--- title: "MU (Trans): Supply tightness to last through 2027" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/42171053.md" description: "Below is Dolphin Research's compilation of Micron Technology's FY26 Q3 earnings call Trans. I. Core takeaways from the print.1) FQ4 guidance (record): revenue of $50 bn (±$1 bn), GPM about 86%, and OpEx around $1.65 bn. EPS guided to a record $31 (±$1), based on approx. 1.15 bn shares. Tax rate ~15%.The FQ4 GPM guide already factors in a clear slowdown in price increases. For FY27, OpEx is expected to rise by about $1 bn to expand R&D." datetime: "2026-06-25T02:39:28.000Z" locales: - [en](https://longbridge.com/en/topics/42171053.md) - [zh-CN](https://longbridge.com/zh-CN/topics/42171053.md) - [zh-HK](https://longbridge.com/zh-HK/topics/42171053.md) author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)" --- # MU (Trans): Supply tightness to last through 2027 **Dolphin Research's memo from Micron Technology's FY26 Q3 earnings call** **I. Core Takeaways** 1\. **Record FQ4 guide**: revenue of approx. $50bn (±$1bn); GPM around 86%; OpEx about $1.65bn; EPS a record ~$31 (±$1) on ~1.15bn shares; tax rate ~15%. The FQ4 GPM guide already embeds a notable slowdown in price hikes. FY27 OpEx is expected to rise by about $1bn for R&D expansion, skewed to 2H. FQ4 FCF is guided to surge again. 2\. **Capital return and balance sheet**: will step up capital returns starting Dec 9, 2026 (two years after CHIPS agreement), targeting 100% of excess cash over the long term. Quarter-end cash and investments hit a record $30.2bn; repaid $4.4bn of debt in FQ3 (including $4.3bn cash tender for senior notes), ending with $5.7bn of debt, $24.4bn net cash, and a WAM to Apr 2035; all three agencies upgraded ratings this fiscal year (to BBB+). 3\. **CapEx**: FQ4 CapEx around $10bn; FY26 about $27bn (net of Gov. subsidies). FY27 quarterly CapEx will run above FQ4 levels, with over half of the YoY increase as construction CapEx to pre-build long-term cleanroom capacity. 4\. **SCA disclosures (material accounting change)**: starting the May quarter, disclosing RPO under ASC 606. FQ3-end RPO was over $5bn; including post-FQ3 signings, total SCA RPO is about $100bn (based on minimum volumes and floor prices, a conservative proxy, not expected revenue; actual revenue is expected to exceed RPO materially). For 14/16 SCAs, cumulative revenue at contract floor prices totals about $100bn. Signed SCAs are expected to bring ~$22bn combined cash deposits and related financial commitments (~$18bn cash deposits, ~$4bn LCs); received over $400mn in FQ3 and expect about $10bn more in FQ4. Customer deposits are financing cash flows and do not affect FCF; they will be refunded in tranches mainly in the back half of the contracts. Once all planned SCAs are in place, about half or more of company revenue is expected under SCAs; fixed or capped-price deals (caps near current CQ2 market prices) should represent ~40% of revenue, and even at floor prices, GPM would still be well above any prior cycle peak. **II. Call Details** **2.1 Key management commentary** 1\. **AI-driven demand and supply-demand setup** a. Data center revenue exceeded $25bn in FQ3, implying a run-rate over $100bn. Data center SSD revenue topped $5bn, more than doubling QoQ. b. DRAM and NAND demand continues to meaningfully outstrip industry supply, with tightness expected through CY27 and beyond. Even assuming supply gradually improves in CY28, management does not see supply catching up with demand. c. Supply is structurally constrained: long greenfield timelines, skilled labor shortages, and complex approvals and power infrastructure. Rising node complexity slows bit growth and lifts wafer needs; each HBM generation’s higher trade ratio further crowds out non-HBM supply; NAND supply is constrained as cleanroom space shifts to DRAM. d. AI systems (GPU/ASIC/CPU) are architecturally dependent on memory performance and capacity, making memory a strategic asset and creating Micron’s largest ever differentiation opportunity. 2\. **SCA strategy (biz. model shift)** a. 16 SCAs signed across data center, consumer, and auto, typically 5-year terms (CY26–CY30; auto generally 3 years), including 4 very large customers, 3 mid-sized, and the rest smaller auto customers. b. Take-or-pay structures with binding multi-year volume commitments; largest agreements generally include a price band with a ceiling near current CQ2 market levels and a floor for the term; a minority are fixed-price or market-following without bands. At the band floors, GPM still far exceeds any historical peak. c. SCAs provide supply assurance to customers and give Micron demand visibility, stability, and stronger long-term financials, representing a fundamental commercial model shift. 3\. **Tech and product roadmap** a. 1-Gamma DRAM and G9 NAND ramps are on track and may be the largest volume nodes in Micron’s history. Next-gen DRAM/NAND nodes are progressing well, with mass production expected from 2H CY27. b. HBM4 12-high is ramping at 2x the pace of HBM3E 12-high; shipments already exceed $1bn of HBM4 revenue. Yields are also expected to reach mature levels meaningfully faster than HBM3E 12-high. c. Mix will continue shifting to higher-performance, higher-value (higher cost per bit) products (LP5→LP6, DDR5→DDR6, next-gen HBM). Along with greenfield capacity adds, blended DRAM cost per bit is expected to rise from here; SCAs include provisions to negotiate future premiums for such new products. 4\. **End markets** a. Data center: CY26 industry DC DRAM/NAND bit shipments expected to more than double vs. two years ago; agentic AI is reshaping DC infrastructure from accelerator racks to CPU and storage racks. CY26 industry server units are revised up to high-teens growth (from low-teens), with double-digit growth in traditional servers and stronger AI accelerator servers; with severe memory allocation, customers prioritize unit maximization, modestly tempering DRAM per server growth. b. PC and smartphone: unit declines but industry revenue should grow, reflecting resilient demand for premium models at higher price points. On-device AI (e.g., OpenClaw, NemoClaw and other agentic platforms) plus a replacement cycle should lift long-term memory demand for PCs/phones. c. Auto: ADAS drives content growth; L2+ and above models carry over 5x the memory content of standard cars, with their mix more than doubling this year to over 20% and projected to exceed 40% by 2030. d. Robotics: physical AI is accelerating; humanoid robots use roughly 10x the memory of a typical L2+ car. A multi-year upcycle in memory demand is expected to begin in the back half of this decade. 5\. **Outlook and capacity expansion** a. DRAM and NAND should remain tight beyond CY27; CY26 industry DRAM bit shipments to grow low-to-mid 20% (slightly above prior), NAND around 20% (in line). Micron’s DRAM supply growth is roughly in line with the industry, while NAND growth is slightly below. b. Multi-year EUV supply agreement with ASML supports the One Delta node and subsequent EUV adoption. c. Global expansion: Idaho ID1 (first wafers mid-2027E), ID2 (late 2028E), and New York campus (broke ground in Jan). Tongluo, Taiwan’s existing 300k sq ft site is expected to reach volume shipments by mid-2027 (about a quarter ahead of plan), with a second EUV-capable cleanroom under construction; Singapore will be the advanced packaging CoE, contributing materially to HBM packaging capacity from 1H 2027; Manassas, Virginia has started one-alpha DDR4 volume for auto/industrial/medical/aero-defense long-tail demand. **2.2 Q&A** **Q: Under floor-price scenarios, how much annual revenue is locked/secured? (14/16 SCAs at $100bn cumulative seems ~$20bn/yr, well below the current guided run-rate, yet you also say 40% of revenue will be under SCAs. Please reconcile.)** A: On signed SCAs, cumulative revenue at floor pricing is estimated at ~$100bn, but we expect actual revenue to be well above that. Importantly, even at floor pricing, our profitability/GPM exceeds any historical peak. Roughly 20% of DRAM and ~30% of NAND volume is already under these SCAs, implying about 25% of revenue visibility across the full term. **RPO is disclosed using floor pricing as the accounting basis, but we fully expect actual revenue to exceed RPO materially.** **Q: How should SCA revenue flow into models quarter by quarter? For example, how much of Aug quarter (FQ4) revenue is under SCAs? When do you reach full run-rate (FQ4 next fiscal year)?** A: You will see the disclosures in the 10-Q — we provide next-12-month revenue tied to each RPO group. For example, agreements closed within FQ3 had over $5bn of RPO, mapping to about $1.8bn of next-12-month revenue (these are smaller, as the CEO noted, mainly auto). In FQ4, we will report about $100bn of RPO for 14 of the 16 SCAs and disclose the next-12-month revenue in the 10-K, giving a sense of the ramp cadence. Remember RPO is the minimum enforceable intersection of volume and price — a floor, not our expected outcome. RPO will change each quarter as we sign new contracts, add priced volume commitments, and recognize revenue against obligations; this ASC 606 disclosure is new for us. We still expect about 40% of revenue to be under such RPO-tied commitments, and even at floors, GPM will be well above historical peaks. **Q: What is the purpose of cash deposits? Are they akin to escrow/collateral (usable if customers cancel)? Any linkage between deposits and RPO?** A: Across signed agreements, deposits and financial commitments total about $22bn, including roughly $18bn of cash deposits. We received over $400mn in FQ3 and expect around $10bn more in FQ4. These cash deposits appear in financing cash flows and do not affect FCF; we hold them through the contract term and refund in tranches as milestones are met, heavily weighted to the back half. The ~$4bn difference between $22bn and $18bn reflects letters of credit. **Q: What is the economic substance of this cash? If customers get it back, why commit? Is it tied to take-or-pay? It is not a prepayment?** A: It is not a prepayment but a separate customer commitment reflecting binding take-or-pay agreements, held by us and evidencing mutual performance obligations. **For Micron, these agreements provide visibility and firm volumes, enabling major capital investments and deeper technical collaboration**; customers gain supply assurance and leading technology. We view this as win-win and are very pleased with the nature of the agreements, their business impact, and what they represent in our model transition. **Q: Are these deposits unrestricted cash usable for CapEx? When sizing capital returns (especially after CHIPS Act benefits fade), do you include them in gross cash, or do you need structurally higher gross cash given future refunds?** A: The cash is unrestricted. We do not plan to change our target gross cash level in the near term. We will maintain ample liquidity to support operations, including future deposit refunds as obligations are met, and to fund priority investments (large supply projects and R&D). Again, refunds are back-half weighted over the contract life. **Q: On HBM revenue — thoughts on CY26 share and total revenue? By 2027, can HBM margins converge toward DDR5, or remain permanently below?** A: We are very pleased with HBM4 and have shipped over $1bn already. Strategically, we aim for HBM share roughly in line with our DRAM share — important because HBM’s trade ratio consumes significant wafers and pressures non-HBM supply. Keeping HBM share near DRAM share lets us serve diversified end markets that need non-HBM supply across DC, consumer, auto, and industrial. We will not comment on next year’s pricing, but HBM is an area of strong leadership for Micron (success across HBM3E 8-high, 12-high and now HBM4, with a strong roadmap and high execution confidence), carries higher $/bit than non-HBM, is critical across the AI stack from DC to edge, and offers strong strategic value and ROI. **Q: Of the 4 very large and 3 mid-sized customers, how many are data center? Will there be more DC announcements? Is the $100bn tied to large/mid or small customers? What does a typical DC SCA look like?** A: Our very large customers include DC, and collectively the large/mid/small span DC, consumer, and auto. As noted, large agreements generally include price bands (floor and ceiling), with ceilings set at CQ2 levels — already reflected in FQ3 actuals and FQ4 guidance and implying unprecedented profitability; at the floors, GPM still far exceeds any prior cycle peak. These are multi-year with strong demand visibility and customer commitments, and include financial commitments such as cash deposits. **Q: GPM at 86% — can it persist? Is there a ceiling? As SCAs phase in, should margins normalize somewhere between the current mid-80s and prior peak low-60s (say mid-70s)? What range should long-term investors model for 2027/2028?** A: We are not guiding beyond FQ4. **At current GPM levels, incremental benefit from further price increases is diminishing.** That said, we have raised our market view and expect tightness to extend beyond 2027. As memory’s strategic value is increasingly recognized and AI capability gains require more and higher-performance memory, we keep allocating bits to higher-performance DC and edge applications and optimize bit placement with customers, including SCA customers. New supply begins to materially ramp from mid-2027 and grows through 2028, with some startup costs that are absorbed as utilization ramps, providing operating leverage over time. Overall, given our tech position, world-class portfolio, and strong execution, we are confident in the trajectory. **Q: On LTA floor pricing, you cited prior peak GPM in the low 60s (~62%). If a 64GB server DRAM module floors around $700 vs. ~$1,500 today, implying ~$10–12/GB floor vs. mid-$20s/GB today — are LTA bands roughly low-teens to mid-$20s/GB?** A: We will not discuss specific pricing. Reiterating: GPM at floor prices is far above the peaks we have ever experienced in prior cycles — well above those highs. We are not disclosing price details. The key is that these SCAs provide real visibility, strength, and durability of demand, materially accelerating our financial performance and business model transition. **Q: You guided DRAM bit growth low-to-mid 20% and NAND around 20% this year, both undersupplied — can you quantify 2027? Will the 2027 supply gap be twice this year’s? How should we think about 2027 imbalance?** A: **We see overall tightness in 2027 and expect it to persist beyond. We are working to increase supply, but adding capacity takes a long time; node migrations yield fewer bits per node, and HBM trade ratios put enormous pressure on overall supply growth.** Even if supply begins to improve gradually in 2028, demand should remain strong — AI is a very long-term trend still in its early days; token economics need more memory, and AI system performance is highly constrained by memory capacity, performance, and bandwidth. As compute needs grow and customers pursue unprecedented transformation opportunities with record infrastructure spend, demand is exceptionally strong, memory sits at the core as a strategic asset, and securing supply is a top priority — reflected in multi-year agreements signaling confidence in demand growth. Therefore, we **are pushing supply higher but still expect tightness beyond 2027.** **Risk disclosures and statements:**[**Dolphin Research Disclaimer and General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer) ### Related Stocks - [MU.US](https://longbridge.com/en/quote/MU.US.md)