--- title: "MU (Trans): Customer demand commitments unchanged; signs first 5-yr partnership agreement ---" type: "Topics" locale: "zh-CN" url: "https://longbridge.com/zh-CN/topics/39358133.md" description: "Below is Dolphin Research's Trans of the $Micron Tech(MU.US) FY26 Q2 earnings call.For our earnings take, see 'Micron (MU): AI-fueled memory surge — can it break the cycle?'Key management highlights: 1) Industry outlook: industry DRAM bit shipments in 2026 are expected to grow in the low-20% range (approx. 22%)..." datetime: "2026-03-19T03:29:50.000Z" locales: - [en](https://longbridge.com/en/topics/39358133.md) - [zh-CN](https://longbridge.com/zh-CN/topics/39358133.md) - [zh-HK](https://longbridge.com/zh-HK/topics/39358133.md) author: "[Dolphin Research](https://longbridge.com/zh-CN/news/dolphin.md)" --- > 支持的语言: [English](https://longbridge.com/en/topics/39358133.md) | [繁體中文](https://longbridge.com/zh-HK/topics/39358133.md) # MU (Trans): Customer demand commitments unchanged; signs first 5-yr partnership agreement --- **Below is Dolphin Research's compiled** $Micron Tech(MU.US) **FY26 Q2 earnings call Trans. For the earnings take, see '**[**Micron (MU): AI-fueled Memory Surge — Can It Break the Cycle?**](https://longbridge.cn/zh-CN/topics/39356514?channel=SH000001&invite-code=294324&app_id=longbridge&utm_source=longbridge_app_share&locale=zh-CN&share_track_id=12482f65-395b-49b9-bb28-e4ef68a460a0)**'.** **Key takeaways from Micron management:** **1) Industry outlook: DRAM bit shipments are expected to grow by low-20s% in 2026 (approx. 22%),** with supply constrained by cleanroom capacity, a rising HBM mix, and declining bits per wafer. **NAND bit shipments are expected to grow ~20% in 2026.** **2) HBM: HBM4 36GB 12-high entered volume production and shipments in FY26 Q1,** designed for NVIDIA Vera Rubin. **HBM4 yield ramp is tracking faster than HBM3E,** and HBM4 16-high (48GB/cube, +33% vs. 12-high) has been sampled. **HBM4E on 1-gamma is progressing well,** with volume production expected in 2027. **3) GPM:** The industry is supply-constrained, and **tightness is expected to persist beyond 2026,** supporting near- to mid-term pricing. 81% of next-quarter guidance already reflects HBM4 growth. At today’s GPM level, **incremental price increases have diminishing marginal benefit to margins.** **4) CapEx:** FY26 CapEx will exceed $25bn (vs. prior guide of $20bn). **FY27 CapEx will rise significantly, with construction-related spend alone over $10bn,** and equipment spend also up YoY. **5) Long-term agreements:** **Signed the first 5-year Strategic Customer Agreement (SCA),** structurally different from LTAs (typically 1-year), with multi-year specific commitments. Negotiations for SCAs are underway with multiple customers across markets. In today’s extremely tight supply environment, customers are highly motivated to enter such strategic agreements. SCAs aim to provide supply certainty to customers and specific demand commitments to us across cycles, with binding terms for both sides. **6) Customer demand coverage:** **For key customers, mid-term fulfillment remains at roughly 50% to two-thirds of needs,** consistent with three months ago. **More analyst small-group Trans to follow. Stay tuned.** **I. Micron earnings recap** 1\. **Shareholder returns:** BOD approved a 30% hike in the quarterly dividend to $0.15 per share. $350mn of buybacks were executed this quarter, limited by CHIPS agreement terms. 2\. **Outlook:** FY26 Q3 revenue guide of $33.5bn (±$0.75bn), GPM ~81%, OPEx ~$1.4bn, and EPS of $19.5 (±$0.40), all record highs. The single-quarter Q3 revenue guide exceeds the company’s full-year FY24 revenue. 3\. **Key metrics:** FY26 Q2 revenue $23.9bn (+196% YoY, +75% QoQ), marking the fourth straight quarterly record. GPM 75% (+1,800bps QoQ), nearly doubling YoY, and non-GAAP EPS $12.20 (+682% YoY). Free cash flow was $6.9bn, a quarterly record and +77% QoQ. DRAM revenue was $18.8bn (79% mix, +207% YoY), and NAND was $5.0bn (21% mix, +169% YoY). 4\. **Capital structure and cash flow:** Cash and investments at quarter-end were $16.7bn, with liquidity over $20bn. $1.6bn of debt was repaid this quarter, bringing total debt down to $10.1bn, and net cash to a record $6.5bn. The company received two rating upgrades to a solid BBB. 5\. **CapEx:** FY26 CapEx to exceed $25bn, raised mainly due to the Tongluo acquisition and U.S. fab builds. FY27 CapEx will increase markedly, with construction spend up by over $10bn YoY and equipment spend also higher. **II. Earnings call details** **2.1 Management remarks** 1\. **DRAM technology and capacity** a. The 1-gamma node is ramping yields at the fastest pace in company history. It is expected to become the DRAM bit workhorse by mid-2026 and could be Micron’s largest-production node ever. b. The 1-delta node will expand EUV usage, leveraging the latest EUV tools to optimize cleanroom space efficiency. This should enhance overall fab productivity. c. **Industry DRAM bit shipments are expected to grow by low-20s% in 2026 (slightly above prior view),** with supply growth constrained by cleanroom limits, a higher HBM mix, and lower bits per wafer. 2\. **HBM** a. **HBM4 36GB 12-high entered volume production and shipments in FY26 Q1,** purpose-built for NVIDIA Vera Rubin. b. **HBM4 yield ramp is expected to outpace HBM3E.** HBM4 16-high has been sampled at 48GB per cube, a 33% capacity uplift vs. 12-high. c. **HBM4E development is on track on the 1-gamma node,** with volume production targeted for 2027 to support next-gen AI compute platforms. d. Customization options on HBM4E open deeper co-development opportunities with customers’ R&D teams. This should strengthen design-in stickiness. e. The advanced HBM packaging site in Singapore is expected to meaningfully contribute to HBM supply in 2027. This will diversify geographic capacity. 3\. **NAND and data center SSD** a. The G9 NAND node is set to become the bit workhorse by mid-2026. QLC bit mix reached a quarterly record. b. High-performance PCIe Gen6 data center SSDs based on G9 are in volume production. This enhances Micron’s performance leadership. c. The 122TB high-capacity SSD is seeing strong market traction. Its per-watt sequential read throughput is 16x that of comparable-capacity HDD setups. d. Data center SSD share rose to a new high for the fourth straight year. FY26 Q2 data center NAND revenue more than doubled QoQ. e. NAND demand far exceeds supply and is expected to remain so. This underpins pricing and mix. f. **Industry NAND bit shipments are expected to grow ~20% in 2026.** 4\. **End markets** a. **AI-driven data center DRAM and NAND bit TAM will, for the first time in 2026, exceed 50% of industry TAM.** This marks a structural shift in demand. b. Traditional server demand is solid, supported by agentic AI workloads and a broad refresh cycle. **Server shipments are projected to grow low-teens% (~12%) in 2026.** c. **PCs and smartphones: constrained DRAM/NAND supply may drive a low-double-digit decline in 2026 unit shipments,** while AI lifts content per box. Agentic AI PCs are recommended at 32GB+, and personal AI workstations at 128GB; flagship phones at 12GB+ DRAM have risen from under 20% a year ago to nearly 80%. d. AEBU revenue hit a record, with Auto + Industrial topping $2.0bn. L2+ ADAS is accelerating, and L4 autonomy will need 300GB+ memory vs. under 16GB today per vehicle on average. e. Robotics is seen as a 20-year growth vector. Demand for humanoid robot compute platforms could rival high-end L4 autos. 5\. **SCA (Strategic Customer Agreements)** a. **Signed the first 5-year SCA,** fundamentally different from typically 1-year LTAs, with multi-year specific commitments. b. **Negotiations are ongoing with multiple customers across multiple markets.** These could broaden coverage and duration. c. SCAs are designed to provide predictability and stability for both parties’ business models, and to support investment confidence. This enhances visibility through cycles. 6\. **Global capacity expansion** a. Tongluo: early close of the acquisition from PSMC. Meaningful product output is expected starting FY28, and a second equal-size cleanroom is planned to start construction by end-FY26. b. Idaho: the first fab is expected to output mid-2027. Site prep has started for a second fab. c. New York: the first fab has broken ground. Progress is ahead of plan. d. Japan (Hiroshima): cleanroom expansion is progressing well. This will add DRAM capacity. e. Singapore: a new NAND fab has started construction, with first output expected in 2H28. This supports long-term NAND scaling. f. India: the new AT facility has begun commercial shipments. It will be one of the largest single-story AT cleanrooms globally. **2.2 Q&A** **Q: How sustainable is the 81% GPM guide? As HBM4 mix rises, how should we think about the Aug. quarter and beyond?** A: The Q3 GPM guide is up 600bps QoQ, and we do not guide Q4 margins. **We have indicated market tightness is likely to persist beyond 2026,** and margins reflect a multi-year AI investment cycle, most of which still lies ahead. AI requires more and higher-performance memory, which shows up in margins. **The 81% guide already includes the impact of HBM4 growth,** and we still expect a strong market backdrop. Note that at this GPM level, **further price increases have diminishing marginal uplift to margins.** Mix and productivity also matter. **Q: How do SCAs differ from LTAs? Are there multi-year volume/price commitments, or does pricing reset annually? What about cancellation if the cycle turns?** A: SCAs are multi-year, whereas LTAs are typically 1-year. **Given extreme tightness, customers are highly motivated to secure planning certainty via strategic agreements.** SCAs aim to provide supply certainty to customers and specific demand commitments to us across different cycle stages. These agreements include binding terms for both parties, but details are confidential. We will not disclose specifics. **Q: How are you allocating supply across end markets? Any risk of crowding out PCs and smartphones?** A: Supply is extremely tight across all end markets, and demand trends are strong everywhere. **Price-sensitive consumer segments like PCs and phones may see some demand impact from higher prices,** but overall demand remains robust. We have always pursued diversified end-market exposure, which is critical for us. Data centers are a growing share of TAM and will naturally command a larger allocation, driving Micron’s growth, but PCs, smartphones, Auto, and Industrial remain important to our mix. **Q: You previously said some customers could only be served at 70% of needs. What is the status now?** A: On the last call we indicated that **for some key customers, we could meet only 50% to two-thirds of demand in the mid term,** and that remains the case today. **Q: Do SCAs include mechanisms that protect a margin floor in a downturn?** A: We will not disclose SCA terms for confidentiality reasons. **We have completed one SCA and are in discussions with multiple customers.** These are multi-year with specific commitments and are binding agreements designed to provide visibility and stability. At this stage we cannot share more detail. Discussions continue across markets. **Q: How will you deploy the large cash build? FY26 FCF could reach $35–40bn, and year-end cash could exceed $50bn. Will you reserve for future buybacks, and could CHIPS-related buyback limits be revisited?** A: We are pleased with business execution and balance sheet improvement, with record net cash and FCF in Q2, up 77% QoQ. Based on Q3 guidance and CapEx, cash flow could roughly double QoQ. We will keep strengthening the balance sheet and net cash. We received two rating upgrades to a solid BBB this quarter, and capital allocation priorities remain a strong balance sheet and organic investment to advance technology and capacity, with current ROIC above 30% and moving toward 50%. We will stay disciplined. The 30% dividend raise announced today reflects confidence in prospects and cash returns, and we will have ample room to return cash via buybacks, including offsetting SBC dilution and opportunistic repurchases. **Q: How broad are SCA discussions? Only hyperscalers, or wider? Any CapEx front-load requirements or ROIC-linked pricing?** A: What we can share is the signed SCA is with a large customer. **The core purpose is to invest confidently for future supply while gaining better demand visibility and business model stability.** SCA discussions are underway with multiple customers across markets. We will not comment further beyond that. Terms are customer-specific. **Q: Any change to the HBM TAM outlook? Last quarter’s 40% CAGR implied about $50bn this year. Are some players favoring DDR5 over HBM?** A: It is true that **non-HBM DRAM currently carries higher gross margins than HBM.** HBM demand remains strong, and we have not updated the HBM TAM outlook we provided previously. Data center demand is robust across DDR5, LP, and HBM, and we continue to manage the portfolio. Beyond data centers, we also focus on maintaining share in other key markets, where we see strong growth opportunities in HBM, LP, SOCAMM, DDR5, and data center SSDs. **Q: Can G9-based eSSDs continue to drive QoQ growth through the rest of this year and next? Thoughts on High Bandwidth Flash (HBF) as a new memory tier?** A: Data center SSDs are a strong growth area ahead. NAND supply is very tight and demand is strong, with data center SSDs a key growth driver for NAND. Micron has a full SSD portfolio in both capacity and performance (TLC and QLC) and is well positioned. As we continue shifting revenue toward higher-profit pools and higher-value markets, we will capture SSD growth opportunities, and we are pleased with the trajectory and plan. On HBF, it offers capacity advantages but also inherits NAND constraints such as write speed, power, and data retention. There may be niche workloads for HBF, but it is very early and will need deeper customer engagement to validate the commercial value proposition. **Q: Are multi-year SCAs driven by next-gen HBM architectures, especially base-die customization needing earlier, longer customer collaboration?** A: We will not specify customer types. However, SCAs indeed deepen our partnerships with customers, extending into tighter R&D collaboration and joint roadmap planning. This is an important source of value from SCAs. It should enhance alignment and speed. **Q: GTC featured much on LPU architectures and increased SRAM use. What is the long-term memory impact, and how do you plan capacity to avoid future oversupply?** A: **On SRAM and LPU, these architectures make AI infrastructure more efficient, and any innovation that improves AI infrastructure efficiency benefits the entire AI ecosystem.** It grows the pie. The LPU architecture works alongside Vera Rubin, which uses significant HBM and DRAM, and the NVIDIA Groq LPX LPU architecture uses 12TB of DRAM per rack. These approaches process workloads more efficiently, improving token economics, token speed, inference scale, and energy efficiency. We view them as complementary, not substitutes, for existing HBM and DRAM demand. Note that enterprise AI deployments are still very low today. Across industries and economies, the runway is vast, and we are excited about HBM, LP, DRAM, SOCAMM, and SSDs meeting future demand. From last year to this year, DRAM content in advanced AI accelerators has doubled. These factors drive the current supply shortfall and extend to edge devices, smartphones, and PCs. **Q: Will HBM4 share start at the targeted 20–25% or ramp over time? Can you grow share in the Vera Rubin generation?** A: We previously shared that by FY25 Q3, our HBM share had reached parity with our DRAM share target. We then said we would manage HBM as part of the overall portfolio and would stop disclosing HBM share quarterly. We are comfortable with HBM positioning. There will be demand for both HBM4 and HBM3E in 2026, and we will supply both, confident in our market position and portfolio management. **Q: How sustainable is the 81% GPM vs. prior peak levels (low-60s%)? Do customers react differently seeing such high supplier margins?** A: **The industry is supply-constrained, with tightness likely persisting beyond 2026, which supports near-term pricing.** AI is a transformational, secular driver, and it needs more and higher-performance memory. Memory helps lower token cost and energy per token, increases token counts, and enhances AI capability, driving harder problem sets and agent applications that need more tokens and memory. Margins reflect a re-rating of memory’s value as an efficient way to monetize AI from the data center to the edge. Mean reversion to historical averages is not the right lens. On supply, there are enduring structural constraints: low inventories, lower bits per wafer from node migrations, a rising HBM mix, and the physical reality that new capacity requires greenfield builds that take time. On demand, rising value and structural supply challenges are both durable, and we are investing in capacity and R&D to raise memory value, with customers recognizing this through long-term agreements. **Risk disclosure and disclaimer:**[**Dolphin Research Disclaimer and General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### 相关股票 - [Micron Tech (MU.US)](https://longbridge.com/zh-CN/quote/MU.US.md) - [NVIDIA (NVDA.US)](https://longbridge.com/zh-CN/quote/NVDA.US.md) - [GraniteShares 2x Long NVDA Daily ETF (NVDL.US)](https://longbridge.com/zh-CN/quote/NVDL.US.md) - [XL2CSOPNVDA (07788.HK)](https://longbridge.com/zh-CN/quote/07788.HK.md) - [XI2CSOPNVDA (07388.HK)](https://longbridge.com/zh-CN/quote/07388.HK.md) - [YieldMax NVDA Option Income Strategy ETF (NVDY.US)](https://longbridge.com/zh-CN/quote/NVDY.US.md) - [Direxion Daily NVDA Bear 1X ETF (NVDD.US)](https://longbridge.com/zh-CN/quote/NVDD.US.md) - [T-Rex 2X Long NVIDIA Daily Target ETF (NVDX.US)](https://longbridge.com/zh-CN/quote/NVDX.US.md) - [T-Rex 2X Inverse NVIDIA Daily Target ETF (NVDQ.US)](https://longbridge.com/zh-CN/quote/NVDQ.US.md)