【编者按】本文是推演文章，编者以老阎的思路写成。凡加粗处均为老阎本人原话，出自润涛阎文集；其余文字是编者依他的逻辑所作的分析。文中凡涉及「女方系间谍」之处，均为美国官员与媒体的指控；当事人的私德，本文不作评判。本文不构成任何投资建议。

大智若愚33

一、表象：两波曝光

先把日子排出来。先后次序摆清楚，「时机」二字才有着落。

时间	事件
2025 年 10 月	《纽约时报》首次调查新加坡 Megaspeed 公司；同月英伟达成为白宫舞厅工程的捐款方（《华盛顿邮报》后来报道金额为 2500 万美元）
2025 年底	据《纽约时报》，特朗普听取了有关 Alice Huang 的情报简报，未采取行动（特朗普本人否认，称「不久前」才知道）
2026 年春	一份由国务卿鲁比奥主持的情报备忘录在少数高官间传阅
2026 年 5 月 12—15 日	特朗普访华
10 月 29 日	《纽约时报》报道：美国情报认定 Alice Huang 为北京收集情报；英伟达称相关指控「去年已彻查并澄清」
10 月 30 日	英伟达失去全球市值第一的位置
11 月 3 日	中期选举。民主党大胜：截至 11 月 10 日，众议院约 254 席，参议院 54 席（两席未定）
11 月 4 日	三位共和党籍众院委员会主席致函黄仁勋，要求他 11 月 18 日出席闭门简报
11 月 6 日	《金融时报》报道备忘录中「浪漫关系」的内容
11 月 7 日	《纽约时报》报道商务部批准约二十家中国企业购买 H200 芯片
11 月 8 日（周日）	黄仁勋致信全体员工，承认婚外情；董事会宣布独立调查；特朗普称他是「一个正派的人」
11 月 9 日	英伟达股价跌逾 4%；股东诉讼提起
11 月 10 日	舒默把此事与 2500 万美元舞厅捐款挂钩，要求调查英伟达对国家安全与出口政策的影响
11 月 12 日	《纽约时报》报道黄仁勋与一位直接向他汇报的女性副总裁也有恋情；《华尔街日报》报道董事会准备撤换他。当天黄仁勋辞去总裁、CEO 及董事职务，即时生效；首席财务官克雷斯出任临时 CEO，独立董事哈维·琼斯出任首位董事会主席。股价开盘一度跌 13%，收跌约 7%
11 月 17 日	英伟达第三季度财报
11 月 18 日	众院三委员会闭门简报
2027 年 1 月 3 日	新国会开幕，民主党接管众参两院的委员会与传票权

日子一排，就看出这件事是分两波出来的：「间谍」那一波出在选举前五天，「婚外情」那一波出在选举后五天。

第一波，按美国的说法叫「十月惊奇」。老阎对它的看法很干脆：

美国的大选都是这么过来的，所谓的十月惊奇，都是炒作，而非真的司法官司。伪造、道听途说、篡改、等等五花八门的招术，大选一过，全部烟消云散。经历过无数大选的中间选民，对这一套根本就熟视无睹。都是给本方选民打鸡血的。

这一次正应了他的话：选民在乎的是钱包、伊朗战争和对 AI 的不安，没人在看英伟达。

第二波，第一个选时机的人是黄仁勋自己——《金融时报》登出「浪漫关系」两天后，他在一个股市不开门的周日写信承认，赶在财报和国会闭门简报之前。老阎的方法是先反着看：

不论什么文章，都要先反着看是否成立，时时刻刻用批判的眼光审视，没通过验证的，越是好听越可能是谎言。

反着看，「有人精心计算了曝光时机」这个说法，至少在这一步上不成立。但备忘录春天就写好了，情报据说一年前就到了白宫——东西早就有了，什么时候拿出来，才是选择。

问到这里，「时机」就只是表象了。真正要问的是：谁和谁在较量？照老阎的逻辑，在英伟达身上较量的是三股力量：资本、国家，以及两种制度。

二、根源之一：资本要什么

资本要的东西，老阎只用了一句话：

资本的本质是逐利的。

而芯片的利润在哪里，他也说得明白：

现代社会的经济发展，不是靠卖家，而是靠巴结买家。

美国的芯片超过一半的买主是中国。

所以英伟达要卖给中国，这不是道德问题，是资本的天性。但老阎也看到了资本自己的两难：

美国财团最恨的不是中国在贸易方面赚钱，而是无论多么先进多么复杂的新科技产品，一旦被中国掌握，立刻变成白菜价，让别人没法玩下去。

既要卖，又怕卖出一个对手——这是资本在中国问题上永远解不开的结。

资本要卖，就要朝中有人。老阎讲 2008 年金融危机，有一句被引了无数遍的话：

巴菲特为何在这个危机处理时刻当机立断用50亿美元买入高盛GOLDMAN SUCKS？因为他知道所谓7000亿救市就是给金融大鳄注资。而财长和美联储主席都与高盛有关系。财长就是高盛出来的。巴菲特为何不买雷曼？因为雷曼朝廷没人。

英伟达朝中有没有人？有，在白宫。2500 万美元的舞厅捐款，特朗普一句「我真的相信他」，选后第四天商务部批准约二十家中国企业购买 H200 芯片——这扇门是开着的。老阎生前早就料到了这一步：

就是川普连任后，他也可能顶不住美国大公司财团们的利益要求而卖给中共芯片。他在大选期间可以做到敢跟财团对抗，到了连任后，他可能会跟财团们妥协。

在这一层上，资本和白宫站在一边。

三、根源之二：国家要什么

国家要的东西，老阎也只用了一句话。他讲 2008 年为什么救花旗不救雷曼时，立下一条压倒一切的规矩：

在美国，党争永远是第二位的，而国家利益才是第一位的。

今天美国的国家利益是什么？是在 AI 上领先中国，是芯片和情报不落到对手手里。在这一点上，资本和国家就分道扬镳了：资本要把芯片卖给最大的买家，国家要防着这个买家。

这里有一点必须说清：这场较量的分界线，不在「商界」和「政界」之间，而在政界内部。白宫站在资本一边；站在国家利益一边的，是国会和情报系统，而且是两党都有：第一封要求黄仁勋到国会说明情况的信，出自三位共和党委员会主席；那份备忘录，由共和党政府的国务卿主持。选后接手国会的民主党，只是让这一边的力量多了传票权。

老阎讲过一个故事，最能说明这种分界线的分量：

当初财长保尔森为何哀求波洛西？因为他出手救一些大银行，而对同样的银行袖手旁观，如果没有波洛西的认可，鲍尔森下台后说不定就受到司法调查……而有了波洛西的支持，他就不担心秋后算帐了。

保尔森手里有财政部，还要去求反对党的议长，为的是不怕秋后算账。白宫的门开着，只管得了今天；国会的门开不开，管的是明天。11 月 3 日以前，英伟达两扇门都开着；11 月 3 日以后，它只剩白宫这一扇。

国家这一边，这次还多了一股民意。选民用选票表达了对数据中心和 AI 的不安，背后是中产的钱包。老阎对美国的市场与政府，有一段很冷静的分工：

市场的职能是资本获利而带动经济的总体发展和科学技术的创新与利用；政府的职能是再分配，通过税收与福利的平衡来控制贫富差距。美国「把权力关进钱笼」其市场职能发挥到了淋漓尽致地步，而政府的职能则伤害了除了房子外没什么产业的「中产阶级」（事实上是劳动人民）。

中产觉得 AI 的好处归了资本、代价归了自己，便用选票要求政府出手。老阎亲眼见过一次民意逼政府出手。那是他刚到美国的上世纪八十年代，对象是如日中天、连纽约洛克菲勒中心都买下来的日本：

美国政府不得不迎合民意，对日本掐脖子。并分散依赖日本制造业的供应链，新的高科技发明创造毕竟还是在美国科学界手中。

接下来便是 1985 年《广场协议》逼日元升值、1986 年《美日半导体协议》、1987 年里根对日本商品加征 100% 关税。结局，老阎也写了：

日本被美国掐脖子后，30年沉寂不前。

但这一次有一个根本的不同：被要求「慢下来」的，不是外国的对手，而是美国自己在跟中国赛跑的那匹马。所以照老阎的规矩，国会可以卡数据中心、开听证、立规矩，拿黄仁勋做靶子；但不会真的把 AI 上的领先让给中国。国家要的是管住资本，不是废掉资本。

四、根源之三：两种制度的较量

资本和国家在美国能这样较劲，本身就是民主制度的特征。老阎把两种制度的区别，压缩成了一对好记的谐音：

美国是「把权力关入笼子」，但没人告诉你，这笼子是什么东西。其实，这笼子是钱，简称「钱笼」（发音「乾隆」好记）；而中国正相反，钱永远被政权压住，也就是说「把钱关在笼子里」，这个笼子就是政权，简称「政笼」（正龙）。有「乾隆盛世」和「正龙拍虎」两个成语。把权力关入钱的笼子里，带来的是盛世（资本主义）；新成语「正龙拍虎」，说的是假货盛行。

（「正龙」一词，出自 2007 年的「华南虎照片」事件：陕西农民周正龙声称拍到了野生华南虎，省林业厅开新闻发布会予以认定，还发了奖金；不久网友发现，照片里的老虎与一幅年画几乎一模一样。2008 年，陕西省政府通报：所谓华南虎，是一张纸老虎道具。周正龙以诈骗罪被判刑，13 名官员受到处理。「正龙拍虎」从此成了弄虚作假、官方背书的代名词。老阎拿它来配「政笼」，再贴切不过：钱被关进权力的笼子，造假也就有了权力撑腰。）

在「钱笼」里，英伟达可以跟国家讨价还价，可以捐款、游说、找白宫开门；国会可以反过来查它、卡它、传它。这场较量是公开的、吵闹的，还要靠选票来裁决。

在「政笼」里，没有这种较量。老阎说：

中国「把钱关进政权的笼子」（官本位），导致资本家无外乎都是掌权者的白手套。

专制体制的最大特点就是可以国家不计成本，不考虑短期效益。

把这两种制度放在同一张桌子上，美人计就有了新的读法。如果指控属实，那就是一个「政笼」里的国家，用它最擅长的手段，去接近「钱笼」里最值钱的那个人。老阎写过：

来个美人计，男人很少有经得起诱惑的。很难很难的。英雄难过美人关。

但他同样写过：

肯定有冤的，也肯定有不冤的。

指控目前未经证实，北京也已否认。本文只说一点：无论这一次是不是局，两种制度在芯片和情报上的较量，都不会因为这件事的真假而停下来。

这场较量的长期走向，老阎有一个很不讨美国人喜欢的判断：

技术，在当今时代，无法彻底封锁。

再过也许10年，芯片主要制造大国可能是中国。

但他并不认为「政笼」会因此笑到最后。「政笼」不计成本、效率高，它的命门却在市场：

经济危机发生了，「政笼」的牢固性将面临高压测试。组成笼子的竹板便发生内讧而展开决战，因为竹板与竹板的和谐往往是表面的。……所以，这种「政笼」体制的稳定程度在市场化后取决于市场经济是否发生危机。发生危机时刻，就是轰然倒塌的时刻。

也就是说，「钱笼」的较量吵在明处，吵完了靠选票收场；「政笼」的较量藏在暗处，平时看不见，一到危机就是生死。

他还看到了一个有意思的方向：两种制度正在向对方靠拢。

中国在由「正龙」（官本位）向「乾隆」（市场）转化；而美国在由「钱笼」向「政笼」（政治正确）方向转化。

这一次的中期选举，正是「钱笼」松动、政治重新压过资本的一个信号。一边是「政笼」怕危机，一边是「钱笼」在松动，英伟达正夹在这两种变化的中间。

五、三股力量为什么在英伟达身上交汇

丑闻落在黄仁勋身上，不是偶然。英伟达同时是三样东西：

资本里最大的对华受益者：中国是它最大的买家之一，H200 许可刚批下来；
国家利益里最核心的资产：美国在 AI 上领先中国，靠的就是它的芯片；
两种制度较量中的首选目标：如果指控属实，对方要接近的，正是这个人。

三股力量在同一个人身上交汇，任何一点火星都会烧起来。

那么，是谁「精心计算」了这把火？老阎不信有这样一只手。他专门写文章否定过「影子政府」：

为何润涛阎不认可有有形影子政府呢？一句话就是：没必要。「有形影子政府」的存在是多余的。因为「无形影子政府」足以在利益的驱使下令有关人物们配合默契。

这个被称为上帝的无形能量，其实用一个字表示就是「钱」——无所不能；用两个字表示就是「利益」——无时不在。

不需要谁坐在密室里下棋。资本、国家、两种制度，三方的利益在选后这一周自然对上了：国会要传票，鹰派要收紧出口，民意要给 AI 踩刹车，竞争者和做空者要股价下跌。彼此不必商量，却「配合默契」。

受损的一方也一样清楚：黄仁勋本人，英伟达的股东，还有站在资本一边的白宫。

六、他为什么会辞职

三股力量交汇的结果，四天后见了分晓。还是先把日子排出来：

11 月 8 日（周日）：黄仁勋写信承认婚外情；董事会宣布独立调查；特朗普称他是「一个正派的人」。
11 月 10 日（周二）：公司表示，他将「不间断地」继续领导英伟达。
11 月 12 日（周四）：《纽约时报》报道，他与一位直接向他汇报的女性副总裁也有恋情（当事双方未回应）；《华尔街日报》报道，董事会准备撤换他。当天，英伟达向证交会提交公告：黄仁勋辞去总裁、CEO 和董事职务，即时生效，公告未说明理由。股价开盘一度跌 13%。

第一段关系曝光后，董事会保了他；第二段曝光当天，他就走了。所以要问的不是「他为什么会出事」，而是同样是婚外情，为什么第一次扛得住，第二次扛不住。照老阎的逻辑，答案有四层。

第一层：私德扛得住，治理扛不住。

第一段关系，对象是公司外的人，按美国的规矩属于私生活。第二段不同，对象直接向他汇报，这就从私德变成了公司治理：上司和下属有了私情，升迁、薪酬、项目都说不清了。老阎在书里借一位美国教授之口说过：

千万别搞办公室恋情，那可是一辈子玩完的事。

他也记过美国政坛的先例：

第一个是金里奇，他信誓旦旦政治家要遵守道德，结果很快被民主党查出他与助手搞婚外恋，便灰溜溜辞职了。

（按：金里奇 1998 年辞去议长时，公开理由是中期选举失利，婚外恋是后来才曝光的。此处照录老阎原话。）

第二层：一次可以是失足，两次就是本性。

老阎记下的另一位议长，倒在「不止一个」上：

他出一百万美元，只要哪个女人出来提出议长跟你上过床便可得到一百万美元。当即有四个女人出来承认……这议长立刻辞职了。

只有一段关系的时候，可以说是一时糊涂；第二段一出来，人们就会问：还有没有第三段？董事会再也不敢替他担保了。老阎对人性早有定论：

要是人人都能坐怀不乱，哪里还有柳下惠的名字流传下来？

第三层：克林顿能挺过去，黄仁勋挺不过去。

克林顿的丑闻比这大得多，却没有下台。老阎的解释是：

候选人第一条件是有领袖魅力……克林顿不需要干什么大事，他那一点头一挥手，就能稳坐钓鱼船。搞了那么大的丑闻，照样没有被弹劾搞下台。

政治人物的命运握在选民和议员手里，魅力可以抵消丑闻，弹劾还需要参议院三分之二的票。CEO 的命运握在董事会手里，董事会对股东负责，不对魅力负责。老阎说：

资本的本质是逐利的。

创始人在资本面前也没有豁免权，老阎举过最有名的例子：

争夺个人利益的竞争非常残酷无情。乔布斯自己开的公司，竟然自己被开除。

黄仁勋一夜之间从公司最大的资产变成最大的风险，股价开盘就跌了 13%。在「钱笼」里，资本对风险的反应，比对恩情的记忆快得多。

第四层：白宫的门开着，但这扇门管不到董事会。

白宫一直站在他这边。可辞不辞职，不由白宫说了算，由董事会说了算；而董事会要怕的，是选后易手的国会。老阎讲过保尔森为什么去求反对党的议长：

而有了波洛西的支持，他就不担心秋后算帐了。

民主党 1 月 3 日接管传票权，国会议员已要求保全英伟达对华出口的文件。董事会若在第二段关系曝光后还留着他，到了听证会上，被追问「你们知道什么、做了什么」的就是董事会自己。这也说明了他为什么连董事席位一起辞掉：切得越干净，公司离秋后算账就越远。

没有一只手，只有利益。

不需要有人「逼宫」，也不需要谁「精心计算」。股东要止损，董事会要自保，国会要传票，鹰派要收紧出口：几方的利益在同一天对上了。第五节说过的那个「无形影子政府」，在 11 月 12 日这一天又配合了一次。

所以，他不是因为婚外情辞职的。婚外情是表象；根源是他从资本的资产变成了资本的风险，而且是在国会换手、对华芯片较量最紧的时候。第一段关系，资本还能替他扛；第二段关系涉及下属，资本就不扛了。

七、照老阎的逻辑推下一年

以下预测只用两样东西：老阎生前反复论证过的规律，和截至 11 月 12 日已经确认的事实；不采用任何他人的评论或预测。

其中最大的变数，是司法部对 Megaspeed 的刑事调查。法庭若认定芯片或信息流向了不该去的地方、违反了出口管制，就是「有罪」；反之则是「无罪」。法官是独立的，但起不起诉、怎么起诉，由司法部决定，而司法部属于行政部门，归特朗普管。照第二节的推理，白宫站在资本一边；老阎也早就说过，特朗普连任后「他可能会跟财团们妥协」。所以照老阎的逻辑，一年之内，「不起诉或从轻」的可能性大于「有罪」。能改变这一点的只有国会：民主党从 1 月 3 日起握有传票权，听证若逼出新的证据，司法部就不得不动。

刑事案从起诉到判决往往超过一年，下面的「有罪」「无罪」也可读作「起诉成立」「不起诉或不成立」。照此，预测分为两类：不受判决影响的，和受判决影响的；后者每条先列更可能的「无罪」情景。

（一）不受判决影响的

数据中心会被卡，AI 挡不住。

老阎的原理：

美国政府不得不迎合民意。

任何人都阻挡不了AI最终取代生物人，只能是时间的早晚而已。

推论：民意已经用选票说了话，州和地方的数据中心审批会明显放慢；但「给 AI 减速」的联邦立法，即便新国会通过，也会被特朗普否决；而国会两院都凑不齐推翻否决所需的三分之二多数。这一条由中产的钱包驱动，与法庭无关。即便判决有罪，AI 成了国安竞赛，联邦层面减速只会更难。慢下来的是速度，不是方向。

中国会加速自造芯片。

老阎的原理：

中国的举国体制在川普搞出中兴事件后会快速加大在芯片领域的巨额投资，用不了十年中国就会在芯片领域追上来。这是川普最愚蠢的一步棋。这等于给中共提出了发展方向与动力。

技术，在当今时代，无法彻底封锁。

短时间内中共造不出5纳米芯片，不代表几年后还是造不出。因为时代无法阻止科学家流动了。

推论：北京的决心早在中兴事件后就下定了。判决有罪，门关得更紧，是又一次「给中共提出了发展方向与动力」；判决无罪，方向也不会改变。

（二）受判决影响的

婚外情：退出新闻，还是换个身份回来。

老阎的原理：

所谓的十月惊奇，都是炒作，而非真的司法官司。……大选一过，全部烟消云散。

肯定有冤的，也肯定有不冤的。

无罪：私事会很快退出新闻。国安这一层一塌，剩下的只有私德，而私德已经由辞职了结了。间谍指控本身，也会像十月惊奇那样烟消云散。

有罪：私事会以新的身份回到新闻里，不再是丑闻，而是案情背景。法庭认定的只是芯片去了哪里，证明不了美人计；但人们会回头追问，那段关系和这条渠道之间有没有联系，国会也会把私事和出口放在一起问。

对华芯片出口：国家和资本，谁说了算。

老阎的原理：

在美国，党争永远是第二位的，而国家利益才是第一位的。

奥巴马办不到，因为不卖给中国芯片等高科技，美国大公司就有损失。

资本的本质是逐利的。

无罪：先紧后松，而且松得更快。国安这面盾牌被法庭拿掉，资本重新占上风；白宫本来就站在资本一边，会顺势开门。

有罪：门会关紧，而且难再打开。判决书证明了芯片外流，国家利益压过资本，两党都会站到国家利益一边，白宫也顶不住。

抓技术间谍：风，还是潮。

老阎的原理：

芯片大战，等于情报系统脱钩。美国还会继续抓在美华人技术间谍，中共也会抓其它领域里的情报间谍。

报复历来都是互相的。

「抓间谍」运动，希望不要抓出很多可疑分子，引发在高科技领域的排华潮流。

无罪：风会小一些，但不会停。情报系统脱钩是结构性的，不是一桩案子造成的，也不会因为一桩案子结束。

有罪：风会变成潮。美国这一方会大张旗鼓，北京那一方会对等抓人；老阎生前最担心的「排华潮流」，最可能在这时出现，首当其冲的是在美国高科技领域工作的华人。

英伟达股价：下坡比上坡快。

老阎的原理：

不论是房市还是股市，上涨就跟推车上坡，而下降往往也跟推车下坡一样，常见拽不住车而猛烈朝下滚。所以，泡沫吹得越大，爆了后能量释放得也越大。

无罪：股价会回稳，但回不到原来的高度。法庭能去掉国安风险，去不掉国会换手这个事实：英伟达仍然只剩白宫一扇门。

有罪：下坡会更陡。除了出口受限，还有罚款、合规和诉讼的风险，等于车已在下坡，又被人推了一把。但 AI 挡不住，基本盘还在，车不会翻。

本条不构成投资建议。

两类预测放在一起看：不受判决影响的两条，由结构决定；受判决影响的四条，判决真正决定的，是国家和资本谁说了算。照老阎的逻辑，一年之内，更可能说了算的是资本。

Surface and Roots

Capital, the state and two systems in contest — reading the Jensen Huang scandal

Editor’s note: this is an analytical piece, written in the editor’s voice but along Runtao Yan’s lines of reasoning. Everything in bold is Yan’s own writing, from his collected works; the rest is the editor’s analysis by his logic. Wherever this piece mentions the claim that the woman involved was a spy, it is an allegation by US officials and the press; it passes no judgment on anyone’s private conduct. Nothing here is investment advice.

I. The surface: two waves

First, line up the dates. Only when the order of events is clear does “timing” mean anything.

Date	Event
Oct 2025	The New York Times publishes its first investigation of Singapore’s Megaspeed; that month NVIDIA becomes a donor to the White House ballroom (later reported by the Washington Post at $25 million)
Late 2025	Per the NYT, Trump is briefed on intelligence about Alice Huang and takes no action (Trump denies this, saying he learned “not too long ago”)
Spring 2026	An intelligence memo overseen by Secretary of State Rubio circulates among a few senior officials
May 12–15, 2026	Trump visits China
Oct 29	The NYT reports US intelligence concluded Alice Huang gathered information for Beijing; NVIDIA says the claims were “thoroughly investigated and debunked last year”
Oct 30	NVIDIA loses its place as the world’s most valuable company
Nov 3	Midterm election. A Democratic landslide: as of Nov 10, about 254 House seats and 54 Senate seats, two races uncalled
Nov 4	Three Republican House committee chairs ask Huang to attend a closed-door briefing on Nov 18
Nov 6	The Financial Times reports the memo’s “romantic relationship” details
Nov 7	The NYT reports Commerce cleared about two dozen Chinese firms to buy H200 chips
Nov 8 (Sunday)	Huang writes to all staff admitting the affair; the board opens an independent review; Trump calls him “an honorable guy”
Nov 9	NVIDIA shares fall more than 4%; a shareholder suit is filed
Nov 10	Schumer ties the matter to the $25 million ballroom gift and calls for an investigation of NVIDIA’s influence on national security and export policy
Nov 12	The NYT reports Huang also had a relationship with a female vice president who reports to him; the WSJ reports the board is preparing to remove him. Huang resigns as president, CEO and director, effective immediately; CFO Colette Kress becomes interim CEO and independent director Harvey Jones the first chair of the board. Shares fall as much as 13% at the open and close down about 7%
Nov 17	NVIDIA third-quarter earnings
Nov 18	Closed-door briefing with three House committees
Jan 3, 2027	The new Congress convenes; Democrats take over committee chairs and subpoena power in both chambers

Line the dates up and you see the story came out in two waves: the “spy” wave five days before the election, the “affair” wave five days after.

The first wave was what Americans call an October surprise. Yan’s view of those was blunt:

This is how every American election goes. The so-called October surprise is all hype, not a real court case. Forgeries, hearsay, doctored material, every trick in the book — once the election is over, it all vanishes like smoke. Swing voters who have lived through countless elections ignore the whole routine. It is all a shot of adrenaline for each side’s own base.

This time bore him out: voters cared about their wallets, the Iran war and their unease about AI. Nobody was watching NVIDIA.

In the second wave, the first person to choose the timing was Huang himself. Two days after the FT printed the “romantic relationship,” he admitted it in a letter to staff — on a Sunday, with the market shut, ahead of both earnings and the closed-door congressional briefing. Yan’s method was to read things backward first:

Whatever you are reading, first test whether the reverse holds. Scrutinize it constantly. What has not passed verification is likelier to be a lie the better it sounds.

Read backward, “someone carefully calculated the moment of exposure” fails at least at this step. But the memo was written in the spring, and the intelligence reportedly reached the White House a year ago. The material had been there all along; when to bring it out was the choice.

At this point, “timing” is only the surface. The real question is: who is contending with whom? By Yan’s logic, three forces are contending over NVIDIA: capital, the state, and two political systems.

II. The first root: what capital wants

What capital wants, Yan put in one sentence:

The nature of capital is to pursue profit.

And where the profit in chips lies, he was just as plain:

Economic growth in modern society depends not on sellers but on courting buyers.

More than half of America’s chips are bought by China.

So NVIDIA wanting to sell to China is not a moral question; it is the nature of capital. But Yan also saw capital’s own dilemma:

What America’s conglomerates hate most is not China making money from trade, but that any new technology product, however advanced or complex, becomes dirt-cheap the moment China masters it, so that nobody else can stay in the game.

To sell, and yet to fear selling yourself a rival — that is the knot capital can never untie on China.

To sell, capital needs a friend at court. Writing about the 2008 crisis, Yan left a line quoted countless times since:

Why did Buffett, at the height of the crisis, put $5 billion into Goldman Sachs — GOLDMAN SUCKS — without hesitating? Because he knew the so-called $700 billion rescue was a capital injection for the financial sharks. And the Treasury Secretary and the Fed chairman both had ties to Goldman. The Treasury Secretary came straight out of Goldman. Why didn’t Buffett buy Lehman? Because Lehman had no one at court.

Does NVIDIA have someone at court? Yes — in the White House. A $25 million ballroom donation; Trump’s “I really believe him”; and four days after the election, Commerce clearing about two dozen Chinese firms to buy H200 chips. That door is open. Yan saw this coming:

Even if Trump is re-elected, he may not be able to hold out against the interests of America’s big corporations and conglomerates, and may sell chips to the Chinese Communist Party. During a campaign he can stand up to the conglomerates; after re-election, he may compromise with them.

On this layer, capital and the White House stand together.

III. The second root: what the state wants

What the state wants, Yan also put in one sentence. Explaining why Washington saved Citi and let Lehman fall, he laid down a rule that overrides everything else:

In America, partisan struggle always comes second; the national interest always comes first.

What is America’s national interest today? To lead China in AI, and to keep chips and intelligence out of a rival’s hands. Here capital and the state part ways: capital wants to sell to its biggest buyer; the state wants to guard against that buyer.

One point must be made plainly: the dividing line in this contest runs not between “business” and “politics,” but through politics itself. The White House stands with capital. Standing with the national interest are Congress and the intelligence community — and both parties at that. The first letter summoning Huang to explain himself came from three Republican committee chairs; the memo was overseen by a Republican administration’s Secretary of State. The Democrats taking over Congress have simply handed that side the subpoena.

Yan told a story that shows how much that dividing line weighs:

Why did Treasury Secretary Paulson go begging to Pelosi back then? Because he was rescuing some big banks while standing by as others like them went under. Without Pelosi’s endorsement, Paulson might well have faced a judicial investigation after leaving office. … With Pelosi’s backing, he no longer had to fear a reckoning after the harvest.

Paulson held the Treasury and still went to beg the opposition’s Speaker — so as not to fear a reckoning later. An open door at the White House covers today; whether Congress’s door is open decides tomorrow. Before November 3, NVIDIA had both doors open. After November 3, it has only the White House.

This time the state’s side has gained something more: public feeling. Voters used the ballot to register unease about data centers and AI, and behind that unease is the middle-class wallet. Yan described the division of labor between American markets and government coolly:

The function of the market is for capital to profit, driving overall economic growth and the creation and use of science and technology; the function of government is redistribution, balancing taxes and welfare to keep the gap between rich and poor in check. America “locks power in a cage of money,” so its market function is exercised to the full — while its government function has harmed the “middle class” (in reality, working people) who own little beyond their houses.

When the middle class feels AI’s gains go to capital and its costs to them, they use the ballot to make government act. Yan had once seen public feeling force the government’s hand. It was the 1980s, when he had just arrived in America, and the target was a Japan at its zenith — rich enough to buy New York’s Rockefeller Center:

The US government had no choice but to bow to public opinion and put a chokehold on Japan, and to diversify supply chains away from dependence on Japanese manufacturing — new high-tech invention was, after all, still in the hands of American science.

What followed was the 1985 Plaza Accord forcing up the yen, the 1986 US–Japan Semiconductor Agreement, and in 1987 Reagan’s 100% tariffs on Japanese goods. Yan recorded the outcome too:

After America put the chokehold on Japan, Japan stood still for thirty years.

But this time differs at the root: the one being asked to “slow down” is not a foreign rival but America’s own horse in the race against China. So by Yan’s rule, Congress can hold up data centers, hold hearings, write rules and make Jensen Huang a target — but it will not actually hand China the lead in AI. What the state wants is to rein capital in, not to break it.

IV. The third root: two systems in contest

That capital and the state can wrestle like this in America is itself a feature of democracy. Yan compressed the difference between the two systems into a pair of memorable puns:

America “locks power in a cage” — but nobody tells you what the cage is made of. In fact, the cage is money: call it the “money cage” (qianlong, which sounds like the emperor Qianlong — easy to remember). China is the opposite: money is forever held down by political power — that is, “money is locked in a cage,” and the cage is the regime: call it the “regime cage” (zhenglong). There are two idioms for these: “the golden age of Qianlong” and “Zhenglong photographs the tiger.” Lock power in a cage of money and you get a golden age (capitalism); the new idiom “Zhenglong photographs the tiger” means fakes run rampant.

(The pun on zhenglong comes from the 2007 “South China tiger photo” affair. A Shaanxi farmer, Zhou Zhenglong, claimed to have photographed a wild South China tiger; the provincial forestry department endorsed the photos at a press conference and rewarded him. Internet users soon found that the “tiger” matched a printed New Year poster, and in 2008 the provincial government confirmed it was a paper cut-out. Zhou was convicted of fraud, and thirteen officials were disciplined. “Zhenglong photographs the tiger” has since meant fakery with official backing — which is why Yan found it such an apt name for the regime cage: lock money inside the cage of power, and fakery has power behind it.)

In the money cage, NVIDIA can bargain with the state — donate, lobby, find an open door at the White House — and Congress can turn around and investigate it, restrain it, subpoena it. The contest is open and noisy, and ballots settle it.

In the regime cage there is no such contest. As Yan put it:

China “locks money in the cage of political power” (official-rank supremacy), so that capitalists are, without exception, white gloves for those who hold power.

The great feature of an autocratic system is that the state can disregard cost and ignore short-term returns.

Put the two systems at the same table, and the honey trap reads differently. If the allegation is true, it would be a state in the regime cage using its favorite method to get close to the most valuable man in the money cage. Yan wrote:

Set a honey trap, and few men withstand it. Very, very hard. No hero gets past a beautiful woman.

But he also wrote:

Some certainly are innocent, and some certainly are not.

The allegation is unproven, and Beijing has denied it. This piece makes one point only: whether or not this was a trap, the contest between the two systems over chips and intelligence will not stop on the answer.

On where that contest leads in the long run, Yan made a judgment Americans will not enjoy:

Technology, in the present age, cannot be completely blockaded.

In perhaps another ten years, the main chip-making power may well be China.

But he did not think the regime cage would have the last laugh. It may disregard cost and move fast, but its mortal weak point is the market:

When an economic crisis comes, the soundness of the “regime cage” will face a high-pressure test. The bamboo slats that make up the cage will turn on one another in a decisive fight, because the harmony between slat and slat is usually only on the surface. … So the stability of this “regime cage” system, once the market has been let in, depends on whether the market economy falls into crisis. The moment of crisis is the moment it comes crashing down.

In other words, the contest in the money cage is loud and in the open, and ends at the ballot box; the contest in the regime cage is hidden, invisible in ordinary times — and in a crisis it is life or death.

He also saw something curious: the two systems moving toward each other.

China is shifting from zhenglong (official-rank supremacy) toward qianlong (the market); while America is shifting from the “money cage” toward the “regime cage” (political correctness).

This midterm election is one signal of the money cage loosening — of politics pressing back down on capital. On one side, a regime cage that fears crisis; on the other, a money cage coming loose — and NVIDIA is caught between the two.

V. Why the three forces meet at NVIDIA

It is no accident the scandal landed on Jensen Huang. NVIDIA is three things at once:

capital’s biggest beneficiary of the China market — one of its largest buyers is China, and H200 licenses had just been approved;
the state’s most vital asset — America’s lead over China in AI rests on its chips;
the prime target in the contest between two systems — if the allegation is true, he is exactly the man the other side wanted to get close to.

When three forces meet in one man, any spark will catch.

Then who “carefully calculated” the fire? Yan did not believe in such a hand. He wrote an essay against the idea of a shadow government:

Why does Runtao Yan not accept that there is a visible shadow government? In a word: there is no need for one. A “visible shadow government” would be redundant, because an “invisible shadow government,” driven by interests, is quite enough to make the relevant people work together in tacit harmony.

This invisible force people call God can in fact be expressed in a single word — “money”: omnipotent; or in two — “interests”: omnipresent.

Nobody needs to sit in a back room moving pieces. Capital, the state and the two systems found their interests lining up of their own accord in the week after the election: Congress wanted subpoenas, hawks wanted tighter exports, voters wanted brakes on AI, competitors and short sellers wanted the share price down. No conferring, yet “tacit harmony.”

The losers are just as clear: Huang himself, NVIDIA’s shareholders, and the White House that stood on capital’s side.

VI. Why he resigned

The result of the three forces meeting came four days later. Again, first the dates:

November 8 (Sunday): Huang writes to admit an extramarital affair; the board announces an independent review; Trump calls him “an honorable guy.”
November 10 (Tuesday): the company says he will go on leading NVIDIA “without interruption.”
November 12 (Thursday): the New York Times reports that he also had a relationship with a female vice president who reports directly to him (neither responded); the Wall Street Journal reports that the board is preparing to remove him. That day NVIDIA files with the SEC: Huang has resigned as president, CEO and director, effective immediately; the filing gives no reason. The shares fall as much as 13% at the open.

After the first relationship came out, the board stood by him; on the day the second came out, he was gone. So the question is not “why did he get into trouble,” but why, with an affair either way, the first could be weathered and the second could not. By Yan’s logic, the answer has four layers.

First layer: private morals can be weathered; governance cannot.

The first relationship was with someone outside the company — under American norms, a private matter. The second was different: she reported directly to him. That turns private morals into corporate governance; once a boss and a subordinate are involved, promotions, pay and projects can no longer be accounted for cleanly. Yan, through the mouth of an American professor in one of his books, put it this way:

Whatever you do, don’t get into an office romance — that’s the end of you for life.

He also recorded a precedent from American politics:

The first was Gingrich, who swore that politicians must keep to morality, and was promptly found by the Democrats to have been having an affair with an aide — and slunk off and resigned.

(Note: when Gingrich stepped down as Speaker in 1998, the stated reason was the party’s midterm losses; the affair became public later. Yan’s words are quoted as he wrote them.)

Second layer: once can be a slip; twice is nature.

The other Speaker Yan wrote about fell on “more than one”:

He offered a million dollars to any woman who would come forward and say she had slept with the Speaker. Four women came forward at once … and the Speaker resigned immediately.

With one relationship, it can be called a lapse; once a second appears, people ask whether there is a third. The board could no longer vouch for him. Yan had long since reached his verdict on human nature:

If everyone could keep still, how would the name Liuxia Hui have come down to us at all?

Third layer: Clinton could ride it out; Huang could not.

Clinton’s scandal was far larger, yet he did not fall. Yan’s explanation:

A candidate’s first requirement is charisma. … Clinton didn’t need to do anything great; a nod and a wave of the hand, and he sat firm in the fishing boat. A scandal that big, and he still wasn’t impeached out of office.

A politician’s fate is in the hands of voters and legislators; charisma can offset scandal, and removal takes two-thirds of the Senate. A CEO’s fate is in the hands of the board, and the board answers to shareholders, not to charisma. As Yan said:

The nature of capital is to pursue profit.

A founder has no immunity before capital either. Yan cited the most famous case:

The competition for personal interest is extremely cruel and merciless. Jobs was actually fired from the company he himself founded.

Overnight Huang turned from the company’s greatest asset into its greatest risk; the shares fell 13% at the open. In the money cage, capital reacts to risk far faster than it remembers gratitude.

Fourth layer: the White House door was open, but it does not reach the boardroom.

The White House stood by him throughout. But whether he stayed was not the White House’s call; it was the board’s. And what the board had to fear was the Congress that changed hands after the election. Yan explained why Paulson went to beg the opposition’s Speaker:

With Pelosi’s backing, he no longer had to fear a reckoning after the harvest.

Democrats take over subpoena power on January 3, and members of Congress have already asked for NVIDIA’s China-export documents to be preserved. Had the board kept him after the second relationship came out, then at the hearings it would be the board itself asked what it knew and what it did. That also explains why he gave up his board seat as well: the cleaner the cut, the further the company stands from the reckoning.

No hidden hand, only interests.

No one needed to stage a coup, and no one needed to “carefully calculate.” Shareholders wanted to stop the losses, the board wanted to protect itself, Congress wanted its subpoenas, hawks wanted tighter exports — and on the same day, their interests lined up. The “invisible shadow government” of Section V worked in tacit harmony once again on November 12.

So he did not resign because of an affair. The affair is the surface; the root is that he turned from an asset of capital into a risk to capital — and did so just as Congress changed hands and the chip contest with China was at its tightest. Capital could carry him through the first relationship; once the second involved a subordinate, capital would carry him no further.

VII. The year ahead, by Yan’s logic

The predictions below use only two things: the laws Yan argued for again and again in his lifetime, and the facts confirmed as of November 12. No one else’s commentary or forecasts are used.

The biggest variable is the Justice Department’s criminal inquiry into Megaspeed. If a court finds that chips or information went where they should not have, in violation of export controls, that is “guilty”; otherwise, “not guilty.” Judges are independent, but whether to prosecute, and how, is decided by the Justice Department — part of the executive branch, answerable to Trump. By the reasoning of Section II, the White House stands with capital; and Yan said long ago that after re-election Trump “may compromise with the conglomerates.” So by Yan’s logic, within the year “no charges, or lenient ones” is more likely than “guilty.” Only Congress can change that: from January 3 the Democrats hold subpoena power, and if hearings force out new evidence, the Justice Department will have to act.

A criminal case often takes more than a year from indictment to verdict, so “guilty” and “not guilty” below can also be read as “charges brought and sustained” and “no charges, or charges that fail.” On that basis the predictions fall into two groups: those the verdict will not affect, and those it will; for the latter, the likelier “not guilty” outcome is listed first.

Not affected by the verdict
Data centers will be held up; AI will not be stopped.

Yan’s principle:

The US government had no choice but to bow to public opinion.

No one can stop AI from eventually replacing biological humans; it is only a matter of sooner or later.

Inference: voters have spoken at the ballot box, and state and local approvals for data centers will slow markedly; but federal legislation to “slow AI,” even if the new Congress passes it, would be vetoed by Trump — and neither chamber has the two-thirds majority needed to override him. This one is driven by middle-class wallets, not by the court. Even with a guilty verdict, AI would become a national-security race, making a federal slowdown harder still. What slows is the speed, not the direction.

China will speed up making its own chips.

Yan’s principle:

After Trump’s ZTE affair, China’s state-driven system will rapidly pour huge investment into chips; in less than ten years China will catch up in chips. This is Trump’s most foolish move. It amounts to handing the CCP a direction and a motive for development.

Technology, in the present age, cannot be completely blockaded.

That the CCP cannot make 5-nanometer chips in the short term does not mean it still cannot in a few years. Because the age can no longer stop scientists from moving.

Inference: Beijing made up its mind after the ZTE affair. A guilty verdict shuts the door tighter — once again “handing the CCP a direction and a motive”; a not-guilty verdict does not change the direction either.

Affected by the verdict
The affair: fading from the news, or returning in a new role.

Yan’s principle:

The so-called October surprise is all hype, not a real court case. … Once the election is over, it all vanishes like smoke.

Some certainly are innocent, and some certainly are not.

Not guilty: the private matter fades quickly. Once the national-security layer collapses, only private morals remain, and the resignation has already settled those. The spy allegation itself will vanish like an October surprise.

Guilty: the private matter returns to the news in a new role — no longer a scandal but case background. A court ruling establishes only where the chips went; it cannot prove a honey trap. But people will ask whether the relationship and the channel were connected, and Congress will put the private matter and the exports side by side.

Chip exports to China: who has the last word, the state or capital.

Yan’s principle:

In America, partisan struggle always comes second; the national interest always comes first.

Obama could not do it, because if America didn’t sell chips and other high technology to China, America’s big companies would lose out.

The nature of capital is to pursue profit.

Not guilty: tighter first, then looser — and looser sooner. With the court removing the national-security shield, capital regains the upper hand, and a White House already on capital’s side will open the door.

Guilty: the door shuts tight and will be hard to reopen. With a judgment proving that chips leaked, the national interest overrides capital; both parties line up behind it, and the White House cannot hold out.

The hunt for technology spies: a wind, or a tide.

Yan’s principle:

The chip war amounts to the decoupling of the intelligence systems. America will keep arresting Chinese technology spies in America, and the CCP will arrest intelligence agents in other fields.

Retaliation has always been mutual.

As for the “spy-catching” campaign, I hope it does not turn up a lot of suspects and set off an anti-Chinese tide in high tech.

Not guilty: the wind eases but does not stop. The decoupling of intelligence systems is structural; one case did not cause it, and one case will not end it.

Guilty: the wind becomes a tide. America will act loudly and Beijing will answer in kind; the “anti-Chinese tide” Yan feared most is likeliest to come now, and Chinese engineers in American high tech will bear the brunt.

NVIDIA’s stock: faster downhill than uphill.

Yan’s principle:

Whether in housing or in stocks, going up is like pushing a cart uphill, and going down is often like a cart rolling downhill — you frequently cannot hold it back and it plunges. So the bigger the bubble is blown, the more energy is released when it bursts.

Not guilty: the stock steadies, but does not return to its old height. The court can remove the national-security risk; it cannot undo the change in Congress. NVIDIA still has only the White House door.

Guilty: a steeper descent. On top of export limits come fines, compliance and litigation — a cart already rolling downhill, given another push. But AI cannot be stopped; the core business remains, and the cart will not overturn.

This is not investment advice.

Taken together: the two predictions unaffected by the verdict are set by structure; for the four it does affect, what the verdict really decides is who has the last word — the state or capital. By Yan’s logic, within the year it is more likely to be capital.