瑞银研报解读:美光 HBM 比预期更紧,NAND 短期噪音不改长期买入逻辑

marsbitPubblicato 2026-08-12Pubblicato ultima volta 2026-08-12

Introduzione

A UBS report on August 10th addressed Micron's near-term outlook. While Q3 NAND contract price increases moderated to ~20% from an expected 28%, pricing momentum is seen as slowing, not collapsing, with strong enterprise SSD demand offsetting consumer/PC softness. The key positive is a tighter-than-expected DRAM/HBM supply situation. NVIDIA has reportedly downgraded the HBM spec for its VR300 GPU due to constraints. However, UBS calculates this will actually increase total 2027 HBM bit consumption to 61.5B Gb from 58.7B Gb. HBM4/HBM4E pricing continues to exceed expectations, with the 2027 blended average selling price (ASP) forecast raised to a 79% YoY increase. Structurally, supply-side checks indicate Yangtze Memory is shifting a significant portion of incremental capacity to DRAM rather than NAND, potentially extending the NAND pricing cycle. DRAM supply tightness is expected to persist at least into Q2 2028. UBS forecasts a significant profitability reset for Micron, with EPS estimates for 2026-2028 well above consensus. The firm reiterates a Buy rating with a $1,625 price target, implying ~85% upside, citing long-term agreements that are shifting Micron's earnings trajectory from cyclical peaks to a structurally higher plateau. Cumulative free cash flow through 2028 is projected to exceed $450 billion.

撰文:Rita

美光三季度 NAND 合约价涨幅从预期 28%回落至 20%附近,瑞银在 8月 10 日研报中承认短期存在定价噪音。但 DRAM 比此前预期的更紧张,HBM4和 HBM4E 定价持续超预期,英伟达因供给受限已下调 VR300的 HBM 规格。瑞银测算 2027年 HBM 总消耗量反而从 587亿 Gb 增至 615亿 Gb,HBM 混合均价同比涨幅从 67%上调至 79%。瑞银维持买入评级,目标价 1625 美元,对应 2029 年预期每股收益约 11 倍。瑞银认为美光的盈利能力正在经历结构性重置,2028 年累计自由现金流有望突破 4500 亿美元。

HBM 供给偏紧,规格下调推高消耗量

HBM4和 HBM4E 的定价持续走强,超出瑞银此前的预期。英伟达因供给约束已调整 VR300的 HBM 规格,首款版本采用 384GB HBM4(8-Hi×24Gb×8×2),后续版本升级至 512GB HBM4E(8-Hi×32Gb×8×2),低于此前预期的 768GB。但瑞银认为规格下调后,2027 年整体 HBM 消耗量反而从 587亿 Gb 增至 615亿 Gb。单颗 GPU的 HBM 容量下降,但可用芯片数量增加,总消耗量反而上升。

瑞银测算 2027 年行业 HBM 混合均价同比上涨约 79%,高于此前预测的 67%。HBM4E 单颗价格可能突破 530 美元。美光 2027年 HBM 营收预计达 263.5 亿美元,同比增 134.6%,ASP 从每 Gb 20.03 美元升至 28.48 美元。DRAM 整体 ASP 预计 2027 年同比上涨 84.3%,2028 年仍有 15.5%的涨幅。DDR DRAM 方面,瑞银维持三季度合约价环比上涨约 20%的判断。CXMT 的产能扩张已被充分计入模型,供给影响可控,DRAM 供给偏紧格局至少持续到 2028 年二季度。

NAND 短期放缓,供给改善延长周期

NAND 是本次调整的主要变量。三季度 NAND 合约价环比涨幅从预期的 28%下调至约 20%,美光自身 NAND ASP 环比涨幅从 35%降至 23%。瑞银的判断是定价节奏放缓,不是需求崩塌。企业级 SSD 需求持续强劲,基本抵消了消费端和 PC 端的疲软。瑞银反而将 NAND 位元需求增速预测从 2026 年的 20%和 2027 年的 23%分别上调至 23%和 26%。

供给端出现一个结构性变化。瑞银的供应链核查显示,长江存储正在将相当一部分增量产能转向 DRAM(年化约 3 万片/月,占总产能增量 45 万片/月的三分之二),不是继续扩张 NAND。这将延长 NAND 的定价周期。瑞银预计 NAND 价格将在 2027 年四季度见顶,2026 年行业 NAND 营收约 3000 亿美元,2027 年升至 4950 亿美元。

盈利能力结构性重置,FCF 将超 4500 亿

美光 2026至 2028 年每股收益预测分别为 74.13 美元、184.89 美元和 265.65 美元,均远高于市场共识。2029 年每股收益预计仍维持在 190 美元以上。瑞银认为这不是周期顶部的短暂峰值,盈利能力的结构性重置正在发生。长期协议的覆盖正在改变美光的盈利结构,2028 年累计自由现金流有望突破 4500 亿美元。

瑞银目标价 1625 美元,对应 2029 年预期每股收益约 11 倍,折回 2028 年后隐含市盈率约 11 倍,与美光三年历史均值一致。以当前股价 877 美元计算,隐含约 85%的上行空间。DRAM 的紧张程度超出预期,HBM 定价持续走强,NAND 的短期放缓被企业级 SSD 的强劲需求和供给端的变化所对冲。美光的盈利轨迹正在从周期性峰值向结构性高位切换,长期协议正在改变市场对存储周期的定价方式。

免责声明

本文系潮向研究对第三方券商研究报告(瑞银,2026年 8月 10 日)的整理与解读,结合公开市场信息的整理。文中引述的评级、目标价、盈利预测及相关判断,均为该券商分析师的观点,仅代表其所属机构立场,不代表潮向研究的观点,也不构成任何投资建议。

市场有风险,决策需独立。本文不应作为买卖任何证券的依据。

Domande pertinenti

QWhat is the main reason UBS maintains its 'Buy' rating on Micron despite near-term pricing noise in NAND?

AUBS maintains its 'Buy' rating primarily due to a structural reset in Micron's profitability driven by strong DRAM/HBM dynamics and long-term agreements, not just cyclical peaks. The firm sees tighter-than-expected DRAM supply, persistently strong HBM pricing, and NAND supply-side improvements extending the cycle, leading to a significant upward revision in earnings and cumulative free cash flow projections.

QWhy does UBS expect total HBM bit consumption in 2027 to increase even though Nvidia has reduced the HBM specifications for its VR300 GPU?

AAlthough the HBM specification per GPU is lower, the reduction allows for a greater number of usable chips to be produced from the constrained HBM supply. This increase in the total number of chips shipped more than offsets the lower capacity per chip, leading to a higher overall HBM bit consumption forecast of 61.5 billion Gb, up from the previous 58.7 billion Gb.

QWhat is the key structural change in the NAND supply side mentioned by UBS, and what is its expected impact?

AThe key change is that Yangtze Memory Technologies Corp (YMTC) is reportedly shifting a significant portion of its incremental capacity expansion towards DRAM production (approx. 30k wafers/month), rather than continuing to expand NAND. This reduces the expected growth in NAND industry supply, which UBS believes will extend the NAND pricing cycle, pushing the price peak into Q4 2027.

QWhat are UBS's forecasts for Micron's HBM revenue and ASP (Average Selling Price) for 2027?

AUBS forecasts Micron's HBM revenue for 2027 to reach $26.35 billion, representing a year-on-year increase of 134.6%. The HBM ASP is projected to rise to $28.48 per Gb, up from $20.03 per Gb.

QWhat is UBS's price target for Micron and what is the implied upside from the current price mentioned in the article?

AUBS's price target for Micron is $1,625. Based on the current price of $877 mentioned in the article, this target implies an upside of approximately 85%.

Letture associate

OpenClaw Founder Peter Steinberg's Latest Speech: The Life and Death Revelation of an Open Source Project, How OpenClaw Emerged from the Eye of the Storm

Peter Steinberg, founder of the wildly popular open-source AI agent project OpenClaw, reflects on its turbulent journey from creation to near-collapse and resurgence. It began in late 2024 from his personal frustration with sending prompts from phone to computer. He built a WhatsApp relay that felt "magical," but initial excitement was limited to friends. After adding Discord support and launching a public server, the project exploded virally overnight, attracting massive attention, rebranding requests (from "Claudis" to OpenClaw), and even praise from NVIDIA's CEO. The intense scrutiny and pressure nearly broke Steinberg. The biggest technical crisis came from over-optimizing for a specific AI model provider, which suddenly canceled subscriptions, forcing a rushed shift to open-weight models. OpenClaw's weekly downloads plummeted before surging to a record high after being "declared dead." Steinberg realized he had stopped enjoying the build process and using his own product, making it a duty rather than a passion. His key lesson: "Fun is speed." Recovery came by refocusing on building for himself, forming a non-profit, and receiving crucial support (e.g., from NVIDIA for security). The team is now using OpenClaw to build OpenClaw itself, emphasizing its core value: a private, open alternative where "your agent, your machine, your life." He shares insights: the first user should be yourself; balance bug fixes with new features to maintain interest; the current bottleneck is reliable toolchains, especially for Mac testing. His story underscores that creation thrives on personal need, endures through storms when the joy of building returns, and persists by "living in the future and building what's missing."

marsbit15 min fa

OpenClaw Founder Peter Steinberg's Latest Speech: The Life and Death Revelation of an Open Source Project, How OpenClaw Emerged from the Eye of the Storm

marsbit15 min fa

US Stocks and Gold Rise Together, Bitcoin Rebound Absent, Have Bottom Signals Appeared?

While traditional assets like U.S. stocks and gold have rebounded strongly, Bitcoin continues to consolidate without following the uptrend. Since hitting an all-time high near $126,000 in October last year, Bitcoin has remained in a corrective phase, trading between $62,000 and $66,000 over the past month. Although U.S. spot Bitcoin ETFs have seen renewed net inflows recently—with the past week marking the best performance since mid-April—this buying pressure has been insufficient to counter persistent selling from miners and crypto companies. For instance, miner-linked wallets have seen their BTC balance drop by about 72% since late 2021. Some long-term indicators are showing early signs of a potential bottom. Metrics such as Reserve Risk, BTC price relative to its 2-year moving average, and the AHR999 index have entered historically oversold territory, suggesting improved long-term value. However, key indicators like MVRV Ratio, NUPL, and the Fear & Greed Index have not yet reached the extreme levels typically seen at major cycle lows. Furthermore, over 53% of Bitcoin supply remains in profit, and stablecoin liquidity is still contracting, indicating the market may need more time for a full capitulation phase. In summary, while select bottom signals are emerging, Bitcoin lacks the broad-based confirmation needed to declare a definitive cycle low. A sustained recovery will likely depend on continued ETF inflows, a reduction in structural selling pressure, and further deterioration in key on-chain metrics toward historical extremes.

marsbit17 min fa

US Stocks and Gold Rise Together, Bitcoin Rebound Absent, Have Bottom Signals Appeared?

marsbit17 min fa

After Yushu Got Hot, a 68-Year-Old Auto Parts Boss Bet 1.85 Billion

After staying relatively low-profile for years, Shanghai Bate Technology, a traditional auto parts supplier, has recently been thrust into the spotlight as a leading humanoid robotics concept stock. This shift is largely tied to the impending IPO of robotics company Unitree, with which Bate collaborates on prototype development. Bate's founder, 68-year-old Jin Kun, is making a bold bet on this new frontier. The company is investing 1.85 billion yuan to build a major R&D and production base in Kunshan for planetary roller screws—a key component in robot joints. With an annual capacity of 2.6 million sets planned, plus overseas expansion, this represents a massive, forward-looking commitment from a firm whose 2025 revenue of 2.3 billion yuan still relies almost entirely on its traditional automotive business (over 98%). The investment highlights a strategic pivot. Bate aims to leverage decades of expertise in metal processing, precision manufacturing, and heat treatment from the automotive sector and apply it to the burgeoning robotics industry. However, this move carries significant risk. The company has not yet secured formal supplier contracts or begun mass deliveries for its roller screws. It is essentially betting on future demand in a market that remains nascent. The broader industry context adds to the uncertainty. While multiple Chinese manufacturers are racing to invest in similar robotics component capacity (totaling over 6 billion yuan), the downstream humanoid robot market is still in its early stages. Estimates put 2025 global shipments at only around 18,000 units, primarily for entertainment, education, and data collection rather than large-scale industrial use. This creates a critical mismatch: the speed of upstream component factory construction and capital investment is far outpacing the commercialization and adoption of the robots themselves. Bate's story, driven by Jin Kun's decisive gamble, encapsulates the high-stakes transition facing traditional manufacturers seeking a "second growth curve" in robotics—a future full of potential but currently backed more by market anticipation than concrete financial results.

marsbit21 min fa

After Yushu Got Hot, a 68-Year-Old Auto Parts Boss Bet 1.85 Billion

marsbit21 min fa

EIP-8363 Explained: Burning Staking Rewards to Save Ethereum? Independent Node Operators Might Be the First to Leave

Analysis of EIP-8363: A Proposal to Burn Staking Rewards and its Potential Impacts on Ethereum A new Ethereum Improvement Proposal, EIP-8363, has sparked significant debate. Its core mechanism is to progressively burn validator consensus rewards, aiming to reduce the issuance rate to 0% once 50% of the total ETH supply is staked. Proponents argue this would cap the staked ETH ratio, enhancing credible neutrality and censorship resistance by discouraging excessive centralization among large staking operators and liquid staking token (LST) providers. However, a critical analysis suggests the proposal may create more problems than it solves. While it compresses the ETH staking range where profitability turns negative (from 70-100 million ETH down to 50-60 million ETH), it does not fundamentally alter the competitive disadvantage of independent stakers. Due to economies of scale, large operators would still retain advantages in costs and MEV opportunities. Independent stakers are predicted to be the first to face negative real returns under the new curve. The economic repercussions could be severe. A sharp decline in base staking yield would disrupt DeFi ecosystems built around this yield, potentially reversing incentives for borrowing ETH and triggering large-scale unwinding of leveraged staking strategies. This could lead to significant selling pressure on ETH as professional staking operations become unprofitable and exit. The proposal is criticized for being based on uncertain future market equilibriums and for potentially imposing costs that outweigh its stated benefits, without fully addressing the underlying issues of centralization it aims to solve.

marsbit21 min fa

EIP-8363 Explained: Burning Staking Rewards to Save Ethereum? Independent Node Operators Might Be the First to Leave

marsbit21 min fa

Trading

Spot
活动图片