Micron Technology Fiscal 2026 Third-Quarter Post-Earnings Analyst Call Notes
目录
TL;DR
I. Opening Remarks and Safe Harbor
II. Capital Return and DRAM Product Mix
III. HBM Orders, Long-Term Demand, and Supply Constraints
IV. Strategic Customer Agreement Structure and Cash Deposits
V. Data Center Non-HBM DRAM, LPDDR, and Greenfield Costs
VI. Signed Customer Agreements, Deposit Returns, and Customer Supply Shortfalls
VII. NAND, Enterprise SSDs, and China Competition
本内容基于公开资料和研报数据整理，不构成任何投资建议，不代表任何个人观点，仅供学习参考，请理性阅读
The core tension in Micron’s post-earnings call is that AI-driven memory demand is far outstripping supply, strategic customer agreements are improving cash-flow visibility, but capacity expansion, rising costs, and Chinese competition still need to be tested. The core tension in Micron’s post-earnings call is that AI-driven memory demand is far outstripping supply, strategic customer agreements are improving cash-flow visibility, but capacity expansion and costs.
TL;DR
The core tension in Micron’s post-earnings call is that AI-driven memory demand is far outstripping supply, strategic customer agreements are improving cash-flow visibility, but capacity expansion, rising costs, and Chinese competition still need to be tested.
Management emphasized that AI data centers, edge computing, and demand for higher-performance memory are driving company cash flow to record highs; fourth-quarter free cash flow is expected to continue growing, and the company will retain sufficient cash to navigate the cycle.
Strategic customer agreements are the most important incremental development in this call. Micron has signed 16 agreements, representing more than $22 billion in cash and related financial commitments, including nearly $17 billion in cash; most agreements have five-year terms.
HBM demand remains materially above supply. Management said customer demand for HBM3E, HBM4, and even HBM4E not only exceeds Micron’s available supply in 2027, but extends into.
Non-HBM DRAM and NAND are similarly tight. Management is unwilling to provide longer-term bit demand forecasts because future shipment growth is no longer primarily determined by demand, but by supply; in DRA.
Capacity expansion will create cost pressure. Higher HBM stack ratios, LPDDR penetration in data centers, and greenfield fab expansion will all increase cost per bit; start-up costs will appear in the fourth fiscal.
Capital expenditure will continue to move higher. Fiscal 2026 capex is around $27 billion; fiscal 2027 will increase significantly, with more than half of the incremental spend allocated to construction; management implied that it will.
I. Opening Remarks and Safe Harbor
Operator: Ladies and gentlemen, thank you for joining Micron’s post-earnings analyst call. After management’s remarks, we will conduct a Q&A; session. I will now turn the call over to Sacha Kumar, Vice President of Investor Relations and Finance.
Sacha: Thank you, and welcome to Micron Technology’s fiscal 2026 third-quarter post-earnings analyst call. Joining me today are Micron’s Chief Business Officer Sumit Sadana, Executive Vice President of Global Operations Manish Bhatia, and Chief Financial Officer Mark Murphy.
Please note that today’s discussion includes forward-looking statements regarding market demand and supply, market trends and drivers, expected results, guidance, and other matters. These forward-looking statements involve risks and uncertainties that could cause actual results to differ materially from today’s statements. Please refer to our filings, including our most recent Form 10-K and our upcoming Form 10-Q, for related risks. Although we believe the expectations reflected in the forward-looking statements are reasonable, we cannot guarantee that future levels of activity or performance will be achieved.
We undertake no obligation to update any forward-looking statements to conform them to actual results. We can now begin the Q&A.;
II. Capital Return and DRAM Product Mix
Operator: We will now begin the Q&A.; To ask a question, please press star 1 to raise your hand; to withdraw your question, press star 1 again. The first question comes from Ben Reitzes of Melius Research.
Ben: Mark, great to speak with you. Looking at next quarter’s numbers, free cash flow will be above $30 billion. I want to confirm whether the buy side and investors are correctly understanding what you said about cash returns. You said 100% would be returned to shareholders, and I understand the vast majority would be buybacks.
If free cash flow starts at $30 billion and continues to grow, and assuming next year is close to this quarter’s level, you could repurchase about 10% of the company next year. Based on a $1.2 trillion market cap, that is roughly 10%. Are you prepared to repurchase at that level?
Mark: Ben, thank you for the question. We are very pleased with the company’s financial trajectory. Memory is highly important across many markets, including AI data centers, the edge, and the capabilities required for this technology revolution. Through our technology products and manufacturing performance, we are generating record cash flow. The cash flow generated over the past two quarters is already close to the total produced across many prior periods in the company’s history.
As you pointed out, we expect cash flow to increase again in the fourth quarter. Over the past year, we have paid down a meaningful amount of debt, and cash will continue to accumulate. We will maintain a level of cash we consider comfortable, allowing the company to continue investing through all phases of the business.
But as you heard today, we have confidence in the durability of business performance. The reasons include long-term growth demand, demand for higher-performance memory, the structural supply challenges we have discussed over the past several years, slower node migrations or lower benefits from node migrations, HBM consuming more wafers, and incremental wafer capacity requiring greenfield construction. We also announced a meaningful number of strategic customer agreements today.
We will hold an appropriate amount of excess cash. We have consistently said we plan to increase the dividend over time. Recently, you also saw us raise the dividend by 30%. But the primary form of capital return will be share repurchases. As I said in my prepared remarks, we plan to increase capital returns starting December 9, which is the two-year anniversary of signing our CHIPS agreement. After that, the cadence and pace of returns will depend on multiple factors, but we are absolutely committed to capital returns.
Ben: Can I ask a follow-up? On strategic customer agreements, you mentioned 40%, and ultimately about half of the business would be under these agreements. I cover Apple, and they have never publicly indicated a willingness to pay full price for a component and pass it on to customers. To me, this seems like a signal that they are willing to do business for full-price DRAM. Would you be more inclined to increase DRAM? Under these conditions, could the DRAM mix be higher than expected? That could make the share of some strategic customer agreements slightly lower, but for good reasons.
Sumit: When you say a higher DRAM mix, relative to what?
Ben: Relative to HBM, NAND, or whatever products you are making capacity allocation decisions around. Consumer-facing commodity DRAM, in my view, will not be included in strategic customer agreements in the same volume as hyperscale cloud customers.
Sumit: A few thoughts. As you know, the mix between DRAM and NAND in our business, with DRAM including HBM, typically fluctuates from about 80% DRAM and 20% NAND to around 75% DRAM. We are relatively comfortable with that mix and will continue serving customers in both product categories.
On HBM, we have made and continued to communicate a strategic decision: our goal is for our HBM share over time to be consistent with our DRAM share. So our intent is to support both customers’ HBM demand and the non-HBM portion of the data business, across all market segments.
We do believe in the power of diversification. If you look at EPBU and NCBU, the two non-data-center business units, together they are close to 40% of company revenue. We like that diversification and will continue ensuring that we serve customers and their demand across markets, supporting their growth across all market segments, including non-HBM DRAM, HBM, and NAND.
Ben: Understood. Thank you very much. Well done.
III. HBM Orders, Long-Term Demand, and Supply Constraints
Operator: The next question comes from Harlan Sur of J.P. Morgan.
Harlan: Thank you for hosting this callback. Last September, the Micron team said HBM3 and HBM3E were fully booked through calendar 2026. At that time, HBM4 was still in qualification and evaluation. Now, particularly with stronger XPU-related demand, have you already locked in volumes and pricing for HBM3E and HBM4 in calendar 2027? I understand HBM4E still needs to go through qualification, but are current-generation HBM3E and HBM4 already sold out for 2027?
Sumit: The HBM demand we are seeing, including HBM3E, HBM4, and of course early demand for HBM4E, involves very large customer requirements for 2027 shipments. Because of these strategic customer agreements, we have been discussing multi-year demand and customer requests. One major advantage of these agreements is that we can discuss demand on a multi-year basis.
If you look at a multi-year timeframe spanning 2027, moving into 2028 and beyond, we have received very high-confidence demand from customers, and that demand is far above the level we can support with existing supply. So demand remains clearly above supply. Even with the multi-year strategic customer agreements we have signed, the volumes included in the agreements are below what customers actually wanted to contract for. In fact, in many negotiations, we spent substantial time helping customers understand that this is all we can deliver within this timeframe.
So it is not just 2027. Even into 2028, demand across various HBM products is materially above our supply capability. By the way, the same is true for non-HBM DRAM.
Mark: Harlan, I would add one point. On HBM, today we said we expect market tightness to continue beyond 2027. Part of the reason is that we are seeing the HBM ramp move higher. Previously, we said the HBM total addressable market would exceed $100 billion in 2028. Now we see the HBM market size easily exceeding $100 billion in 2027.
Harlan: That is very helpful, thank you. Another question: you have not updated your medium- to long-term industry DRAM and NAND bit demand view for more than a year. A lot has changed over the past 12 months. Inference workloads, training workloads, and inference workloads themselves are evolving and becoming more complex. On the server CPU side, CPU customers are also forecasting 30% to 40% annual growth, driven by higher CPU content and memory sensitivity. I believe medium- to long-term bit demand figures would also be part of your multi-year strategic customer agreement discussions. Can you update your medium-term view on DRAM and NAND bit demand over the next few years?
Sumit: That is a good question. We previously provided some updates on changes in 2026 bit demand forecasts. The DRAM forecast was revised slightly higher, while NAND was relatively similar.
We have not provided much longer-term forecasting because, for the foreseeable future, bit demand and shipment growth are no longer truly determined by demand, but more by supply. Demand is already far above the industry’s supply capability, so supply growth will effectively determine shipment growth, rather than demand growth.
Therefore, we will provide some data points on how we see supply growth and demand growth this year or next year, but we will not provide a much longer-term outlook. We continue to evaluate how supply conditions are changing. Our expectation is that supply will remain below the level required to meet demand. We do not see when supply can catch up with demand, and we will not make a forecast on that at this point.
Harlan: Understood. Thank you.
IV. Strategic Customer Agreement Structure and Cash Deposits
Moderator: The next question comes from Tom O’Malley of Barclays.
Tom: I want to go back to the long-term agreements you have signed. What happens if a customer cancels an agreement? What financial constraints do you have? Can you retain all of the cash under the agreements? Any color would be helpful.
Sumit: First, these strategic customer agreements are non-cancelable. There is no clause in the agreements that allows customers to exit. These agreements are structured as take-or-pay arrangements. Outside of automotive, they are typically five-year agreements. Each year has an annual purchase-volume commitment. Take-or-pay means that, whether or not the customer wants to buy those bits, it is obligated to pay price times volume.
In many of the larger agreements, the pricing itself has a price corridor, with a ceiling and a floor. Pricing is negotiated quarterly based on market conditions, but in any case it cannot exceed the ceiling or fall below the floor. As a result, the value of these agreements can be determined relatively easily.
The agreements also include premiums for newer products that are more complex, higher performance, and higher capacity. For example, LPDDR6 versus LPDDR5, DDR6 versus DDR5, new versions of HBM, and so on all have provisions that allow these products to be priced at a premium to existing products. All of these provisions are included. What is not included is any ability for the customer to exit the agreement.
Beyond the take-or-pay obligations, these agreements also have financial obligations created by price times volume, as well as upfront cash deposits. Customers provide support in the form of upfront cash deposits and related financial commitments, such as letters of credit, representing a small portion of the total. Across just the 16 agreements already signed, aggregate cash and related financial commitments exceed $22 billion, of which nearly $17 billion is cash.
As you can imagine, when we reach our target number of agreements, moving from roughly 20% of demand coverage today, about one-third of bits, to ultimately covering about half of company revenue, or even slightly more, the cash associated with these agreements will increase significantly from the current $22 billion. This cash represents customers’ commitment to this new business model. That is how these agreements are structured.
Tom: Thank you, that is very helpful. A follow-up on the cash deposits. My understanding is that, at the end of the agreements, you ultimately return this cash to customers. Generally speaking, what is the strategic rationale for collecting this cash upfront and being able to use it freely? How is the cash arranged? Why is it returned to customers at the end rather than recognized as revenue as part of sales under the agreements?
Sumit: Customers will perform under the terms of the agreements and purchase the corresponding volumes over multiple years. The cash is a contingent arrangement, and it is also a demonstration of customers’ goodwill and confidence in the new business model. It means that, in the unlikely event a customer cannot or does not purchase the corresponding volumes at the prices specified under the agreement, we have the right to deduct from the cash balance. It ultimately gets returned to them as one available remedy, but it is not the only remedy.
These arrangements do not relieve customers of their responsibility to purchase the agreed volumes at the agreed prices over the term of the agreements. The cash is only one component of the overall transaction. The cash is not returned all at once at the end of the agreement. It is returned over a period of time, with the return schedule weighted toward the back half of the agreement term.
Tom: Thank you.
V. Data Center Non-HBM DRAM, LPDDR, and Greenfield Costs
Moderator: The next question comes from Melissa Weathers of Deutsche Bank.
Melissa: Thank you. On non-HBM DRAM in the data center, you have talked a lot about SoCAMM and the use of low-power DRAM in data center applications. Particularly as agentic AI drives a higher server CPU mix, can you update us on SoCAMM attach demand growth and adoption trends?
Sumit: Agentic AI will drive a lot of GPU demand and GPU server growth, which is the trend we are seeing. These CPU-based servers come from multiple different suppliers. Many companies are introducing CPU products for the data center, including x86 CPUs, NVIDIA CPUs, and CPUs announced by other companies. Over time, many different types of CPUs will be used in the data center to drive AGI and AI usage.
We have CPUs using DDR5, and we also have customers planning to increase the use of LPDDR in the data center. When LPDDR is used, it will be in the SoCAMM form factor. As you know, Micron has been an industry pioneer in driving LPDDR adoption. We were the first company in the industry to launch the relevant products, and for a long time we were the sole source of data center LPDDR. We were also among the first to bring the SoCAMM form factor to market.
We continue to expect this to be an area of differentiation for us relative to customers. We are recognized as a leader in this area and have market leadership in these products. We have very strong collaboration with customers that plan to use LPDDR to reduce power consumption, improve performance, and even reduce memory footprint. SoCAMM can help achieve these goals.
