Ladies and gentlemen, welcome to KeyBanc Capital Markets Annual Technology Leadership Forum. For 27 years, this is where tech's visionaries, investors, and industry leaders come together to shape what's next. Please welcome Pat Kratus, managing director and group head of technology, to the stage.
Thank you so much. I'm going to need you every 8:00 A.M. to introduce me in the morning. That's a way to wake up. Welcome everybody to our tech conference. It means a lot to us to have you make the time and effort to come here in the second week of August. We know that this necessarily isn't on every one of your plans. We know that you're going to get a lot of great value from it, but from my perspective, we know it takes a lot of effort, and so it's our job to make sure that we put on a great program for you. Also, thank you for giving us the opportunity to work with you.
I see a lot of familiar faces here, some people who've been coming to this conference for the entirety of its 25-year run when it was in Vail and now here in Deer Valley. There are some new faces. For those of you that we actually do formally work with, thank you for giving us that opportunity. For those of you where our relationships are just building, to use a bad sports analogy, just give us a chance to swing the bat. We have some tremendous professionals on the banking research, and throughout the investment bank. We would love to show you what kind of value we can bring and provide for you. I have three quick housekeeping items, just to make sure everybody's aware of. The first is, you should have received an email today to download your app.
Sometimes schedules, meeting places do change, and the best way to make sure that your meetings are happening as efficiently as possible is to download that app. Also, very importantly, that app is going to be one of the ways that we can communicate with you. In the back of the room, I've got [Paul DeMarco]. Paul is one of the two KeyBanc security officers that we have here. He's in charge of physical security. Many of you received yesterday an update that there's fires. The fire is about, I believe, an hour and 20 minutes away. Paul's been in contact with both the hotel as well as the local Deer Valley and Park City fire departments.
Just to put your mind at ease, in case there is anything that we have to do, Paul and the KeyBanc's physical security team will make sure that we are getting that message out to you. The message we received yesterday was a blast to everybody within basically a two-hour roundtrip circle. The third housekeeping item is, I know that it's beautiful out on the patio. The sun is shining. The vista is spectacular. But we would encourage everybody, when you have the chance, please come in to see some of the great panels that are happening here, in this room. Gartner has forecasted global IT spending up 13.5% this year, driven by AI. Gartner is also projecting that data center spending is going to grow over 55% this year.
Very few people and very few companies are in the middle of this more than Micron, which is one of the reasons why I'm so excited to invite my friend John Vinh up here to help bring Micron up to talk about some of the trends that they're seeing. With no further ado, John, Sumit.
Great. Good morning, everybody. It's my pleasure to welcome Sumit Sadana, Executive Vice President and Chief Business Officer at Micron. In this role, Sumit is responsible for Micron's business units, P&L, helping drive the company's growth strategy, customer partnerships, product roadmaps, and long-term positioning across memory and storage markets. Sumit joined Micron in 2017, brings more than three decades of experience across the semiconductor industry, having had leadership roles at SanDisk, Freescale, and IBM. As AI drives unprecedented demand for memory bandwidth capacity and storage performance, Micron is playing an increasingly important role in enabling next generation AI infrastructure. We're fortunate to have one of the industry's most respected executives with us today to share his views on the AI markets, memory, and Micron's strategic direction. Please join me in welcoming Sumit.
Thank you.
Thank you.
Thank you. Appreciate it.
Sumit, you've been a veteran in this industry. You've seen multiple cycles. Maybe just take a minute and talk to us about what's different about this cycle that we're going through versus previous cycles that you've experienced.
Sure. John, thank you for having me here, and appreciate all of you being here. I'll just start with some quick opening comments. I'll be making some forward-looking statements, so please look at our SEC disclosures for detailed risk assessments on our disclosures. Our business is on a terrific trajectory. Our business and financial performance continues to be exceptional. Since our earnings, our aggregate demand signals from our customers have increased even further. We have said that we do expect these very tight industry conditions to continue beyond calendar year 2027. Based on these increased signals of demand from our customers, we now expect that 2027 calendar year will be even tighter than 2026. Because the growth in demand that is taking shape is faster than the growth in supply for calendar 2027 on an industry basis.
We have been responding to this with significant investments across our global network of manufacturing, especially in key areas like Japan, Taiwan, and Singapore, and investments in India in back-end manufacturing and so on. Our key investments are obviously focused in the U.S. with an increase from $200 billion of investments to $250 billion over the next many years. These investments are also going to be aided by investments in our supply chain. We made this $500 million investment in GlobalWafers for raw wafers. We are also making as part of a $3 billion investment in our supply chain. With all of the efforts that we are doing, we still don't have line of sight as to when the supply is going to be able to meet demand, because demand continues to escalate at a very rapid pace over time.
With those opening comments, I'll just talk to you, John, about the specific question that you asked. Certainly, we have been through many different time horizons and industry conditions that have changed through this time period across many different growth drivers in the industry over the decades. This time feels very different for a number of different reasons. First, AI is a transformative capability that has come on the scene. Of course, AI has been researched and developed and progressed on for many years, decades. But the ChatGPT moment and the whole generative AI field has put it on steroids. I strongly believe we are headed towards artificial general intelligence and beyond that, artificial super intelligence.
That AGI moment is getting closer and closer, in part because AI has gotten so good so quickly that you're seeing these AI improving AI approaches that are accelerating the innovations that our frontier labs are able to bring to the market. This recursive self-improvement capability is super exciting and will create a lot of breakthroughs. What this means is that if you step back and look at the demand picture from a longer-term perspective, we are still in the very early stages of how this demand is going to proliferate. For example, most of the interaction today with AI still happens using chat interface. If you think about the number of users around the world that are going to grow who get access to AI, billions of people over time will be using AI who are not using it today.
If you look at companies who are trying to leverage AI, they are in the very early innings of being able to leverage AI capabilities, whether you look at Fortune 1000 companies or just the global large companies who can benefit tremendously from AI, both in terms of innovation and driving the top line as well as productivity for the bottom line. We are at such early stages, and you can see the breakthroughs happening with agentic AI. What is very important to understand about agentic AI is that the interaction of these agents with the hardware infrastructure creates and needs 5 - 30 times the tokens that a chat interface needs for a similar type of task. If it is a deep reasoning type of an agentic approach, it is even higher.
We are looking at the next few years of such dramatic growth coming from AI. When you step back and think about it, a lot of this growth is coming in the digital domain, but the supply that we have to create for it happens in the physical domain. This difference between demand growing in the digital domain and supply growing in the physical domain is a structural challenge because I do not think there is adequate appreciation for how long it takes for that supply vector to catch up. On top of that, we have this issue of HBM growing very significantly, and HBM is needed in these AI systems because in a lot of the workloads, the processor, whether it is a GPU or an ASIC or a CPU, is sitting idle for 50% of the time because it is waiting for data from the DRAM.
That is a huge underutilization of an important asset. You need to really get a much higher performance memory bandwidth and a much higher capacity of memory in the system so that AI can be deployed on an efficient and scalable basis. That means there is so much demand coming from that, and this HBM increase, consequently, is creating this 3:1 trade ratio that we have mentioned in the past, where to produce 100 bits of HBM, we have to reduce 300 bits of DDR supply because there is that 3:1 trade ratio between HBM3E and DDR. When you go to HBM4 and HBM4E, by the time you get to 4E, that trade ratio has worsened to closer to 4 : 1.
All of this growth in HBM pressures the supply that is left for everything else, which means that the wafer supply has to increase dramatically, and that is just not easy to do. It takes a long time because the whole industry got itself into a state where most of that expansion was needed to be done in greenfield expansion, which means you go to an empty space that is nothing but trees, and you have to create an entire massive fab cluster there. That is a very challenging endeavor. While our customers have to build data centers and secure power and real estate, building a data center, no offense to our customers, is considerably easier than building a sophisticated leading-edge technology front-end fab. It is one of the most complex engineering projects in the world from a construction perspective to build such fabs.
It takes a long time for these fabs to come online, and this leading-edge technology that gets deployed takes a long time to ramp. Because of all these reasons, we do not have line of sight as to when that supply vector will intersect the demand vector, which continues to escalate with every passing year as our customers have locked up all these contracts for power and real estate and data centers and so on. They are all telling us that the number one constraint they have today, and this is not just one or two customers, but a common theme across our customers, is not power or real estate or data center capacity or logic wafers. The number one constraint they have is DRAM.
Great.
We have done as a result is we have obviously created all these strategic customer agreements that have also changed how we think about this business. Memory has become a very strategic asset for all of the reasons I have mentioned. This cycle, if you call it a cycle, but this time is a very different time, and we see that we are going to. We expect to have very robust business and financial performance for a very long time.
Great. Thanks, Sumit. Lots to unpack there. I had a brief follow-up from your opening statements regarding aggregate demand has increased for you. Can you just give us some perspective and color on where is that aggregate demand increasing? I assume it's within the data center AI complex, but maybe just some additional color would be helpful.
Sure. Yeah. The aggregate demand across all of the segments is at much higher levels than it has been. Our customers are telling us that despite the fact that the prices are at very high levels, that they are eager to get more supply because they are not able to meet their own business case requirements from driving their own units and volumes and revenue and so on based on the opportunity that they see in their end markets. It's not just focused on the data center; it's across different parts of the market. As we engage with customers on these longer-term agreements and even more near-term supply, their constant feedback is that they're not happy about the extent of volumes we are making available to them.
They are signing up in these agreements, but they feel like they're leaving considerable opportunity on the table, like I said, across market segments. Of course, it's most acute in the data center, where quite often we are not able to meet any more than half of the demand our customers have. I think when we look at all of that, it's not just for 2026. These are multi-year signals that we're getting from customers.
That's great. I think when I just saw you this morning and catching up with you were telling me you've been incredibly busy meeting and talking to customers these days. I've got to imagine that the main thing that they're asking you for is: How can I get more capacity? Beyond that, can you just talk about what your customer meetings and conversations are about? I'm wondering, just with this major shift in AI demand, has your relationships with your customers changed?
Yeah. That's a great question because the nature of our engagements with customers is at levels that I have not seen in my entire career. What I mean by that is because this role of memory has become so important and strategic to our customers, it's not just about the supply-demand environment. It's also very importantly because of the phenomena I described earlier about the way that memory is determining system performance. Whether it is a data center server or it is a consumer product, or it is something that goes into an industrial system. As our customers think about how to implement AI capabilities in their products across these various segments, they're persistently hitting up against this wall, which you have heard about, this memory wall, where the bandwidth from memory is just not adequate, the capacity of memory is not adequate.
They're all focused on AI is a once in a lifetime opportunity to create differentiation for our customers in their end markets. They are very keen on my biggest bottleneck is: how do I get more out of memory from a performance perspective? How do I use memory to create differentiation in my products? What that means is we have been able to drive significant inroads with our customers on the engineering partnership front. So we have visibility now and deep engagements with our customers on the engineering side that stretches out to roadmaps at our customers that are beyond 2030 timeframe. When we started in the memory industry many years ago, it was rare to even understand what next year's products are going to be at our customers.
Now we have, I would argue, even longer-term visibility than most in the semiconductor industry have across the different semiconductor markets. We have these deep engagements, and we have become, because of our leading technology and our leading product portfolio, the company of choice for our customers to partner with in order to solve some of their most pressing challenges and in order to innovate with us, and co-engineer their products with us. We have many examples of how we have done that. If you look at the HBM3E product, we came out with capability that was 30% lower power consumption than anyone else, the next best product. That is such meaningful capability. You look at how LPDRAM, L ow Power DRAM, was deployed in the data center. Micron innovated that capability first in the industry, with NVIDIA.
We took LPDRAM, which was only meant for smartphones and laptops, into the data center for the very first time with higher density and much lower power and very robust performance. For the longest time, we were sole sourced in that with very robust ROI because it was such a differentiated product. We became multi-sourced with NVIDIA only because we were just not able to meet and keep up with the demand. This is exactly how we expect a lot of other innovative vectors to occur with our customers, where we will be first to innovate with customers, get a level of sole source, at most dual source. More and more of our portfolio is going to be ending up in that category. You are seeing our leadership in data center SSDs as another example, first to market with PCIe Gen6.
Very high levels of demand, far above our ability to supply with that product. Very high levels of demand for the industry's highest capacity SSDs with our 245 TB SSDs. There are many examples of over the last many years, how we have totally transformed the product portfolio, and that is another reason why our customers are coming to us for these deep engineering interactions and innovations. The fact that we are the only company who is going to be producing memory in the U.S. is another huge differentiator because a lot of our customers are looking for that improved resilience in the supply chain, and Micron is right there as the only company investing $250 billion in the U.S., so it is another huge differentiator.
It has totally changed the complexion of how we work with our customers, and they recognize that the next 10, 20 years, memory is at the center of AI, and that relationship has to be different as well.
That is great. Clearly, your customer relationships are clearly deepening here. That is great. Maybe along those lines, I wanted to ask you about your strategic customer agreements. Also known in the industry as long-term agreements that you have put in place. I think you have talked about putting in place SCAs that will potentially cover half of your revenues, and then you have also, as part of these agreements, have upfront capital commitments in aggregate of north of $20 billion, which is pretty impressive. I am wondering if we could just talk about how these agreements came together. Did you kind of proactively approach your customers about putting these agreements in place, or did they reach out to you, just given the shortage situation that we are having?
Yeah, great question, John. As far as these SCAs are concerned, we started working with our customers many months ago on them, and we pioneered this approach, to do these strategic customer agreements. We call them strategic customer agreements first and foremost, because they are very different than the historical LTAs or long-term agreements. First of all, those LTAs were somewhat of a misnomer because there was nothing long-term about them. They were just 12-month agreements for the next calendar year. Another important difference is that those LTAs had no binding terms in them. It was more of a handshake, kind of an understanding with customers about ensuring that there is good level of supply chain planning that we do, and documenting it so that they are putting more thought into what kind of supply they are intending to purchase from us. That was the LTA time.
When we saw some of this AI demand for a multi-year timeframe become so urgent for our customers, there was an increased level of anxiety at our customers to secure the supply. We came up with this proposal and idea of strategic customer agreements. We have pioneered this concept in our industry, and we have been the first ones to work on it with our customers. We have also, we believe, signed the most number of SCAs across our industry. Since our earnings, we have signed up more SCA agreements with our customers. We really see these agreements as being transformative of the business model that we are used to. For starters, these SCAs cover a long time horizon.
Some customers that are smaller, like automotive customers, have mostly three-year SCAs, but the SCAs that cover the overwhelming amount of the revenue under SCA is going to be five-year type of terms through the end of calendar 2030. Number one, these provide quite a long-term visibility to us. Second, these SCAs are binding commitments on purchases of these volumes by year, by customer. These are take or pay agreements and there are no contractual outs for our customers from these agreements. These are very much different terms, very stringent and binding terms on the purchases. These are backed up by tremendous amounts of upfront cash and cash-like commitments, like letters of credit. But the overwhelming amount of the commitment is upfront cash that we are going to have on our balance sheet.
At the time of the earnings, we announced 16 SCA agreements with $22 billion of cash and cash-like commitments, of which $18 billion was just cash alone that we will have on our balance sheet. Of course, as time goes by, we will have more SCAs that we sign. These SCAs also have something that our LTAs never had, which is pricing. Some SCAs are going to be floating pricing, so the pricing will be consistent with whatever the market price is at that time. But most of the volume under these SCAs is going to have a price band. There is a ceiling price. The ceiling price of the SCAs that we announced at the time of earnings, those 16 SCAs are CQ2 pricing. Since then, as I said, we have signed more SCAs.
SCAs that we sign in the future are going to have the ceiling pricing that is going to be consistent with whatever the market price is at the time the SCAs are signed in the future. That is how the ceiling price works. There will be some SCAs that will not have any pricing mechanism and will just be focused on market pricing. The floor pricing, which is a very important part for us, is set at a level that provides a gross margin for us that is well above any prior peak in the cycles of the industry. These provide very high ROI that we can leverage to make long-term investments in capital, and in our capacity. With that said, we are going to have continued focus on disciplined investments.
We build out all these clean rooms, but we are going to put equipment investments in them consistent with our view of medium-term demand from our customers. These SCAs are extremely transformative, but they also go beyond just volumes and pricing mechanisms. They also enable very deep engagements with our customers on the R&D and product roadmaps as well. Very transformative in aggregate.
Thanks, Sumit. Concern that I hear from investors is, obviously we have gone through past cycles, and I think coming out of COVID, there were a lot of LTAs that were put in place that had legally binding contractual commitments, and a lot of those were broken once supply normalizes. The question I get is, what is to prevent customers from breaking these SCAs at some point in the future once supply normalizes as well?
Yeah. There are a couple of things I can think about on that front. First, memory itself has become a very strategic asset, and our customers are viewing memory for the long term in a very different way. To them, these SCAs are not just a short-term mechanism to get commitments on supply. They know that when they look out the next 10, 15, 20 years, because of AI and the growth driven by AI, the challenges in bringing up adequate supply, but also the fact that I spoke about how memory from a system performance perspective is so critical, our customers recognize that they need to change the relationship, between the company, the customers, and a company like Micron. When there is that strategic mindset, it changes how they think about the relationship from a long-term perspective.
In addition, like I said, these LTAs that we used to have never had any binding terms. These SCAs have tremendous number of binding terms, and there are no contractual way to get out of these SCAs. On top of that, these SCAs have an evergreen structure to them, which means over time, more years can be added to the back end on these SCAs to keep extending them. The reason our customers have gone to that kind of a structure with us is they don't expect that this is just a five-year kind of a thing. They expect that over time, this will become a good construct to engage with us.
Because of how AI is coming in right now with huge amount of growth driven from the data center, but over time, this growth is going to go into a lot of other areas of the market in a very meaningful way, proliferate from the data center to the edge with consumer devices, smartphones, autonomous driving, and then huge amount of growth coming from robotics, which is going to be a ginormous growth driver. Our customers recognize and understand that these spurts of growth are going to happen many times, much faster than the supply can grow. This is not the last time you're going to see a significant deficit of supply in the industry.
My view is that, if some customers are wanting to be adventurous on the terms of these SCAs and complying with the terms of these SCAs, what happens when there is the next shortage? They may not be able to get much allocation at that time. All right? There have been customers who have treated memory very tactically in the past, and those customers are struggling even more right now to get allocation, in a tight environment. I do think that customers recognize the strategic nature of the engagement and intend to think of this as a long-term partnership.
Great. That makes a lot of sense. Sumit, you had referenced this a few times this morning. What are your thoughts on physical AI? It's emerging category. It's getting a lot of attention. You're seeing a lot of semiconductor companies start to invest in it. Where do you think the opportunity is in physical AI?
I think, physical AI is in its infancy right now, and it has a tremendous and hugely exciting opportunity ahead. A lot of what has happened in AI has happened in the digital domain. Our customers recognize that the next massive growth vector that could potentially even dwarf the growth that is occurring today, could happen in the physical domain when you think about things like robotics and so on, could become massive growth vectors in the future. If you think about how our customers are approaching this, they have always had a view of robotics becoming, for example, a big growth driver. There is a lack of adequate data to train these robots from the physical world. There is a lot of focus on accelerating the training.
These robots and humanoids are going to have an exponential learning curve based on their ability to cumulate the learnings from each individual unit of robot can be taught something different and ultimately, aggregated learning from all of those can multiply the capability of each individual robot. There is tremendous excitement on that exponential curve, and these exponential curves grow faster than we can grasp and analyze and project. We very much expect that the physical AI domain is in its infancy now. It is going to grow rapidly over the next few years, and when it hits its inflection point, it is going to be another vector of growth that we are quite likely going to struggle to supply for years to come.
I thought it was interesting you described it as potentially dwarfing the opportunity that is in front of you today, which is health size. Why do you feel that this could dwarf that opportunity?
I think if you think about, for example, just humanoids alone, there is massive opportunity there and each single humanoid robot is expected to have hundreds of gigabytes of DRAM and terabytes of SSD. That is a massive amount of DRAM. The reason you need that DRAM is you need these humanoid robots to have very fast response times and be able to have a level of functionality even if there is not an access to the cloud and back to the data center. The amount of onboard compute capability that is needed for functionality, safety, security, all kinds of issues, is going to be driving that level of capacity and driving a level of performance requirements that will enable us to come up with really innovative ideas and solutions for our customers.
That aggregate amount of capacity multiplied by the extent of deployment that these robots are going to have over the next decade and beyond, is going to just be astronomical. We feel like, again, robotics in its infancy now, but will start to grow rapidly later this decade and get into that exponential curve early part of the next decade.
Great. I think one thing that you had mentioned on your earnings call is you had said the outlook for the server market is actually increased because you are starting to see some level of despec-ing by your customers there because of the increasing prices of memory. I think we are seeing RDIMM densities decrease. Can you maybe just talk to that? Is that potentially something that could be harmful to the industry if you start to see customers start to despec as a trend?
Yeah. In our interactions with our customers, it is not really driven by pricing on the server side as much as it is driven by just lack of adequate supply. Really, our customers are struggling to get their hands on adequate DRAM supply because the extent of supply that they feel is available to them is not going to enable them to ship the units that they need to ship to meet the opportunity that is ahead of them. Consequently, they are focused on balancing how much capacity of DRAM to put in the system with the number of units that they want to ship. That is where the modulation of the average capacity in the system is happening, whether it is in servers or some consumer products. It is very heavily driven by just availability of supply.
Despite all of that, when you multiply the units with the average capacity that our customers are planning, the aggregate demand coming from even those new levels of average capacities that our customers are planning is rising at a pace that is just I made my earlier comment about aggregate demand increasing despite all of that change that our customers are planning in average capacity. The aggregate demand is increasing despite that. I will also mention that there is this phenomenon that I spoke about earlier about system performance getting impacted if there is not adequate memory capacity. That kind of changes in memory capacity in the system is going to further reduce the potential utilization that the processor was going to have. That means that there is tremendous opportunity to improve the system performance simply by increasing the amount of memory in the system.
Once the system is configured and qualified with a certain amount of capacity, increasing the capacity of memory in the system is an easier lift for our customers, and they would be able to introduce a higher performing SKU with a higher level of capacity when they see the additional supply becoming available. There is all of this latent demand out there that comes from being able to increase the average capacities and create SKUs that are higher performing and are able to have longer context windows, much better capability from an overall system performance perspective when that additional supply is available. But again, it is not clear on a multi-year timeframe when that additional supply will be available. This is the optimizations that our customers are trying to do to try and continue to maximize the units they ship.
Great. Maybe related to that, I think something that you have recently disclosed is that you are starting to moderate your price increases, to some extent. Historically, as you know, in memory, when you start to see price increases start to moderate, it typically signifies that you are getting closer to the end of the cycle. But you just recently just said this morning that your aggregate demand in 2027 is actually increasing. That does not really make sense to me. I am wondering if you could reconcile the price actions versus demand getting even stronger for you.
Yeah. I think if you look at our financial performance, it is at extraordinarily robust levels, right? You have seen our gross margin performance and the operating margins that we are delivering. 81% operating margin in the last quarter is very robust level of margin. Of course, there is pricing opportunity in the future. We are obviously going to try and optimize what the right level of pricing ought to be based on enabling our customers' long-term demand, as well as ensuring a good financial performance and ROI for us. There are lots of opportunities to continue to grow our revenue and profits over time. And we do expect this constrained environment to last for a very long time.
Between the improvements that we will have in our portfolio mix, the improvements that we will have in our overall supply in terms of shipments, and the opportunities that we have on the pricing front, we will optimize all of those parameters to ensure a very balanced outcome that is in the best long-term interest of our customers and ourselves.
Great. In a few minutes here, we are going to open it up for questions. My last question for you, Sumit, is if you think about Micron over the next five years, what is the one thing you would want investors to know about Micron?
Yeah. I think Micron today is a very different company and has a tremendously stronger business model looking out into the future than it has ever had. We are executing at the best levels that we have executed as a company across technology, products and portfolio, manufacturing and operations. Just a tremendous momentum. AI is a total game changer for memory and has made memory a strategic asset. We believe that the next five years or next 10 years, we have tremendous tailwinds, a completely changed and robust business model, high levels of ROI expectations, and these SCAs that have been transformational, and we expect to be transformational for our business.
Great. I think there are microphones around the room. If you have a question, feel free to raise your hand. Great.
Hi. Thank you. This has been really helpful and insightful. You noted one of Micron's key differentiators as being an American producer. How does Micron view the expansion of Samsung's fabs in Texas and SK hynix entering the Indianapolis, Indiana area, as a differentiator going forward?
Our understanding has been that Samsung's investments are in logic foundry, not in memory. Hynix's investments are in back-end manufacturing, assembly, packaging type of investments. Micron is the only company investing in front-end fab manufacturing in the U.S. We have Idaho 1, Idaho 2, leading-edge memory fabs coming online middle of next year. First fab, second fab, end of 2028. Then we have New York 1, and a cluster of New York fabs beyond that. We have our investments in Virginia for long life cycle technology and bringing 1-alpha DRAM into Virginia. Then, of course, investments in U.S. supply chain as well.
When you think about all of these investments that we are making, the investments we are making in our communities, the investments in talent, and all of the investments in technology, here in the U.S., we have a very unique position compared to everyone else. Our customers recognize that, they value that. We do believe that we will be able to get a premium for our U.S. supply, from a pricing perspective. That premium is something that we have even baked into our SCAs. So we feel very, very good about our U.S. investments, our positioning, as well as the differentiation that it's going to bring to our customers, the resilience it's going to bring to our customers on supply chains, and how it'll enable our customers to meet their own goals about how to leverage more U.S.-made content.
Great.
I just wanted to ask about the custom HBM opportunity, when it can be material for Micron, and what it could mean for your market share.
When we talk about HBM, as you know, we have a hugely differentiated HBM3E product that continues to be in high volume production. HBM4, very robust capability, which we announced some milestones at the time of our last earnings. HBM4E is when there is going to be a custom SKU of HBM as we work with customers on, and there is plenty of opportunity for differentiation in HBM. The one thing that may not be very well understood on HBM is, even beyond the custom opportunities that start with HBM4E and continue with future generations of HBM, there is also this phenomena where the whole core design and qualification of HBM is such a time-consuming, R&D-intensive, and expensive process that it is not practical for our customers to be able to do business with all three HBM suppliers for all projects.
Consequently, many projects and platforms and several customers are going to just end up using one or two HBM suppliers, and a very small number of customers will end up using all three HBM suppliers. HBM, whether it is on a platform basis or on a customer basis, we see is going to become a two-supplier market, and oftentimes, just a single-sourced opportunity. It is not just true for HBM. Like I said earlier, since memory is such a critical asset now, there is this significant push to create differentiation using memory across a range of opportunities. I gave the LPDDR6 example for data center, but there is now a proliferation of opportunities to create differentiation in other end markets as well.
We see those opportunities again as being either single-sourced or dual-sourced for a long period of time because it is not easy for our customers to do that kind of co-engineering work with three different companies. They typically will do it with one, then maybe bring in another. We think that the market is going to become very much like an ASIC-like market more and more, and we see that more and more of our portfolio over time is going to migrate to those sort of capabilities, which is incredibly powerful from an ROI perspective long term.
Great. I think we have time for one more.
Congratulations, first of all, on both not only the financial performance of Micron, but also the technological advances that you have made, which are pretty impressive.
Thank you.
The gap between DRAM on a per storage unit and flash has grown dramatically over the last couple of years, yet there's still a lot of cold data that ends up in DRAM. Are you seeing any threats in terms of people finding ways to dynamically offload their cold data off of Micron's DRAM over to flash, despite its speed disadvantage, and then port it back over to DRAM and therefore reducing the DRAM requirements and needs? Otherwise, do you see anything else that's happening on the flash side that you might be concerned about?
Yeah, we are doing a lot of work with some of the most leading-edge customers and the frontier labs and so on architectures for processor and memory and storage, and the entire hierarchy of how data is stored and moved across the hierarchy. What we are seeing is that the way AI works requires so much DRAM for a more balanced performance, that our customers just don't have adequate capacity of DRAM in the system. When they don't have that adequate capacity of DRAM, then things like KV cache spill on to the flash side of things and go into NAND, and you start to use more of the NAND for those purposes. AI systems work well when there is ready access to high-performance bandwidth between processor and memory.
The one gate is the performance and bandwidth from processor to memory, and that's why you see all these HBM-like capabilities and architectures. But then the next gate is just the raw capacity of DRAM in the system. Our customers find that as they make AI systems do more reasoning and do more intelligent work and get into agentic modes, the context windows lengthen, and that requires more DRAM, and it ends up spilling over into NAND flash. We continue to see an environment where both the performance of DRAM leading to more innovations, and the capacity of DRAM is heavily stressed in these next generation AI systems.
Great. Thanks for your time this morning, Sumit.
Thank you.
Appreciate it.
Thank you very much.