2.0B, so I think they probably hit that milestone sometime in August. Management has gotten quieter about their HBM production starting in FQ3 of this year. Based on revenue guidance for FQ1-26, their HBM revenue will be more than $2.3B, and probably closer to $2.5B. $2.5B would be 25% of Micron’s total DRAM revenue, above their market share. They said last quarter that HBM pricing was “steady.” I think that is probably still the case, because they had agreements on pricing with customers in 2025 and the market has not weakened such that those customers would push to change the terms. The way management has pulled back from how much information they previously shared on HBM makes me wonder about that market. Management says they expect to have higher HBM market share in 2026 than they did in 2025, which implies they will take share from Hynix and/or Samsung. The current strength of the DRAM market hinges on the insatiability of HBM demand. The Big Three DRAM companies are all charging forward headlong to ramp their HBM capacity. Can the market absorb all those new bits? If it can’t, recall what Micron’s management team kept saying for several quarters about the HBM trade ratio. They touted how each HBM wafer took out three non-HBM wafers, and that trade ratio was reducing non-HBM bit supply. Well, if the HBM market reaches saturation and those HBM wafers start getting converted back to non-HBM wafers, that trade ratio cuts the other way. For every HBM wafer that got swapped back over to non-HBM, three wafers will be created. That means oversupply in the HBM segment will spread quickly to other segments.

What happens from here? Management has confidence in the strength of the cycle. They expressed this by paying off $1B worth of debt and announcing a year-over-year capital expenditures increase of between 25% and 35%, mostly for DRAM. The stock is priced for growth to continue its strong recent pace. The current TTM free cash flow yield is only 2.8%, well below a high-yield savings account. If revenue grows to over $54B in FY26, I estimate Micron’s free cash flow yield will be 5.7% at the current stock price. That is a good yield, but is it enough to compensate for the risk that this time in the memory cycle isn’t different?

– Smooth Hughes (short MU)