🐦 推特Tae KimQwen · 实盘不做把握 4/10不做原因:已被定价
北京时间·🕒 发布 10:16:49时区推断·⟳ 抓取 10:17:12·⚡ 决策 09-02 10:23:40 +6m28s·55.6 天前
Dell Q2财报大超预期:收入470亿美元同比+58%、EPS $7.04同比+203%,单季AI服务器订单609亿美元创历史纪录,供应链全面紧缺
Dell财报确认AI需求爆发和供应链全面紧缺,方向明确正面。但A股AI硬件供应链全线逆利好下跌(光模块-5%、PCB-2%、电源-4%),唯一正反应的英维克已20日涨33%。市场正在对AI硬件板块进行系统性减仓/轮动,正面需求确认无法阻止下跌,说明当前定价受估值、CPO替代担忧、仓位调整等结构性因素主导,逆趋势做多无正期望。
📄 原文Tae Kim@firstadopter↗ 原文
Dell earnings call. Best hand picked excerpts: "Revenue was $47 billion, up 58%, and earnings per share was $7.04, up 2…
不做 中际旭创(300308.SZ)
Dell确认光模块短缺方向正面,但股价今日逆利好下跌-4.71%,20日跌-15.92%,处于深度下行通道;案例库胜率仅35%、中位-1.07%;市场正在定价CPO替代风险和估值消化,逆趋势做多无正期望
不做 英维克(002837.SZ)
唯一正响应候选(+1.97%),Dell点名CDU紧缺直接利好;但20日已涨+33.32%,短期涨幅过大追高风险高,日波动率4.17%意味着止损空间大;等回调至合理位置再考虑
不做 新易盛(300502.SZ)
今日-5.66%逆利好暴跌,20日-10.09%,距60日高-34.91%;与中际旭创同属光模块板块,市场正在系统性减仓,追多无正期望
不做 麦格米特(002851.SZ)
昨日出现从139.6到124.26的日内暴跌(-10.3%),典型分配/出货形态;今日继续下跌-3.99%,短期趋势极弱,不宜介入
展开详情(成交明细 / 研究档案 / 原文全文 / 决策记录链)
研究档案
结论摘要
Dell FY27Q2收入$47B(+58%)、EPS $7.04(+203%)、单季AI订单$60.9B均创纪录,管理层确认DRAM/光模块/CDU/ABF载板全面短缺。事件对A股AI硬件供应链方向明确正面。然而2026-09-02开盘后,中际旭创-4.71%、新易盛-5.66%、沪电股份-1.96%、深南电路-1.55%、麦格米特-3.99%,全线逆利好下跌;麦格米特昨日出现从139.6到124.26的日内暴跌(分配形态)。唯一正反应的英维克(+1.97%)已20日涨33%。案例库显示中际旭创历史胜率仅35%。结论:市场正在系统性减仓AI硬件板块,正面需求确认无法扭转趋势,追多无正期望。
关键理由
- Dell单季AI订单$60.9B创纪录、backlog从$51B飙升至$95B,确认AI基础设施需求加速
- 管理层明确点名光模块、CDU、ABF载板、电源机架短缺,直接利好对应A股供应商
- 需求持续超越供给(连续两个季度),且agentic AI推理需求87倍增长预期支撑长期景气
主要风险
- A股AI硬件板块正处于系统性减仓/轮动阶段,全线逆利好下跌说明市场定价受估值消化、CPO替代担忧和仓位调整主导,短期趋势无法被单一正面财报扭转
- 英维克20日已涨33%,追高面临回调风险;光模块股距52周高仍有35-40%空间但持续下行,抄底缺乏催化
反向证据
- 做多前查反向 · 来源 kb_search · 日期 2026-09-01 · 影响 weakens
中际旭创9月1日公告40-80亿元回购计划(上限1200元/股),属正面但未能阻止下跌;CPO替代叙事持续压制光模块估值
检索:光模块 中际旭创 新易盛 近期利空 下跌原因 CPO 竞争 原文 ↗ - 做多前查反向 · 来源 stock_quote · 日期 2026-09-02 · 影响 invalidates
全线逆利好下跌:中际旭创-4.71%、新易盛-5.66%、麦格米特-3.99%、沪电-1.96%、深南-1.55%;麦格米特昨日出现日内-10.3%的分配形态(139.6→124.26)
检索:A股AI硬件板块今日走势 - 做多前查反向 · 来源 search_case_library · 影响 weakens
中际旭创20笔ENTER决策胜率仅35%,中位收益-1.07%,历史表现持续为负
检索:中际旭创历史决策胜率
先例如何影响了这次决策
对本次决策的影响:先例不足,但当前市场实时反应(全线逆利好下跌)已给出明确方向信号
方向:无可用先例
涨幅量级:无法校准
形态:无
工具执行结果 · 成功 14 · 降级 0 · 失败 1 · 取消 0
- search_processed_events 成功 stock_tools HTTP · 245 ms
{"result": "search_processed_events:agent query=`dell ai server demand supply chain constraints dram optical | dell earnings ai server orders $60.9 billion | dell q2 revenue $47 billion eps $7.04`;结构化索引覆盖完整 72 小时;旧记录兼容扫描近 71.8/72 小时、4852 条本臂决策(完整),返回 0 条候选。未命中只表示这些 agent query 表达式没有召回候选,不能据此证明消息首发或不 - fetch_decision_experience 成功 stock_tools HTTP · 22 ms
{ - supply_chain_lookup 失败 stock_tools HTTP · 25 ms
FactSet Supply Chain 表结构缺失 - query_modeling_kb 成功 stock_tools HTTP · 50 ms
{"result": "modeling_kb 命中 15 条 (库快照 2026-07-27 · AI 算力链 19 页 · 本次命中证据最新 2026-06-15 · 其中 ⟨预测⟩ 9 条) —— 这是**定期重生成的快照**,不是活数据;比快照更晚的事实以你手上的信号/kb_search 为准。\n\n• [A] AI 超高容叠层数 · 表3 高容化技术路线图(AI server-grade 超高容 MLCC 关键参数演进) · CY2027E = 1200–1600(随尺寸迭代)/1300–1400(330μF 目标对应) 层 ⟨预测⟩ ⟨区间 1200.0~1600.0⟩[1][2 - kb_search 成功 stock_tools HTTP · 2,000 ms
{ - kb_fetch_document 成功 stock_tools HTTP · 648 ms
{ - stock_quote 成功 stock_tools HTTP · 594 ms
匹配到多个,请从 candidates 里选对的那只,用它的 canonical 代码再调用一次 stock_quote 确认。 - stock_quote 成功 stock_tools HTTP · 856 ms
{ - stock_quote 成功 stock_tools HTTP · 87 ms
{ - stock_quote 成功 stock_tools HTTP · 114 ms
{ - stock_quote 成功 stock_tools HTTP · 149 ms
{ - search_case_library 成功 stock_tools HTTP · 63 ms
{"result": "search_case_library「002463.SZ,002837.SZ,300308.SZ,300502.SZ」无历史决策命中(我们过去没在这些标的上决策过,或历史尚未回填)。\n◆ 002463.SZ:相关历史决策 39 条 · 其中该标的 ENTER 2 条(5.1%) · 已结算 2 笔 · 胜率 100% · 中位 +7.05% · 均值 +7.05% ⚠️只有 2 笔,样本太薄,只能当线索不能当规律\n◆ 002837.SZ:相关历史决策 66 条 · 其中该标的 ENTER 5 条(7.6%) · 已结算 5 笔 · 胜率 60% · 中位 +1.3 - kb_search 成功 stock_tools HTTP · 3,333 ms
{ - stock_quote 成功 stock_tools HTTP · 135 ms
{ - stock_quote 成功 stock_tools HTTP · 287 ms
{
原文全文
展开原文 · 3,045 字
Dell earnings call. Best hand picked excerpts:
"Revenue was $47 billion, up 58%, and earnings per share was $7.04, up 203%.”
“Over the past 12 months, we have booked more than $130 billion in AI server orders.”
“Our AI server momentum continues to accelerate. We booked $60.9 billion of AI orders in this quarter, the most in our history. We are also seeing AI-related tailwinds in traditional servers and networking, along with early signs of increased storage demand as customers prepare, manage, and protect growing volumes of data.”
“Demand is broadening across Neoclouds, sovereigns, and enterprise customers, and our customer count has surpassed 6,500.”
“We are seeing a growing trend of customers that require meaningful CPU compute capacity to support AI and agentic workflows.”
“If I look at the longer-term trends, I know you're a believer of this, but as we see it, agentic demand is reshaping the data center and the underlying infrastructure. Inference is past training and is pure demand on our industry. We think the tokens that inference drives is going to grow 87x to 3,600 quadrillion tokens by 2030. Training demand grows 5x to 850 zettaflops by 2030. Enterprise agentic is expected to be the single largest workload by 2028."
"We're expecting AI to be 75% of all data center demand by 2030, adding 200 GW of power over that same timeframe, and half of that, we believe, is right in our sweet spot with our customers, the Neocloud sovereigns and enterprises. If you look at that math, we think the opportunity in front of us is more than a trillion dollars over that time frame.”
“If we think about this across agentic workloads as we head towards physical AI and what's going to happen in manufacturing and IoT sensors and robotics, which drive tremendous amounts of multimodal unstructured data, Arthur likes to call it unstructured repositories. There's a lot of structured data in databases. The growth of that is immense, and we actually see it accelerating, not slowing down.”
“The constraints remain the same. DRAM, DRAM, DRAM followed by NAND, NAND, NAND. We have spotty CPU shortages. There are shortages with disk drives. If you go further down in the supply chain, just about every product going through a leading node is constrained. Mature nodes that are building MOSFETs, power ICs, microcontrollers, drivers are constrained. There's shortages of ABF substrate, T-glass, all of which we monitor. There's shortages in optical. The AI supply chain is working red line all out to build CDUs, power racks. Welcome to the life of a supply chain person at Dell. This is what we do, chasing parts. We love it.”
“But the underlying demand for the technology is significant. I think about the new use cases, that's all new use cases, all new growth, which is being driven by agentic AI, essentially running the harness, if that makes sense. We continue to be optimistic about the prospects. Again, demand outran supply last quarter. Demand outran supply this quarter. The pipeline remains robust.”
"Revenue was $47 billion, up 58%, and earnings per share was $7.04, up 203%.”
“Over the past 12 months, we have booked more than $130 billion in AI server orders.”
“Our AI server momentum continues to accelerate. We booked $60.9 billion of AI orders in this quarter, the most in our history. We are also seeing AI-related tailwinds in traditional servers and networking, along with early signs of increased storage demand as customers prepare, manage, and protect growing volumes of data.”
“Demand is broadening across Neoclouds, sovereigns, and enterprise customers, and our customer count has surpassed 6,500.”
“We are seeing a growing trend of customers that require meaningful CPU compute capacity to support AI and agentic workflows.”
“If I look at the longer-term trends, I know you're a believer of this, but as we see it, agentic demand is reshaping the data center and the underlying infrastructure. Inference is past training and is pure demand on our industry. We think the tokens that inference drives is going to grow 87x to 3,600 quadrillion tokens by 2030. Training demand grows 5x to 850 zettaflops by 2030. Enterprise agentic is expected to be the single largest workload by 2028."
"We're expecting AI to be 75% of all data center demand by 2030, adding 200 GW of power over that same timeframe, and half of that, we believe, is right in our sweet spot with our customers, the Neocloud sovereigns and enterprises. If you look at that math, we think the opportunity in front of us is more than a trillion dollars over that time frame.”
“If we think about this across agentic workloads as we head towards physical AI and what's going to happen in manufacturing and IoT sensors and robotics, which drive tremendous amounts of multimodal unstructured data, Arthur likes to call it unstructured repositories. There's a lot of structured data in databases. The growth of that is immense, and we actually see it accelerating, not slowing down.”
“The constraints remain the same. DRAM, DRAM, DRAM followed by NAND, NAND, NAND. We have spotty CPU shortages. There are shortages with disk drives. If you go further down in the supply chain, just about every product going through a leading node is constrained. Mature nodes that are building MOSFETs, power ICs, microcontrollers, drivers are constrained. There's shortages of ABF substrate, T-glass, all of which we monitor. There's shortages in optical. The AI supply chain is working red line all out to build CDUs, power racks. Welcome to the life of a supply chain person at Dell. This is what we do, chasing parts. We love it.”
“But the underlying demand for the technology is significant. I think about the new use cases, that's all new use cases, all new growth, which is being driven by agentic AI, essentially running the harness, if that makes sense. We continue to be optimistic about the prospects. Again, demand outran supply last quarter. Demand outran supply this quarter. The pipeline remains robust.”
决策记录链 · 只追加
每次领取产生一条记录,重新决策追加一条并以 supersedes 指向被替代的记录(ADR 0014)。
这条决策早于决策账本(2026-10-08 起逐条追加),没有追加记录。