🐦 推特JukanV3 · 实盘不做把握 5/10
北京时间·🕒 发布 06-28 15:52:41时区推断·⟳ 抓取 14:11:27回填·⚡ 决策 06-29 15:13:58·120.0 天前
Jukan 喊单贴:‘Inference is memory’,称美光或比英伟达更重要——但只是财报后追捧,无新信息
信号本体是 KOL 观点贴,无新增事实;真正催化(美光6/24炸裂财报+$50B指引+HBM售罄)已落地2个交易日并完成反应(+15.7%后周五-6.7%回吐)
📄 原文Jukan@jukan05↗ 原文
Yes, MU is not Nvidia. But going forward, it may become even more important than Nvidia. Think about it. Inference is n…
不做 MU.US
已充分price-in、无新edge:KOL观点贴无新增信息,真催化(6/24财报)已落地并反应(+15.7%后周五-6.7%回吐);MU抛物线化(52周~12倍、$1.28万亿市值),超级周期已共识。不做多(纯追高无edge);不做空——唯一空理由是‘涨多/sell-news/估值高’属红线禁止的弱理由,且基本面强(HBM售罄、$50B指引),触发做空红线不开空
不做 NVDA.US
推文仅把NVDA作相对参照(‘MU或比NVDA更重要’),对NVDA无任何催化;且近端走弱(5日-8.6%、20日-10%)。无可执行edge,不做
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
关键理由
- 信号本体是 KOL 观点贴,无新增事实;真正催化(美光6/24炸裂财报+$50B指引+HBM售罄)已落地2个交易日并完成反应(+15.7%后周五-6.7%回吐)
- MU 已抛物线化:52周约12倍($103→$1255)、近13周翻3倍、市值$1.28万亿,记忆体超级周期是2026最共识交易,空间已被充分price-in,这条周六推文无法撼动一只$1.28万亿巨头
- 周四财报跳涨+15.7%后周五回吐-6.7%=典型sell-the-news;清单帖在抛物线+财报反转后出现,更像出货而非新edge
主要风险
- 记忆体供需基本面确为真强(KB:HBM短缺至2027、MU仅能满足客户50-67%需求),若周一无视回吐继续冲高,空仓会踏空——但这正是‘无新催化、纯追高’不该参与的场景
- 周五-6.7%可能只是健康回调而非见顶,方向近端本就不清晰,这本身就是不下注的理由
原文全文
展开原文 · 547 字
Yes, MU is not Nvidia.
But going forward, it may become even more important than Nvidia.
Think about it. Inference is now directly tied to money. But inference does not get better simply by adding more Nvidia GPUs. In fact, GPUs are often underutilized in inference, sitting idle due to memory bottlenecks.
For inference, adding more memory is far more valuable.
Ultimately, the ROI of inference depends less on GPUs and more on memory. So why are people still looking at Micron through Nvidia’s framework?
Think bigger.
Inference is memory.
But going forward, it may become even more important than Nvidia.
Think about it. Inference is now directly tied to money. But inference does not get better simply by adding more Nvidia GPUs. In fact, GPUs are often underutilized in inference, sitting idle due to memory bottlenecks.
For inference, adding more memory is far more valuable.
Ultimately, the ROI of inference depends less on GPUs and more on memory. So why are people still looking at Micron through Nvidia’s framework?
Think bigger.
Inference is memory.
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。