Serenity 持仓雷达 › 深度长文 › 最恐慌的那三天,巨头在偷偷签合同
最恐慌的那三天,巨头在偷偷签合同
7.16–7.23 · 他这周发了 47 条,我把最关键的 9 件事按时间轴排了一遍
2026-07-23 · 约 8 分钟 · #周度复盘 #AI 供应链 #存储 #光模块
这一周很分裂。
前半周,屏幕上是韩国散户被强平、他自己单月回撤近一半;后半周,谷歌财报炸了、AMD 甩出 50 亿美元、光模块和存储集体反弹。
同一批资产,三天里被市场判了两次完全相反的结论。他这周发了 47 条帖,我按时间轴挑出 9 件有实锤来源的事,还原一下中间到底发生了什么。
第一幕 · 崩的是杠杆,不是合同
7 月 16 日,美光宣布和高通签下存储长期协议(LTA)。这种协议的意思是买方提前锁量、锁产能,通常还带 take-or-pay 条款——签了就得买,买不满也得付钱。按常理这是需求确定性的证据。
结果消息出来当天,美光股价反而跌了 5.37%。
Today, $MU announced it signed memory LTAs with $QCOM. Interesting reaction to see Micron proceeded to drop 5.37% right after. Doesn't quite feel like there's anything individually wrong with memory or AI names with all these structural agreements signed? More like the tail end of deleveraging / margin cascades.
两天后,一个数字给这个判断做了注脚。
Wow, apparently there's claims ~3.4% of the adult population in Korea faces liquidation risk? "As of July 13, over 1.2 million leveraged retail traders' accounts hit margin call lines, with 320,000 to 360,000 accounts force-liquidated by brokers." Regardless, just evidence recent crash was accelerated by liquidation cascades/deleveraging rather than fundamentals.
第二幕 · 他自曝亏了一半
7 月 18 日,他发了这周传播最广的一条,160 万次浏览。内容是自曝当月回撤 49.4%。
Feels bad, -49.4% drawdown this month after the recent crash. My portfolio is mainly AI chokepoints and bottlenecks. In the memory, photonics, robotics, and upstream semis, (on margin) which all tend to be higher beta than others. But reduced leverage recently from the crash. I see a lot of people making fun of the drop or AI names… But I have conviction that all these themes are backed by structural revenue growth or technological shifts. Especially considering a lot of retail view things on a week to week basis: no, my thesis isn't wrong yet if I project revenue inflection in H2 2027 and it's 2026 now.
这条帖里有个细节容易被忽略:他说自己「最近已经降了杠杆」。也就是说,他一边坚持论点,一边确实动手降了风险敞口。这两件事在他这里不矛盾。
第三幕 · 恐慌的同时,合同一份接一份
就在同一天,纽约时报报道了一笔交易:Meta 正在洽谈把自己的算力租给 Anthropic,规模 100 亿美元。
$META in talks to lease compute to Anthropic in a $10B dollar deal. Seems like they saw how profitable $SPCX $45B compute deal was. But just goes to validate Neocloud business models like $NBIS, $IREN, and co if hyperscalers are copying their homework. Source: NYT
四天后,AMD 出手,方向一样但更狠。
$AMD to invest $5B in Anthropic. Anthropic in turn will sign a 2 GW deal to buy AMD's latest chips, worth tens of billions of dollars. Looks like they're copying the $NVDA playbook with Anthropic last year.
同一周还有第三件事:英伟达通过 SEC 文件披露,它持有 Neocloud 龙头 Nebius 9.3% 的受益所有权。
$NVDA disclosed it owns 9.3% beneficial ownership of Nebius via SEC filings. Not exactly too new since it dates back to Nvidia's existing share position of 1.19M shares + $2B prefunded warrant of ~21M shares. But great for sentiment having Nvidia be a large shareholder of the Neocloud leader.
第四幕 · 谷歌把话挑明了
7 月 23 日凌晨,谷歌财报出来,这周的分裂被一份财报直接终结。
$GOOGL reported earnings today: EPS: $9.11 actual vs $2.91 expected Revenue: $119.7B actual vs. $116.98B expected Google Cloud Revenue: $24.77B +82% Y/Y, vs. 63-65% expected — this is a very large beat and most material part so far. Cloud Operating Income: ~$8.81B, implying 35.6% margin vs. 32.9% Q1. So initial read through is Google Cloud accelerating growth with expanding margins is genuinely bullish for AI demand.
他特意提醒了一句:EPS 那个 9.11 对 2.91 的巨大差距别太当真,因为里面很可能包含 SpaceX、Anthropic 等持股的账面收益,属于会计噪音。真正重要的是云业务的增速和利润率——那是真金白银的经营数据。
半小时后,更关键的信息来了:资本开支指引上调。
$GOOGL updates FY 2026 capex guidance: To $195B-$205B. Up from $180B-$190B. Google is also expected to significantly increase capex in 2027. This is perhaps the most bullish read through on upstream semis from a hyperscaler.
他在另一条帖里补了个关键换算:上季度财报里,谷歌说过数据中心和网络大约占资本开支的 40%。按新指引的中值算,这一块对应大约 800 亿美元。
至于这些开支为什么停不下来,他给了一个需求侧的数字。
$GOOGL now has 950M monthly Gemini users and processes 22B API tokens/min. Compared to 750M back in February. Absurd growth adding 200 million more users in a few months. Kinda explains why they ran out of compute, then cut allocations to $META and others. Very unlikely AI capex will slow down given.
第五幕 · 真正的卡点,写在交期表里
这是我认为他这周最有价值、但传播最少的一条:一张上游元器件的交期清单。股价会一天涨 30% 一天跌 20%,但交期不会——它是物理世界的排队长度。
他还提醒了一个容易误读的地方:MLCC 这类品类里,消费级和 AI 数据中心用的规格是混在一起统计的,所以不能简单地把交期拉长直接等同于 AI 需求。这句话本身比那张表更能说明他的方法。
存储那条线,他引了韩国券商的判断。
Kim Sunwoo of Meritz Securities: "This Is Not the Time to Sell Samsung Electronics and $SKHY" They claim markets are excessively misunderstanding the situation with semis. And that DRAM shortage will intensify in the second half of this year. - H2 2026: "suppliers can fulfill only 75%–80% of DRAM demand" - 2027: "fulfillment is expected to fall into the 60% range" Think the $INTC CEO said it best with his quote "no relief in memory supply or pricing until at least 2028".
然后,市场用一天把话收了回去
7 月 21 日到 22 日,反弹来了,而且是他口中「历史级」的单日反弹。存储、光模块、Neocloud、上游半导体几乎全线拉起,其中不少标的单日涨幅在两位数。
Memory names from Kioxia finished up a whopping +17.18%. Samsung up +6.15%. Optical networking from $SIVE, $LITE, and $AAOI are showing signs of recovery, all up 4%+. Neoclouds from $NBIS to $IREN all up 4%+ premarket. Does the Cramer effect work on the entire AI trade?
从最恐慌到「历史级反弹」,中间隔了不到 72 小时。而这 72 小时里,产业侧新增的确定性信息只有一个方向:合同在签、指引在上调、交期在拉长。
他的方法,其实就三句话
把这周 47 条帖压缩一下,他的判断框架大概是这样:
最后,两条花絮
以上全部内容为对公开帖文和公开财报/研报的整理与翻译,每条都标注了原始出处。
我不判断他说得对不对——他自己这个月也回撤了近一半。
方法可以复刻,判断永远是你自己的。
看全部深度长文 →以上为对公开信息的整理与翻译,不构成任何投资建议。