Micron (MU.US) Q4 conference call: Management shouts “can't see the point of balance between supply and demand” 75% of shipments next year are locked in 2028 is tighter than 2027
The Zhitong Finance App learned that after the US stock market on September 30, EST, Micron Technology (MU.US) announced the results report for the fourth fiscal quarter of fiscal year 2026. Management sent an optimistic signal during the conference call, saying that demand for AI-driven memory continues to be strong, supply and demand will remain tight in 2027 and 2028, and the company's confidence in long-term growth has increased. Management said that more than 75% of shipments in FY2027 have been locked in, and SCA covered about 35% of sales by 2030; HBM's growth rate is faster than traditional DRAM, and pricing has increased significantly in 2027. The supply side is constrained by cleanroom construction, declining returns from technological transformation, and rising HBM trade ratios. The company has yet to see a balance between supply and demand.
Management's core opinion is that demand for AI-driven memory is far from seeing a balance between supply and demand. The market will continue to be tight in 2027 and 2028, and even 2028 may be tighter than previously anticipated. President and Chief Operating Officer Manish Bhatia said that server units continued to grow in 2027, and the rapid rise of proxy AI also brought demand for CPU-driven memory; more than 75% of the company's shipments in fiscal year 2027 have been locked in, and customer allocation negotiations have been extended to 2028. The supply side, on the other hand, is constrained by structural constraints such as declining returns from technological transformation, HBM's faster growth than traditional DRAM, rising HBM capacity consumption ratio (trade ratio), and long cleanroom construction cycles, so the company “cannot see when supply and demand will balance.”
HBM remains in focus. Management said that HBM shipments will grow faster than traditional DRAM, and its share of the industry's production capacity will continue to increase by 2028; HBM pricing for the 2027 calendar year has been significantly raised and will be reset at the beginning of the year to narrow the profit gap with traditional DRAM. In terms of HBM's market share, the company does not pursue a single number, but rather maintains roughly the same share of DRAM as a whole. On the HBM4E side, Scott DeBoer said that it has been over a year since the company and Nvidia co-designed the NV HBM. This will be the first major customized HBM product. The differences are reflected in power consumption, speed, and product quality, which is expected to bring high value and strong ROI.
In terms of strategic customer agreements (SCA), the company has signed 26 SCAs, covering about 35% of sales by 2030, covering DRAM and NAND; of these, the amount of DRAM is slightly less than 35%, and the NAND bits are slightly higher. The 10 new customers, ranging from small to large, are covered by all business units. About three-quarters of SCA revenue has a clear pricing framework, and one-quarter is open to regular negotiations or market-based pricing. Most have upper and lower limits, but the new agreement has been repriced according to current market conditions and tight future expectations.
In terms of capital expenditure, CFO Mark Murphy said that the increase in construction capital expenditure in FY2027 was mainly invested in clean rooms launched at the end of 2028 to accelerate production capacity space; however, the expenditure will not be immediately converted into digits; however, the company will flexibly allocate production capacity according to demand trends and equipment supply agreements. On the Chinese side, management said China's exposure continues to decline and is expected to drop to single digits in the 2027 fiscal year; the technology is ahead of the Chinese rival by at least 2 nodes. 1-gamma DRAM is already in the majority, 1-delta will be mass-produced in the second half of next year, and EUV is the key differentiation.
For mobile and clients, management acknowledged the month-on-month decline in bitumen shipments, but revenue grew due to improved pricing and combinations, and demand for high-end clients and flagship phones was strong. On the NAND side, the industry is expected to grow by about 20% mid-2028, and the market remains tight; Singapore clean rooms are used for advanced R&D and HBM pilots, resulting in lower supply growth than the industry this year, but G9 climbing will provide high ROI supply. Overall, management believes AI hardware demand, SCA lockdown, and supply constraints will support strong long-term financial performance, although slowing startup costs and price increases may partially offset the expansion of profit margins.
The following are the minutes of Micron Technology's fourth quarter earnings conference call for FY2026:
Satya Kumar
Investor Relations and Treasurer Corporate Vice President
Welcome to Micron Technology's Q4 FY2026 post-earnings analyst conference call. Today, I was joined by Manish Bhatia, President and Chief Operating Officer, Dr. Scott DeBoer, President and Chief Technology and Product Officer, and Mark Murphy, Chief Financial Officer.
Remind everyone that our discussions today include forward-looking statements on market supply and demand, market trends and drivers, expected performance and guidance, and other matters. These forward-looking statements are subject to risks and uncertainties, and may cause actual results to differ materially from today's statements. Please refer to our SEC filings, including the latest Form 10-K and the upcoming 10-Q Form for risks that may affect results.
Although we believe that the expectations reflected in the forward-looking statements are reasonable, there is no guarantee of future results, levels of activity, performance, or achievements. We are under no obligation to update any forward-looking statements to match actual results. Q&A is now open.
Q & A session
Benjamin Reitzes
Melius Research LLC
Congratulations on your new appointment, Scott. I want to talk about 2028. You added comments on 2028 and said that supply and demand in 2028 will be tighter than this year and 2027. I want to know what factors have changed and what that means for profit margins. You guys have given more margin reviews around 2027 than before, which is great, but I don't think 2028 profit margins will deviate from the level you've hinted at for 2027. Please explain more about 2028.
Manish Bhatia
President and Chief Operating Officer
Of course, Ben. I'll answer first, and Mark will say more if he has anything to add. Thanks for the congratulations too. For 2027 and 2028, we are seeing stronger demand drivers than ever before. We mentioned that server units will continue to grow into 2027. At the same time, proxy AI is growing rapidly and is forming a CPU-driven demand stream. As a result, when formulating our FY2027 plan, we have locked in more than 75% of shipments throughout the year, indicating stronger demand and enabling us to extend distribution negotiations with our customers until 2028.
This is one source of confidence: the demand situation in 2027 is stronger than previously anticipated. At the same time, as we engage in more long-term agreement (SCA) negotiations with our clients, including agreed extensions, we are more confident about our long-term prospects. The combination of the two gives us a very strong demand outlook until 2028.
On the supply side, the structural constraints we discussed earlier are still there: the return of technological transformation is declining; HBM will grow faster than traditional DRAM by 2028, which means that HBM's share in the industry's output capacity will increase; at the same time, not only for the current HBM's capacity consumption ratio (trade ratio), but the higher capacity consumption ratio of more complex HBM will also limit supply in the future.
In addition, it takes a long time to build, certify, and equip new clean rooms throughout the industry, and even if put into operation, it will take several quarters to form meaningful shipments. Therefore, based on comprehensive demand and supply prospects, we cannot currently see a point where supply and demand are balanced.
Mark Murphy
Executive Vice President and Chief Financial Officer
Ben, let me add one more point. As you indicated, in 2027, we have indicated that profit margins for the whole year will expand compared to the first quarter because prices continue to rise, but the increase is more moderate. We have also said many times before that the price increase will eventually slow down. At that time, in addition to a slower rate of price increase, the product portfolio will improve, and we will optimize the combination based on technology and product leadership.
As we said, market conditions are expected to remain tight and will provide support until 2028. Part of the offsetting of the price and the favorable combination is the start-up cost. But these are all matters we are managing, and we expect to maintain strong financial performance.
Benjamin Reitzes
Melius Research LLC
One more question. Sanjay took a prominent seat at Trump's luncheon. Is he optimistic about the industry's growth and ability to self-regulate? Does this support its optimistic guidance? Does the conference have more information and is memory frequently mentioned?
Manish Bhatia
President and Chief Operating Officer
Ben, Sanjay isn't online, I'm just explaining the conversation as I know it. We are certainly excited to participate in this forum. Micron was invited along with model companies and accelerator companies to show the importance of memory. The white paper framework released at the conference has been signed by many model companies. The framework is constructive for continuing to advance AI infrastructure, especially AI hardware infrastructure.
I've learned that the most discussed concept is that the way to manage some security issues is to use security solutions, which will require more advanced hardware, including higher performance, lower latency memory, and higher bandwidth memory. Because future gateways may be set up to manage security issues, their responsiveness will largely depend on the availability of high performance and low latency memory.
Melissa Weathers
Deutsche Bank Research Division
Congratulations to the two new recruits as well. Over the past few quarters, you have given your views on HBM's general TAM for 2028 and 2030. Clearly, the pricing landscape has changed. Would you guys like to give out new official figures for HBM TAM? In terms of direction, can you help us estimate how much of this comes from bits and how much from pricing? And an updated take on the size of the market?
Manish Bhatia
President and Chief Operating Officer
Melissa, thank you. We are not currently updating this TAM outlook. However, we have said that HBM shipments are expected to grow faster than traditional DRAM, which means that HBM's share of the industry's production capacity will continue to rise by 2028. In terms of pricing, at least for us, we have drastically raised HBM pricing for the calendar year 2027 and will reset it at the beginning of the calendar year to close the profit gap with traditional DRAM. Other than that, we have not commented on the specific future of HBM TAM. The market continues to grow and remains an important driver, and HBM deployment continues to be the key to unlocking the potential of a wider range of AI. As such, it is an important part of the market.
Melissa Weathers
Deutsche Bank Research Division
Understood. In terms of market share, have you updated on HBM's share targets? In the past, you said you wanted to increase to a market share close to the company average, that is, around 20%. Is that still the case today? Also, I note that you mentioned HBM4E and cooperation with Nvidia in your advance remarks. Please provide more information on 2027 and HBM4E customer developments.
Manish Bhatia
President and Chief Operating Officer
I'll answer the first one. HBM4e was answered by Scott because he was responsible for developing this product. As far as HBM market share is concerned, we reached a milestone about a year ago, where HBM's market share reached the overall DRAM market share. At the time, we stated that the target would change according to a number of factors. Other than indicating that HBM's share is expected to be roughly equivalent to the overall DRAM share, we have not updated specific share targets, nor are we targeting a certain number as an absolute target.
Shares will fluctuate due to a number of factors. However, as mentioned in the previous question, HBM is a very important part of the market, enabling us to be close to the cutting edge of accelerator platforms designed and deployed by customers, and is also a key driver for other AI to deliver on its promises and potential.
Scott DeBoer
President and Chief Technology and Product Officer
I added about the partnership with Nvidia. This will be the first major custom HBM product on the market. We have been working with Nvidia on HBM4E for over a year, the so-called NV HBM. We believe there is a significant opportunity because the product was co-designed with key customers, making it significantly more valuable than the standard HBM4E. I think this will have a real impact on the industry and show how future systems can optimize products.
Atif Malik
Citigroup Research Division
The first question is about 26 SCAs, which will account for 35% of sales by 2030. Do these cover both DRAM and NAND? Can it be split?
Manish Bhatia
President and Chief Operating Officer
Atif, yes. We're not splitting, but the SCA agreement does cover DRAM and NAND and will last until 2030. Specifically, the amount of DRAM is slightly less than 35%, and the amount of NAND bits is slightly higher. Going forward, we have more available capacity. As we said, that number is likely to be higher in the future as negotiations on the SCA continue.
Atif Malik
Citigroup Research Division
Understood. Regarding the impact of competition in China, can you first confirm that China's sales exposure is small? Can Scott comment on how his Chinese rivals are closing the gap technologically?
Manish Bhatia
President and Chief Operating Officer
First question, yes. Our exposure to China has continued to decline over the past few years and quarters, and is expected to fall to the single-digit range in the 2027 fiscal year.
Scott DeBoer
President and Chief Technology and Product Officer
On the technical side, our technology is currently at least 2 nodes ahead of our Chinese rivals. It's important to emphasize that we're focused on staying ahead of technology and truly differentiating our products and how we compete. As Manish stated previously and during the conference call, our 1-gamma DRAM technology already accounts for the majority of our digits and is expected to become the largest node in the company's history. The technology relies on EUV. The next 1-delta is progressing well, and we are focusing on mass production in the second half of next year. EUV technology is critical to all future advanced DRAM nodes. Our expertise in this field, including cooperation with technology suppliers, masking technology, etc., continues to be a key differentiator for Micron.
Karl Ackerman
BNP Paribas Research Division
You're seeing strong demand across most product portfolios, but mobile and client unit shipments seem to have declined for the second straight quarter this quarter. Are higher memory prices weakening demand in this area? Additionally, SCA adoption is slow in this field; does your SCA growth come from this customer base?
Manish Bhatia
President and Chief Operating Officer
Thanks, Karl. We did see a month-on-month decline in mobile and client business segments, but revenue grew as pricing increased and the mix improved. It's important to note that the high-end segment of clients and flagship smartphones has strong demand for higher content, higher performance solutions, and this is an area we are focusing on. Driven by this, we expect PC and mobile industry revenue to grow despite declining unit sales.
On the SCA side, we have SCA in all business units, including mobile and client. We are not specifically splitting, but maintaining SCA helps maintain diversified supply to all end markets.
Karl Ackerman
BNP Paribas Research Division
Understood. Just ask Scott a question. How do you view the competitiveness of your internal optimization base chip on HBM4E compared to its peers? Some customers are using custom solutions. Does complexity and economic value flow primarily to computing customers or HBM suppliers?
Scott DeBoer
President and Chief Technology and Product Officer
Clarify a bit first. On the HBM4E, we co-designed it with Nvidia, but we didn't use an internal base chip like the HBM4. The HBM4E co-design is based on the foundry process and is used for both customized products and JEDEC standard products.
The differentiation aspect, as we have shown on HBM products in the past, is ultimately reflected in power consumption, speed performance, and margin of cooperation between the product and the customer. The performance of different suppliers in previous generations of products was not the same, and we believe this will continue to be Micron's advantage in terms of product quality and capability.
Manish Bhatia
President and Chief Operating Officer
Karl, when it comes to economic value, HBM is a high-end product. As Scott stated, the customized NV HBM is also expected to be a high value product. We believe HBM will continue to contribute and become a high ROI product.
James Schneider
Goldman Sachs Group Research Division
Congratulations Manish and Scott. I want to know about the 10 new customer SCAs that were signed this quarter. What are the customers asking for? Obviously they want supplies, and for a longer period of time. But has the pricing structure changed? Given the tight forecast for the 2027 and 2028 calendar years, are you less inclined to require fixed upper and lower price limits if more upside can be captured in the next few years?
Manish Bhatia
President and Chief Operating Officer
Thanks, Jim. The SCA framework is similar, but the difference is that negotiations reflect current market conditions and future pricing prospects. As a result, pricing has been moving upward. These factors have been reflected in discussions with customers, and the previous agreement was set under CQ2 market conditions. We have explained that about three-quarters of SCA revenue has a clear pricing framework, and about one-quarter is open to regular negotiations or pricing according to market dynamics.
The overall framework is similar. Most pricing frameworks have upper and lower limits, but the new agreement has taken into account current market conditions and the tension we see in the future during negotiations.
James Schneider
Goldman Sachs Group Research Division
Understood. Does this mean that all upper and lower price limits have been reset to the current higher market conditions? Or are the pricing conditions also different in nature? Also, are you disclosing the signing of the initial hyperscale customer? Are these included in the TAM?
Manish Bhatia
President and Chief Operating Officer
There are a variety of frameworks. What I'm saying is that most frameworks that have pricing have upper and lower limits, but we continue to use multiple frameworks. Since the last conference call, 10 new SCAs have been signed, covering small to large customers. We have not specifically split, but only stated that currently all business divisions have SCA, and that the size of customers ranges from small to large, including the 10 newly signed ones. There are 26 in total.
Christopher Caso
Wolfe Research, LLC
First question, please discuss the comparison between capital expenditure and capital expenditure for fab construction and tool procurement. I've heard from the comments that construction capital spending is growing faster, and I'm not sure if I understand this correctly. Clean room space limitations are limiting the ability to introduce tools this year. However, the increase in construction capital expenditure is worth watching, as it will not bring in bits of production until at least 2029, and possibly longer.
Manish Bhatia
President and Chief Operating Officer
Chris, I'll answer first, Mark added. That's true. The industry's main constraint is clean room space, because the strong increase in AI demand is relatively recent, and it takes a long time to build a clean room. So we're focusing on this area. You correctly pointed out that the capital expenditure for construction in fiscal year 2027 increased compared to the previous plan, and most of it was used for clean rooms launched at the end of 2028 or later. This not only explains the length of the construction cycle, but also why construction investment must be initiated now, and shows our confidence in long-term needs, not only from recent market trends, but also from discussions with the SCA structure and customers extending their commitment beyond 2030. These SCAs are transformative for us, matching supply with future demand, and supporting investment with confidence.
Mark Murphy
Executive Vice President and Chief Financial Officer
Chris, add one more point. Most of the increase was indeed construction capital expenditure, mainly to accelerate cleanroom space availability in 2028 and beyond. We expect this trend to continue beyond 2027. The important point you mentioned is that this expense doesn't immediately convert to bitcoins. These fabs will be equipped and produce wafers when needed, based on our market judgment and SCA. These SCAs help us keep up with the pulse of the market and ensure return on capital expenditure.
Manish Bhatia
President and Chief Operating Officer
Another point, Chris, we will equip clean rooms and build production capacity according to demand trends. We have been implementing long-term supply agreements with equipment suppliers to ensure equipment is available when needed. But of course, clean rooms will still be equipped and production capacity will be built according to demand trends at the time.
Christopher Caso
Wolfe Research, LLC
Understood. The next question concerns the impact of CPU strength on overall bit demand and the balance between supply and demand. This is probably the biggest incremental accident since the beginning of the year. CPUs don't have the trade ratio effect of HBM. How much does this contribute to the imbalance between supply and demand?
Manish Bhatia
President and Chief Operating Officer
The recognition that proxy workloads execute on the CPU is an important driver. LP, DDR memory, and SSD all have high attachment rates to support these proxy workflows. These workflows have begun to be implemented and are driving real value on the enterprise and consumer side. This is also one of the reasons for the strong growth in server units. As we mentioned, the growth rate is about ten percent higher. This also shows that logic chip growth is another vector to take advantage of AI trends. As a result, this pushes up demand for logic chips and enters the overall demand for AI computing. At the same time, it creates more constraints on DRAM, indicating that DRAM is the main constraint, not logic or data center power.
I mentioned that there are various software implementations for enterprise proxy workloads. Meta's Muse progressed rapidly just a few weeks after launch, showing how fast proxy workloads are achieving real value for consumers.
Joseph Moore
Morgan Stanley Research Division
Supply growth will slow next year, which is somewhat surprising in the context of capital expenditure. You mentioned the inhibitory effect of HBM, but the difference should not change that much. Please explain why supply is slowing in the face of high capital expenditure? What are the factors?
Manish Bhatia
President and Chief Operating Officer
Joe, are you asking about DRAM and NAND?
Joseph Moore
Morgan Stanley Research Division
Yes, but mostly DRAM.
Manish Bhatia
President and Chief Operating Officer
We've shown that HBM is growing faster than traditional DRAM. As the industry shifts from HBM3 to HBM4 and towards HBM4E in late 2027, the trade ratio continues to rise. If both occur simultaneously, they will inhibit the ability to grow bits. At the same time, the bit growth for us and other industry players comes from new technology nodes. The timing of the transformation affects bit growth, and the return of new nodes decreases.
These are all factors limiting supply. But the main factor is still the cleanroom space of all manufacturers. Although our Idaho plant will produce wafers for the first time in the mid-calendar year 2027, and other vendors in the industry may open clean rooms, significant supply growth will take several quarters since then. These are the reasons we think supply shipments in the DRAM industry will decrease next year.
Joseph Moore
Morgan Stanley Research Division
Makes sense. If you are conservative about supply in the industry and supply more next year, there seems to be a large amount of suppressed demand. Some AI racks have reduced specifications due to necessity. If supply increases, will the specifications be raised again? Am I being overly optimistic?
Manish Bhatia
President and Chief Operating Officer
That's the point. Sanjay mentioned during the main conference call that we do think customers are choosing to maximize compute chip shipments with the available memory supply. This creates a potential need for more memory to be attached to these computing chips, thereby improving system performance and the end customer experience. As a result, if more memory is available, it can easily be used for higher content growth in AI workloads, whether attached to accelerators or CPUs.
Mehdi Hosseini
Susquehanna Financial Group, LLLP, Research Division
A few follow-up questions. You emphasized that 2026 NAND bit shipments were below the industry average. What should be expected from 2027 to 2028? Can NAND bit demand or shipments increase by 25% on average in the industry?
Manish Bhatia
President and Chief Operating Officer
Mehdi, frankly, we haven't commented on DRAM or NAND that far away. We do expect overall conditions for NAND to be tight, even though the industry expects mid-term growth of 20% in the 2027 and 2028 calendar years. Our supply growth is affected by several factors: we are currently using some clean room space in Singapore for advanced R&D lines for future NAND growth; at the same time, we are preparing to climb the slope at the Singapore HBM plant next year, and some existing clean rooms will be used to pilot HBM operations. This has led to lower supply growth than the industry this year. However, we are confident in the technology, and we expect continuing to provide a high ROI and cost-effective supply to the G9 hill climbing. Additionally, the new cleanroom, which broke ground earlier this year, will be launched in the second half of the 2028 calendar year.
Mark Murphy
Executive Vice President and Chief Financial Officer
Mehdi, time is a matter of time. Please ask the second question before I comment.
Mehdi Hosseini
Susquehanna Financial Group, LLLP, Research Division
Sorry, Mark. Quick question, I'd like to know what the team thinks. Observe DRAM, especially at the wafer level, and the device is the same. I think wafer-level DRAM is replaceable, and there are differences in the back end. This kind of wafer-level substitutability has not been seen before, as past cycles were driven by a single product and a high concentration of customers. Am I correct in assuming that wafer-level DRAM replaceability enables you to better manage DRAM costs?
Scott DeBoer
President and Chief Technology and Product Officer
Let's talk a little bit first. I think this is partly true, but I'm probably of the opposite opinion. If you look at front-end variability, or the differences we have made in DRAM processes to optimize HBM, high-performance SOCAMM, LPDRAM, and DDR6, the current differences are probably greater than at any time in DRAM history. Different products need to be built on the same node, and more differentiation is built in. Manish, did you add?
Manish Bhatia
President and Chief Operating Officer
In the short term, we run different products on the same manufacturing line, which really helps to adjust the mix. Instead of running in different fabs, we can run on the same line. But as Scott explains, each product has its own optimization vector. HBM requires unique process steps to optimize bandwidth and TSV; DDR and LP also each have unique process steps. The most important part of fungibility is that we can flexibly adjust where the demand is, or even adjust the mix to meet customer requirements. This is all within the same fab and is the most important part. But I don't think it necessarily helps with the costs.
Mark Murphy
Executive Vice President and Chief Financial Officer
Mehdi, I think the conference call is nearing its end. I'm happy that the issues focus on the long-term solid foundation and performance of technology and business. I thought there would be some routine issues earlier, but now I've added: The first quarter guidelines include a month-on-month increase in DRAM and NAND bits and a double-digit increase in costs, so please pay attention when modeling. Also, excluding the impact of incentive pay, the year-on-year growth in R&D will exceed the $1 billion mentioned in the previous quarter and over $1 billion in 2027, as we increase R&D activities. Once again, make sure this information is used for modeling.