Micron (MU.US) Q4 earnings call: Management declares "No sign of supply-demand balance," 75% of shipments for next year already locked in; 2028 expected to be tighter than 2027

According to Zhitong Finance APP, after the US stock market closed on September 30th Eastern Time, Micron Technology (MU.US) released its financial results for the fourth quarter of fiscal year 2026. During the earnings call, management conveyed an optimistic outlook, stating that AI-driven memory demand remains robust, supply-demand is expected to remain tight in 2027 and 2028, and the company has strengthened its confidence in long-term growth. Management indicated that over 75% of shipments for fiscal year 2027 are already locked in, SCA covers about 35% of sales through 2030; HBM is growing faster than traditional DRAM, and 2027 pricing has been significantly increased. On the supply side, constraints include cleanroom buildout, diminishing returns on technology transitions, and a rising HBM trade ratio. The company does not yet see a point of supply-demand balance.

The management’s core viewpoint: AI-driven memory demand is far from reaching equilibrium, with market tightness expected to persist in 2027 and 2028—2028 could even be tighter than previously anticipated. President and Chief Operating Officer Manish Bhatia stated server units will continue growing into 2027, while proxy AI’s rapid emergence is also driving CPU-based memory demand. Over 75% of shipments for fiscal 2027 are secured, and customer allocation negotiations have been extended into 2028. On the supply side, structural constraints remain: diminishing technological transition returns, HBM growth outpacing traditional DRAM, rising HBM production consumption ratio (trade ratio), and long build-out cycles for cleanrooms. As such, the company “does not see when supply and demand might balance.”

HBM remains in focus. Management stated that HBM shipment growth will outpace traditional DRAM, with HBM’s share of industry capacity continuing to rise into 2028; 2027 calendar year HBM pricing has been significantly increased, and will be reset at year-start to narrow profitability gap with traditional DRAM. In terms of HBM market share, the company does not pursue a singular number but aims to roughly maintain parity with overall DRAM market share. Regarding HBM4E, Scott DeBoer mentioned that the company has been co-designing NV HBM with Nvidia for over a year, which will be the first major custom HBM product, differentiated by power efficiency, speed, and product quality, expected to deliver high value and strong ROI.

Regarding Strategic Customer Agreements (SCA), the company has signed 26 such agreements, covering around 35% of sales through 2030, including both DRAM and NAND; DRAM volume is just below 35%, while NAND bits are slightly higher. The newly signed 10 customers range from small to large, covering all business units. About three-quarters of SCA revenue has a clear pricing framework, the remaining quarter is either subject to regular negotiation or market-based pricing. Most agreements have set price bands, but new agreements have been repriced based on current market conditions and expectations of future tightness.

On capital expenditures, CFO Mark Murphy stated that capex increases in fiscal 2027 are primarily directed toward cleanrooms coming online in late 2028 and beyond, to accelerate capacity availability; however, this spending will not immediately translate to bits. The company will flexibly provision capacity based on demand trends and equipment supplier agreements. For China, management reported that China exposure continues to decline and is expected to fall to single digits in fiscal 2027; the company maintains a lead of at least 2 nodes over Chinese competitors, with 1-gamma DRAM now the majority of bits, 1-delta scheduled for mass production in the second half of next year, and EUV as a critical differentiator.

For mobile and client segments, management acknowledged sequential bit shipment declines, but revenue grew on pricing and mix improvements, with strong high-end client and flagship smartphone demand. For NAND, the industry is expected to show around mid-20% growth in 2027-2028, with tight market conditions maintained. The Singapore cleanroom is used for advanced R&D and HBM pilots, resulting in lower supply growth than the industry, but G9 ramp will provide high-ROI supply. Overall, management believes AI hardware demand, SCA lock-ins, and supply constraints will support long-term robust financial performance, though ramp-up costs and tempered price increases may partially offset margin expansion.

Below is the Micron Technology Q4 FY2026 earnings call transcript:

Satya Kumar

Vice President, Investor Relations and Treasurer

Welcome to Micron Technology's FY2026 fourth quarter post-earnings analyst call. Here with me today are President and Chief Operating Officer Manish Bhatia, President and Chief Technology & Product Officer Dr. Scott DeBoer, and Chief Financial Officer Mark Murphy.

Please note, today's discussion includes forward-looking statements regarding market supply-demand, trends and drivers, projected performance, guidance, and other matters. These forward-looking statements involve risks and uncertainties, and actual results may differ materially from what is discussed today. Please refer to our filings with the SEC, including our latest 10-K and upcoming 10-Q forms, for risks that could affect our results.

While we believe the expectations reflected in these forward-looking statements are reasonable, we cannot guarantee future results, activity levels, performance, or achievements. We are not obligated to update any forward-looking statements to conform to actual results. With that, we’ll open up for Q&A.

Q&A Session

Benjamin Reitzes

Melius Research LLC

Congratulations on your new role, Scott. I’d like to discuss 2028. You added comments regarding 2028, saying supply-demand would be even tighter than this year and 2027. What changed, and what does this mean for margins? You made more margin comments about 2027, which is good, but I don’t assume 2028 margins deviate from what you implied for 2027. Could you elaborate on 2028?

Manish Bhatia

President and Chief Operating Officer

Sure, Ben, I’ll take this and Mark can add if necessary. Thanks for the congratulations. Regarding 2027 and 2028, we’re seeing stronger demand drivers than before. Server units will continue to grow into 2027. At the same time, proxy AI is rising rapidly, leading to CPU-driven demand. When planning FY27, we’ve locked in over 75% of shipments for the year, showing sustained demand and enabling negotiations with customers for allocations into 2028.

This gives us confidence: 2027 demand is stronger than previously expected. Additionally, as we’ve negotiated more long-term SCA agreements with customers, including recent extensions, our confidence in the long-term outlook has improved. Both factors mean we have a strong demand view out to 2028.

On supply, structural constraints remain as we previously discussed: diminishing returns from technology migration; HBM growing faster than traditional DRAM, which means HBM’s share of industry output will continue rising into 2028; and not just with current HBM trade ratios, but also more complex, higher HBM trade ratios in the future, which will further constrain supply.

Further, new cleanrooms in the industry take time to build, qualify, and equip. Even after going into operation, it takes several quarters to achieve meaningful shipments. Given demand and supply outlooks, we don’t currently see a point of supply-demand normalization.

Mark Murphy

Executive Vice President and Chief Financial Officer

Ben, just to add, as you noted, for 2027 we expect full-year margins to expand from the first quarter, mainly because prices continue rising, though at a slower pace. We’ve said previously that price increases will eventually moderate. At that point, besides modest price gains, mix improvement will also help, as we optimize based on technology and product leadership.

As mentioned, tight market conditions should persist through 2028, supporting results. Start-up costs do partially offset the benefits of price and mix, but we expect to maintain strong financial performance overall.

Benjamin Reitzes

Melius Research LLC

And a follow-up: Sanjay had a prominent seat at Trump’s luncheon. Is he optimistic about industry growth and self-regulation? Does this underpin the company’s bullish guidance? Any further details from the meeting, or frequent mention of memory?

Manish Bhatia

President and Chief Operating Officer

Ben, Sanjay is not on the line, but to the best of my knowledge, yes, we were pleased to participate. Micron was invited alongside model and accelerator companies, underscoring the importance of memory. The white paper framework from the event, signed by many model companies, is constructive for ongoing advancement of AI infrastructure—especially hardware.

I understand a frequently discussed idea was managing safety issues through secure solutions, requiring more advanced hardware, including higher-performance, lower-latency, and higher-bandwidth memory. Gateways that might be set up to manage security issues will rely heavily on the availability of high-performance, low-latency memory.

Melissa Weathers

Deutsche Bank Research

Congratulations to both of you as well. In past quarters, you’ve offered views on HBM’s total TAM for 2028 and 2030. The pricing environment has obviously changed. Could you give an updated HBM TAM number? Directionally, can you help us estimate how much is bits versus price? What’s your latest view of market size?

Manish Bhatia

President and Chief Operating Officer

Thanks, Melissa. We’re not updating that TAM outlook at this point. But as noted, HBM shipments are expected to outgrow traditional DRAM, so HBM’s share of industry capacity will keep rising into 2028. For us at least, we have significantly increased 2027 HBM calendar year pricing, and will reset it at the start of the year to narrow the profitability gap with DRAM. Beyond that, we haven’t commented on HBM TAM specifics. The market’s growth continues to be a major driver; the deployment of HBM is a key to unlocking wider AI potential and is an essential market component.

Melissa Weathers

Deutsche Bank Research

Understood. On market share, do you have an updated HBM share target? Before, you aimed for company average, around just over 20%. Is that still the case? Also, you mentioned HBM4E and the Nvidia partnership in the prepared comments. Can you share more about 2027 and HBM4E customer progress?

Manish Bhatia

President and Chief Operating Officer

I’ll answer the first, Scott will cover HBM4E. A year ago, we hit a milestone where our HBM market share aligned with our DRAM market share overall. At the time, we indicated that the target would change based on multiple factors. We haven’t provided a fresh specific target, other than that HBM’s share is expected to mirror overall DRAM share—we’re not targeting a fixed number.

Share will fluctuate for many reasons. But as discussed, HBM is a vital part of the market; it keeps us close to customers’ design and accelerator platforms and is a key enabler for AI’s promise and potential.

Scott DeBoer

President and Chief Technology & Product Officer

Regarding the Nvidia partnership, this will be the first major custom HBM product in the market. We’ve been working with Nvidia on HBM4E (the so-called NV HBM) for over a year. We see significant opportunity because, by co-designing with a key customer, the product will deliver substantially more value than standard HBM4E. This will set a real industry milestone and showcase how future systems can optimize products.

Atif Malik

Citi Research

First question regarding the 26 SCAs, covering 35% of sales through 2030. Does this include both DRAM and NAND? Any breakdown?

Manish Bhatia

President and Chief Operating Officer

Atif, yes. We don’t break it down, but SCA agreements certainly cover both DRAM and NAND through 2030. Specifically, DRAM is slightly less than 35%, NAND bits a bit more. Looking ahead, we have more availability, and as we continue negotiations, this number may rise.

Atif Malik

Citi Research

Got it. On China’s competitive impact, first, can you confirm your China sales exposure is small? Scott, could you comment on Chinese competitors’ progress on closing the technology gap?

Manish Bhatia

President and Chief Operating Officer

To the first, yes. Our China exposure has declined continually over recent years and quarters, and we expect to be in single digits by fiscal 2027.

Scott DeBoer

President and Chief Technology & Product Officer

Technologically, we’re at least two nodes ahead of Chinese competitors. It’s important to stress our focus on maintaining leadership and true product and competitive differentiation. As Manish discussed, our 1-gamma DRAM is the majority of bits and is expected to be the company’s largest-ever node, depending on EUV. The next-gen 1-delta node is progressing well, scheduled for 2H next year ramp. EUV is pivotal to all advanced DRAM nodes. Our expertise in this area—including tech supplier partnerships, mask technology, etc.—continues to be a core Micron differentiator.

Karl Ackerman

BNP Paribas Research

You’re seeing robust demand in most portfolios, but for mobile and client, bit shipments seem down for a second straight quarter. Are higher memory prices hurting demand there? Also, SCA adoption has been slower in this segment. Is SCA growth coming from this customer group?

Manish Bhatia

President and Chief Operating Officer

Thanks, Karl. We do see sequential bit declines in mobile and client, but revenue grew due to higher pricing and better mix. High-end client and flagship smartphone demand remains strong for higher content and higher-performance solutions, which is our main focus. This demand means that, despite unit declines, PC and mobile industry revenues will grow.

On SCA, all business units have SCAs, including mobile and client. We don’t break out specifics, but maintaining SCAs enables diversified supply to all end-markets.

Karl Ackerman

BNP Paribas Research

Understood. A question for Scott. How do you view your own base die optimization for HBM4E compared to peers? Some customers are adopting custom solutions. Does complexity and value gravitate mainly to compute customers or HBM vendors?

Scott DeBoer

President and Chief Technology & Product Officer

Just to clarify, for HBM4E, this is co-designed with Nvidia and isn’t using an internal base die as in HBM4. The HBM4E co-design is being done with the foundry process, both for custom and JEDEC standard products.

Differentiation, as our past HBM products have shown, ultimately comes down to power, speed, and how much margin the product provides in customer collaboration. Previous generations saw clear differences across vendors, and we expect this to remain a strength for Micron in product quality and capabilities.

Manish Bhatia

President and Chief Operating Officer

Karl, on value, HBM is a high-value product, and as Scott noted, custom NV HBM is also set to be a high-value product. We believe HBM will continue to contribute and be a high-ROI offering.

James Schneider

Goldman Sachs Research

Congrats to Manish and Scott. I’d like to understand the 10 new SCA customers you signed this quarter. What are they asking for? Obviously, they want supply and duration, but has the pricing structure changed? With expectations for market tightness in 2027 and 2028, are you less inclined to set fixed price bands, to preserve upside?

Manish Bhatia

President and Chief Operating Officer

Thanks, Jim. The SCA framework is similar, but negotiations reflect current market conditions and future pricing. The price direction has been up, and that’s reflected in customer conversations. Prior agreements were set under CQ2 market conditions. About three-quarters of SCA revenue has defined price frameworks, the other quarter is subject to periodic negotiation or market dynamics.

The overall structure is the same; most have pricing bands, but new agreements have been repriced for current market conditions and expected future tightness.

James Schneider

Goldman Sachs Research

Understood. So does this mean all price bands are reset higher, or are there other differences in pricing terms? Also, did you disclose initial hyperscale customers among those signed? Are these included in TAM?

Manish Bhatia

President and Chief Operating Officer

There are multiple frameworks. Most price-based frameworks have bands, but we use several. Of the 10 new SCAs since the last call, customers range from small to large, and we didn’t provide line-item detail, only that all business units now have SCAs and customer sizes vary, including the most recent 10. There are 26 in total.

Christopher Caso

Wolfe Research, LLC

First question, please talk about capex and how fab construction capex compares to tool purchases. I heard construction capex is rising faster, just want to be clear. Cleanroom space seems to be constraining tool installs this year. Why is construction capex a focus if it won’t yield bits until 2029 or later?

Manish Bhatia

President and Chief Operating Officer

Chris, I’ll start and Mark can add. That’s right; the main industry constraint is cleanroom space because AI demand ramped so quickly, and it takes time to build cleanrooms. That’s the focus. You’re right, for FY27, construction capex is up mainly for cleanrooms coming online after late 2028. This shows both the length of the build cycle and our confidence in demand, which is supported by recent market trends and by the SCA structure and commitments out beyond 2030. These SCAs are transformative for us, matching supply to demand and supporting confident investment.

Mark Murphy

Executive Vice President and Chief Financial Officer

Chris, to add: most of the increase is construction capex, mainly to accelerate cleanroom availability beyond 2028. This trend will continue. Importantly, this spending does not immediately add bits. Fabs will be equipped to produce wafers as needed, in line with market judgment and SCAs. SCAs help us sense the market and ensure ROI on capex.

Manish Bhatia

President and Chief Operating Officer

One more thing, Chris. Capacity will be built and equipped based on demand trends. We always have long-term supply agreements with equipment vendors to secure tools when required, but equipment deployment will still follow demand.

Christopher Caso

Wolfe Research, LLC

Understood. Follow-up on how CPU intensity impacts overall bit demand and supply-demand balance. This may have been the biggest incremental surprise this year. CPUs don’t have the HBM trade ratio. How big of an effect does this have on the imbalance?

Manish Bhatia

President and Chief Operating Officer

Proxy workloads running on CPUs are a key factor. LP, DDR memory, and SSDs are all highly attached to these proxy workflows—they’re being implemented and are driving real business value in enterprise and consumer. That’s one reason server units are growing strongly, with high double-digit increases. There’s also logic chip growth as another vector to capitalize on the AI trend, which means an increased need for logic chips and more DRAM constraints, not logic or datacenter power, as the main bottleneck.

Proxy workloads for enterprises are being rolled out via multiple software methods. After Meta’s Muse launch, implementation has progressed rapidly, showing how proxy workloads can quickly deliver tangible value to consumers.

Joseph Moore

Morgan Stanley Research

Supply growth next year looks muted, which is surprising given capex. You cited HBM’s impact, but the delta shouldn’t be that big. Why is supply lower with high capex? What factors are at play?

Manish Bhatia

President and Chief Operating Officer

Joe, are you referring to both DRAM and NAND?

Joseph Moore

Morgan Stanley Research

Yes, but mainly DRAM.

Manish Bhatia

President and Chief Operating Officer

We’ve described that HBM is growing faster than DRAM. As the industry transitions from HBM3 to HBM4, then to HBM4E later in 2027, the trade ratio goes up. These transitions constrain bit growth. Also, for us and the industry, bit output additions are from new technology nodes, but transition timing impacts bit growth and newer nodes yield diminishing returns.

These all limit supply. But the main constraint is still cleanroom space industrywide. Even though our Idaho fab will produce initial wafers by mid-2027, and other industry players may launch new cleanrooms, significant supply growth will take several more quarters. That’s why we anticipate DRAM industry supply shipments to be down next year.

Joseph Moore

Morgan Stanley Research

Makes sense. If you’re conservative on supply and there’s more supply next year, that suggests pent-up demand. Some AI racks have had to reduce specs for now—if supply increases, are specs likely to rise? Am I too optimistic?

Manish Bhatia

President and Chief Operating Officer

That’s exactly right. As Sanjay said on key calls, we do believe customers are maximizing compute chip shipments based on available memory. This creates incremental memory demand to add to those compute chips, lifting system performance and end-user experience. If more memory is available, it will easily go into higher memory content for AI workloads, on accelerators or CPUs.

Mehdi Hosseini

Susquehanna Financial Group, LLLP, Research Division

A few follow-ups. You emphasized that 2026 NAND bit shipments are below the industry average. How should we expect 2027 to 2028? Could your NAND bit demand grow at the industry average of 25%?

Manish Bhatia

President and Chief Operating Officer

Mehdi, honestly, we haven’t commented that far out for DRAM or NAND. Overall, we do expect continued tight NAND conditions even if the industry grows at about mid-20% for 2027 and 2028. Our supply growth is affected by a few factors: we’re using part of our Singapore cleanroom for advanced R&D to support future NAND growth, and prepping for the Singapore HBM pilot ramp next year, so some cleanroom is dedicated to piloting HBM. This results in lower supply growth this year than the industry. But we have confidence in our technology; ongoing G9 ramp will supply high-ROI, cost-effective output. Plus, a new cleanroom that broke ground this year will come online in 2H 2028.

Mark Murphy

Executive Vice President and Chief Financial Officer

Mehdi, in the interest of time, ask your second question—I’ll then comment.

Mehdi Hosseini

Susquehanna Financial Group, LLLP, Research Division

Sorry, Mark. Quickly, I just wanted the team’s view. If you look at DRAM at the wafer level, the equipment is the same. My thinking is that wafer-level DRAM is interchangeable, while differentiation is in the backend. This was not seen in prior cycles, as past cycles were driven by single products and customer concentration. Am I right that wafer-level interchangeability lets you manage DRAM cost better?

Scott DeBoer

President and Chief Technology & Product Officer

Let me start—that’s partly correct, but I look at it differently. If you examine front-end variability, or the process differences we use to optimize HBM, high-performance SOCAMM, LPDRAM, DDR6; differences at the process level today may be greater than at any previous time in DRAM history. We have to build various products on the same node, with more differentiation built in. Manish, want to add?

Manish Bhatia

President and Chief Operating Officer

In the short term, running multiple products on the same line definitely helps adjust mix, so we don’t need separate fabs; we can adjust on a single line. But as Scott outlined, each product optimizes differently: HBM for bandwidth and TSV, needing unique process steps; DDR and LP each having their own. Flexibility’s most important aspect is letting us adjust to demand and meet customer expectations in the same fab, but I wouldn’t say it necessarily helps costs.

Mark Murphy

Executive Vice President and Chief Financial Officer

Mehdi, as we wrap up, I'm pleased questions focused on the long-term technical and business foundation. I expected more routine queries earlier—let me add: Q1 guidance assumes single-digit sequential DRAM and NAND bit growth, with double-digit cost increases, so please factor that into your models. Also, net of incentive comp, R&D will be up more than $1 billion year-over-year as guided last quarter, and stay over $1 billion in 2027 due to increased R&D activities. Just making sure this is modeled as well.