Micron Technology, Inc. (MU) Earnings Call Transcript & Summary

[Audio Gap] Freescale scale IBM. So 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.

So 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 kind of previous cycles that you've experienced?

Sure. So John, thank you for having me here, and I 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 of our business and financial performance continues to be exceptional. And since our earnings or aggregate demand signals from our customers have increased even further. Now, 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 and Taiwan and Singapore and investments in India and back-end manufacturing and so on. And 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. And these investments are also going to be aided by investments in our supply chain and we made this $500 million investment in global wafers for raw wafers. We're 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. So 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, 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 Chat GPT moment and the whole generative AI field has put it on steroids. And I strongly believe we are headed towards artificial general intelligence, and beyond that artificial super intelligence. And 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. So 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. And 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, 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. And we are such early stages and you can see the breakthroughs happening with Agentic AI. Now what is very important to understand about Agentic AI is that the interact -- the interaction of these agents with the hardware infrastructure creates and needs 5x to 30x the tokens that a chat interface needs for a similar type of task. And if it is a deep reasoning type of an agentic approach, it is even higher. So we are looking at the next few years of such dramatic growth coming from AI. And 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. And this difference between demand growing in the digital domain and supply growing in the physical domain is a structural challenge because I don't 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's a GPU or an ASC or a CPU is sitting idle for 50% of the time because it is waiting for data from the DRAM and that is a huge underutilization of an important asset. And so you need to really get much higher performance memory bandwidth and much higher capacity of memory in the system so that AI can be deployed on an efficient and scalable basis. And that means there is so much demand coming from that and this HPM increase consequently is creating this 3:1 trade ratio that we have mentioned in the past, where to produce 100 bps of HBM, we have to reduce 300 bits of DDR or supply because there is that 3:1 trade ratio between HBM 3E and DDR. And when you go to HBM4 and HBM4E by the time you get to that trade ratio has worsened to closer to 4:1. So 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's 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 where there's nothing but trees and you have to create an entire massive fat cluster there. So that's a very challenging endeavor. And while our customers have to build data centers and secure power in real estate, building a data center offense to our customers is considerably easier than building a sophisticated leading-edge technology, front-end fab. It's one of the most complex engineering projects in the world from a construction perspective to build such fabs. And so 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. And so because of all these reasons, we don't 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. And they're all telling us that the #1 constraint they have today -- and this is not just 1 or 2 customers, but a common theme across our customers is not power or real estate or data center capacity or logic wafers, the #1 constraint they have is DRAM. So what 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. And memory has become a very strategic asset for all of the reasons I have mentioned. So this cycle, if you call it a cycle, but this time is a very, 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. A lot 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 more additional color would be helpful.

Sure. Yes. I mean 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're 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. And 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 are 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 and so I think when we look at all of that, it's not just for 2026. I mean, these are like multiyear signals that we're getting from customers.

That's great. I think when I just saw you this morning and catching up with you, you're 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. But beyond that, can you just talk about what your customer meetings and conversations are about. And I'm wondering just with this major shift in AI demand as your relationships with your customers changed?

Yes. I mean, 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. And what I mean by that is because this role of memory has become so important and strategic to our customers. And it's not just about the supply-demand environment. It's -- also very importantly, because of the phenomenon 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. So they are all focused on AI is once in a lifetime opportunity to create differentiation for our customers in their end markets. And 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? And 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 road maps at our customers that are beyond 2030 time frame. I mean, 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. And now we have -- I would argue, even longer-term visibility than most in the semiconductor industry have across the different semiconductor markets. And 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 and we have many examples of how we have done that. I mean if you look at the HBM3E product, we came out with capability that was 30% lower power consumption that anyone else, the next best product, and that is such a meaningful capability -- you look at how LP DRAM, low-power DRAM was deployed in the data center, Micron innovated that capability, first in the industry with NVIDIA. We took LP DRAM, 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 and for the longest time, we were sole sourced in that with very robust ROI because it was such a differentiated product, and we became multisourced with NVIDIA only because we were just not able to meet and keep up with the demand. And 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, atmos dual-source and more and more of our portfolio is going to be ending up in that category. I mean you're seeing our leadership in data center SSDs as another example. First to market with Gen 6, very, 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 terabyte SSDs. There are many, many examples of over the last many years, how we have totally transformed the product portfolio and that's another reason why our customers are coming to us for these deep engineering interactions and innovations. And the fact that we are the only company who's 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's and other huge differentiator. So it's totally changed the complexion of how we work with our customers, and they recognize that the next 10, 20 years, memories at the center of AI and that relationship has to be different as well.

That's great. Clearly, your customer relationships are clearly deepening here. That's 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've put in place. I think you've talked about putting in place SEAs that will potentially cover half of your revenues and then you've also, as part of these agreements have upfront capital commitments in aggregate of north of $20 billion, which is pretty impressive. I'm wondering if you 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're having?

Yes, great question, John. So 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 the strategic customer agreements. And we call them strategic customer agreements, first and foremost because they are very, 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, right? 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. So when we saw some of this AI demand for a multiyear time frame become so urgent for our customers -- there was an increased level of anxiety at our customers to secure the supply, and we came up with this proposal and idea of strategic customer agreements, and we have pioneered this concept in our industry, and we have been the first ones to work on it with our customers, and we have also, we believe, signed the most number of SCA'a across our industry. Since our earnings, we have signed up more SCA agreements with our customers. And we really see these agreements as being transformative of the business model that we are used to. For starters, these SCA's cover a long time horizon. Some customers that are smaller, like automotive customers have mostly 3-year SCA's but the SCA's that covered the overwhelming amount of the revenue under SCA is going to be 5-year type of terms through the end of calendar 2030. And so 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. And these are take-or-pay agreements, and there are no contractual outs for our customers from these agreements. These are very much very, very 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. So 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. And of course, as time goes by, we will have more SCAs that we sign. Then 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 is going to be under these SCAs is going to have a price band. And so there is a ceiling price the ceiling price for the SCA that we announced at the time of earnings or 16 SCAs, our CQ2 pricing. Since then, as I said, we have signed more SCAs. SCAs that we signed in the future are going to have pricing at 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. So that's 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. And so 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. So we'll build out all these clean rooms, but we are going to put equipment investments in them consistent with our new -- our view of medium-term demand from our customers. So 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 road maps as well. So very, very transformative in aggregate.

Thanks, Sumit. Concern that I hear from investors is, obviously, we've 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 when supply normalizes. The question I get is what's to prevent customers from breaking these SCAs at some point in the future once supply normalizes as well.

Yes. There are a couple of things I can think about on that front. So first, memory itself, has become a very strategic asset, and our customers are viewing memory for the long term in a very different way. So 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 in a system 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. And so 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. And the reason our customers have gone to that kind of a structure with us is they don't expect that this is just a 5-year kind of a thing. They expect that over time, this will become a good construct to engage with us. And because of how AI is coming in right now with a 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 a 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 so this is not the last time you're going to see a significant deficit of supply in the industry. So 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. right? And so 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. So 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, right? 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. And 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. And 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, but there is a lack of adequate data to train these robots on from the physical world. So there's a lot of focus on accelerating the training and these robots and human nits are going to have an exponential learning curve based on their ability to accumulate 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 robots. So there is tremendous excitement on that exponential curve, and these exponential curves grow faster than we can grasp and analyze and project. And so we very much expect that the physical AI domain is going to -- in its infancy now is going to grow rapidly over the next few years and when it hits its inflection point, it's 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's in front of you today, which is outsized. Why do you feel that this could dwarf that opportunity?

I think if you think about, for example, just humanoid alone, there is massive opportunity there. And each single humanoid robot is expected to have hundreds of gigabytes of DRAM and terabytes of SSD. And that is a massive amount of DRAM. And the reason you need that DRAM is you need these human eyed robots to have very fast response times and be able to have a level of functionality even if there isn't an access to the cloud and back to the data center. And so 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. And 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. So we feel like again, robotics and its infancy now, but will start to grow rapidly later this decade and get into that exponential curve early part of the next decade.