Dolphin Research
2025.09.24 01:50

Micron (Minutes): Broad Guidance Lifts, Is AI Finally Boosting Traditional Semiconductors?

The following are the minutes of Micron MU's FY2025 Q4 earnings call organized by Dolphin Research. For earnings interpretation, please refer to "Micron: Is the AI Capex frenzy finally kicking off a memory supercycle?"

I.$Micron Tech(MU.US) Review of Core Financial Information

1. Q1 Guidance non-GAAP:

a. Revenue: Expected to be a record $12.5 billion (±$300 million).

b. Gross Margin: Expected to be 51.5% (±100 basis points).

c. Operating Expenses: Expected to be approximately $1.34 billion (±$20 million), with growth primarily driven by data center product R&D investment.

d. Earnings Per Share (EPS): Expected to be a record $3.75 (±$0.15).

e. Net Capital Expenditure (CapEx): Q1 is expected to be approximately $4.5 billion, which can be considered a reasonable benchmark for each quarter of FY2026.

f. Tax Rate: Expected to be approximately 16.5% for Q1 and the full year of 2026.

Special Note: FY2026 will be a 53-week fiscal year (one week longer than FY2025), which will impact Q4 operating expenses. The guidance does not include any potential new tariff impacts.

II. Detailed Content of Micron's Earnings Call

2.1 Key Information from Executive Statements

1. Performance:

a. Q4 Performance: Revenue, gross margin, and EPS all exceeded the upper limit of company guidance.

b. FY2025 Full-Year Performance: Revenue reached a record $37.4 billion, up nearly 50% year-over-year. Gross margin expanded to 41% (an increase of 17 percentage points).

c. Growth Drivers: Primarily driven by high-value data center products (HBM, high-capacity DIMMs, etc.), DRAM pricing advantages, and data center SSD business.

d. Q1 Guidance: Revenue and EPS are expected to reach new highs in the next quarter (FY2026 Q1).

2. Technological Progress:

a. DRAM Technology:

- 1-gamma node: Achieved mature yield, 50% faster than the previous generation, and has begun shipping to major customers for servers. 1-gamma will be the main driver of future supply growth.

- 1-beta node: Capacity will shift to support HBM growth in 2026.

b. HBM: Q4 revenue was nearly $2 billion (annualized nearly $8 billion). Market share is expected to be in line with the company's overall DRAM share (20-25%) in calendar Q3.

- HBM4/4E: HBM4 samples have been provided to customers, leading in performance (bandwidth >2.8TB/s) and energy efficiency. Will collaborate with TSMC to produce the base logic chip for HBM4E and offer customization options for higher margins.

- The number of customers has increased to six, and most of the HBM3E supply for 2026 has signed pricing agreements, with active negotiations for HBM4 orders.

c. NAND Technology: G9 NAND production ramp is smooth, and the industry's first G9 data center products, including the first PCIe Gen 6 SSD, have been launched.

d. Server Memory: As the exclusive supplier of LPDDR5 for NVIDIA's GB product family, this business grew over 50% quarter-over-quarter, reaching record revenue.

e. Product Portfolio Adjustment: Development of managed NAND has been halted, focusing resources on high-return areas such as HBM and data center SSDs.

3. Market Outlook and Demand Analysis:

a. Data Center Market (Core Growth Engine): 2025 server shipment expectations have been raised to approximately 10% growth (previously mid-single digits). Both traditional and AI servers are showing strong growth. Data center business now accounts for 56% of total company revenue, with a gross margin of 52%.

b. PC Market: 2025 PC shipment expectations have been raised to mid-single-digit growth (previously low single digits), benefiting from the Win10 upgrade cycle and AI PC proliferation.

c. Smartphone Market: Shipment expectations remain at low single-digit growth. AI phones drive DRAM capacity demand, with flagship models of 12GB and above continuing to increase in share.

d. Automotive, Industrial, and Embedded Markets: Demand is strengthening, with ADAS and AI cockpit experiences driving memory and storage demand growth.

4. Industry Supply and Demand Outlook:

a. DRAM:

- Demand: 2025 industry bit demand growth expectations have been raised to "high teens" percentage (16-19%).

- Supply: Supply is tight. Low inventory, limited technology node migration, increased plant costs, and cycles will constrain supply growth in 2026. It is expected to "further tighten" in 2026.

b. NAND:

- Demand: 2025 industry bit demand growth expectations have been raised to "low-mid teens" percentage (11-15%).

- Supply: Market conditions continue to strengthen, and HDD shortages are expected to benefit NAND demand.

c. Mid-term Outlook: The compound annual growth rate (CAGR) of bit demand for DRAM and NAND is expected to be "mid teens" percentage (15%).

5. Capital Expenditure (CapEx):

a. FY2025 CapEx: $13.8 billion. FY2026 CapEx is expected to be higher than FY2025, mainly for 1-gamma DRAM, HBM-related investments, and wafer fab construction.

b. Global Plant Progress:

- Idaho, USA: New plant has received CHIPS Act funding, with the first wafers expected in the second half of 2027; design of the second plant has begun.

- New York, USA: Environmental assessment is underway.

- Japan: Successfully installed the first EUV equipment for 1-gamma production.

- Singapore: HBM packaging and testing plant construction is progressing smoothly, expected to start production in 2027.

- Virginia, USA: Investment in expansion to support long-cycle demand for D4/LP4 and other mature products.

2.2 Q&A

Q: Regarding the expected $1.2 billion revenue growth next quarter, can you break down the contributions of DRAM and NAND? Also, what factors will affect changes in gross margin?

A: For next quarter's growth, DRAM's contribution will be greater than NAND. We expect gross margin to improve by 580 basis points quarter-over-quarter, mainly due to three factors: favorable product mix, strong pricing, and excellent cost control. The current market environment is very favorable, with tight DRAM supply and significant improvement in NAND conditions.

On the demand side, data center, traditional server, PC, mobile, and automotive markets are all growing; on the supply side, structural factors are also tightening supply. Therefore, we believe that with excellent execution, we can achieve this margin expansion.

Q: Do you have an updated forecast for the HBM market size? Will this number be larger than before? Additionally, can you comment on your outlook for next year's market situation?

A: I reiterate our forecast: the HBM market size is expected to reach $100 billion by 2030. HBM's growth rate will continue to outpace the overall DRAM market, including in 2026. As performance requirements increase, the value of HBM is also rising. Micron is well-prepared for this, with our HBM4 products leading the industry in performance and energy efficiency. Considering the multi-trillion-dollar data center infrastructure investments in the coming years, memory is at the core of the AI revolution, presenting a significant long-term opportunity for HBM. We are confident in the 2026 market and Micron's advantageous position with strong products, execution, and customer trust.

Q: How do you view the transition from HBM3E to HBM4? When do you expect the crossover in shipment volume or revenue between the two generations to occur next year (2026)? You mentioned that HBM3E pricing for 2026 has been finalized. Compared to current prices, will 2026 pricing be higher or lower? Do you expect Micron's HBM3E market share to remain stable or change next year (2026)?

A: Regarding HBM4, we will closely synchronize with customers' timelines to lead the production ramp. Our HBM4 products are industry-leading in performance and power consumption, with the first products planned for shipment in Q2 2026 and production ramping up in the second half based on customer demand.

We will not comment on specific HBM3E pricing but can confirm that most of the 2026 supply volume has signed pricing agreements with customers.

(P.S.: It is worth noting that negotiations for the upcoming year are at a critical stage. Given that NVIDIA is a key customer and Samsung is a potential new supplier, there are market concerns regarding price pressure from NVIDIA, which could adversely affect the gross margin for the HBM business. The company has revised its statement from "all capacity" being locked in to only the "vast majority" of its supply being secured.)

Overall, we expect the HBM market share to grow in 2026. Current supply is tight, and we anticipate a healthy supply-demand environment for the entire DRAM market next year, which is a positive signal for the profitability of both HBM and non-HBM products.

Q: How should we view the company's gross margin trend for the remainder of this year (2025)? Can the forecasted 51.5% for this quarter be seen as a baseline? As long as sales continue to grow, can the gross margin continue to expand at this level? I noticed that your cloud data center business operating margin has reached a strong 48%. How much room for improvement is there on this basis?

A: We do not provide specific guidance for future quarters, but we expect gross margin to improve sequentially. This is mainly due to several persistent factors: first, tight DRAM supply is driving a favorable pricing environment; second, the NAND business continues to improve; finally, we are shifting capacity more towards high-value markets while maintaining excellent cost control.

From a supply-demand fundamentals perspective, demand is driven by growth in data centers, edge computing, and automotive sectors; on the supply side, there are multiple constraints: our DRAM inventory is below target levels, industry support for DDR4 has slowed process conversion speed, and new capacity is both expensive and time-consuming, especially with HBM's high wafer consumption exacerbating capacity tightness.

Therefore, the market landscape entering FY2026 is very favorable for us. We expect gross margin to continue rising in the second quarter and reiterate that both HBM and non-HBM businesses will maintain healthy margins in FY2026.

Q: In the past month or two, DRAM demand led by hyperscale data centers (especially AI inference demand) seems to have reached a turning point. Can you talk about the breadth of this demand and its sustainability? You mentioned that supply tightness will continue into FY2026. Given this, should we expect to see normal seasonal slowdowns in the February quarter (the company's second fiscal quarter), or is the current supply constraint severe enough to offset seasonal effects, resulting in market performance far better than usual?

A: While we do not provide specific quarterly guidance, we can confirm that the AI-driven demand trend is very strong and extends beyond training to inference applications. This trend is expanding from data centers to edge devices such as smartphones and PCs.

Specifically, there are three main drivers:

  1. Data Centers: In addition to AI servers continuously driving demand for HBM, LPDRAM, and high-density modules, we also observe a recovery in traditional server demand.
  2. Smartphones: Newly released AI phones have significantly higher DRAM loads than previous generations.
  3. PCs: The rise of AI PCs and the end of the Windows 10 lifecycle together form another growth driver for DRAM demand.

In summary, the strong AI demand spans data centers, smartphones, and PCs, which is the core reason we are optimistic about 2026 demand. Combined with the previously mentioned supply tightness factors, we expect 2026 to usher in a very healthy supply-demand environment.

Q: You mentioned that net capital expenditure (Net CapEx) for FY2026 is approximately $18 billion. Can you roughly break down how much of this expenditure will be for front-end equipment and how much for clean room construction? What is the implied total capital expenditure (Gross CapEx) for FY2026 corresponding to this $18 billion net capital expenditure?

A: We have not yet provided a detailed breakdown of FY2026 capital expenditure, but it is clear that the vast majority of spending will be on DRAM, specifically including new plant construction, equipment for process conversion, and initial equipment installation for new wafer fabs (greenfield projects).

Regarding specific amounts, we typically communicate in terms of "net capital expenditure" inclusive of government subsidies. The $18 billion you mentioned is correct, which is the net capital expenditure target framework we have set for FY2026. We will not disclose the total capital expenditure amount for 2026 at this time.

As a reference, in FY2025, our total capital expenditure was $15.8 billion, and after receiving $2 billion in subsidies from the US, Singapore, and Japan governments, the net capital expenditure was $13.8 billion. You can understand our future capital expenditure structure based on this model.

Q: You mentioned that inventory days have reached target levels. Does this mean DRAM inventory is actually below target? Considering the strong demand for 12-Hi products (high-capacity HBM) and non-AI DRAM in the 3E process, do you expect DRAM inventory days to continue to decline by the end of this year (i.e., the end of the fiscal first quarter)? Given the overall tight supply, are your lead times extending? Do customers need to place orders earlier as a result? Does this customer pre-ordering behavior provide you with better market visibility? To what extent does it enhance your confidence that the market tightness will continue into the 2026 calendar year?

A: We expect inventory days to remain at or better than last quarter's levels. Specifically:

- DRAM: Due to very tight supply throughout the year, its inventory is expected to remain below our target levels.

- NAND: With market improvement and our strict supply discipline, NAND inventory days are also expected to decline.

Customers fully understand the current strong demand and tight supply situation, and we are working closely with them. Looking ahead, our supply strategy is to meet non-HBM product demand with the 1-gamma process while supporting HBM products with the 1-beta process. We will continue to maximize production efficiency while driving technology conversion using existing clean room space.

Q: To achieve performance 40% higher than the JEDEC standard, does the team need to redesign the base logic die? Has the launch of this higher-performance HBM4 version delayed the customer qualification timeline, or is the qualification progress on track? Even after achieving such high speed and bandwidth, does Micron's HBM4 still outperform competitors' solutions in terms of power consumption?

A: Our HBM4 product achieves exceptional performance beyond customer expectations, thanks to a combination of multiple advantages: innovative design, unique memory architecture, and advanced CMOS technology used in both the DRAM core chip and the base logic die. Notably, the base logic die is self-developed and manufactured by Micron, giving us a unique competitive advantage.

It is this comprehensive strength that enables us to achieve a bandwidth of up to 2.8 TB/s and speeds exceeding 11 Gb/s, fully preparing for the production ramp of HBM4. With such leading specifications, we will be able to closely follow customer demand and be at the forefront of HBM shipment growth.

Q: Assuming HBM capacity is fully sold out in the 2026 calendar year, can you quantify how large Micron's supply capacity or market opportunity will be at that time? If HBM demand continues to exceed expectations and your capacity is quickly sold out, do you have the ability to further increase HBM supply in the 2026 calendar year?

A: We do not disclose specific supply volumes, but the current progress is as follows:

- HBM3E: Most supply pricing and volume agreements have been locked in.

- HBM4 and overall 2026 supply: As customer demand becomes clear, we expect to complete all HBM (including HBM4) supply agreements for 2026 in the coming months.

Our industry-leading HBM4 product specifications put us in a very favorable position. Regarding your question about increasing production, the answer is that we have the flexibility to increase production. Our HBM market share is already on par with the overall DRAM share, and non-HBM business profitability is also healthy. Therefore, we can now flexibly manage the mix of HBM and non-HBM products. This flexibility comes from two aspects:

- Front-end manufacturing: HBM and some other products share 1-beta process wafers, providing substitutability.

- Back-end testing: Our investments over the past few quarters have ensured ample testing capacity.

In summary, we have the ability to adjust HBM supply based on market opportunities, but any decisions will be guided by return on investment (ROI) and overall investment discipline.

Q: You offer both Micron's self-developed base logic die and customized logic chips in collaboration with TSMC. Which option do you expect customers to prefer? What will the mix ratio of these two options be? From Micron's perspective, how difficult is it to switch between these two options based on customer demand? Is the conversion easy?

A: First, it needs to be clarified that the HBM4 we are about to mass-produce in 2026 will use our internally developed base logic die. The collaboration with TSMC to provide standard and customized solutions is for the next-generation HBM4E product, which is expected to be launched around 2027.

We firmly believe that the value of HBM will continue to increase, and by the HBM4E era, by offering customized solutions, we expect to achieve higher margins.

I want to emphasize again that the key to the success of our HBM4 product lies in the integration of multiple self-developed technology advantages: not only the design and architecture of DRAM but also the core use of Micron's own CMOS logic chip. This vertical integration capability brings us industry-leading performance and a unique competitive advantage.

<End Here>

Risk Disclosure and Statement of this Article:Dolphin Research Disclaimer and General Disclosure