大田原正幸｜医療AI/SIガバナンス｜Cursorvers

全网都在关心老黄到底睡了几个，但真正让华尔街后背发凉的，根本不是裤裆里那点事，而是这起黑天鹅撕开的三个残酷真相：$NVDA 第一个真相：为什么董事会敢在96小时内，把硅谷唯一的“造物主”扫地出门？ 很多人想不通：老黄一手创办英伟达33年，把一家做显卡的小作坊干成万亿帝国，怎么说踢就踢？ 去翻一下英伟达的公司章程就懂了——老黄死在了自己的“现代企业美德”上。 英伟达没有AB股（双重股权结构），黄仁勋手里只有个位数百分比的股份和投票权。按章程，董事会11名董事，只要6个人举手，就能直接罢免CEO，甚至不需要任何理由。 对比一下隔壁Meta：扎克伯格死死攥着超级投票权，同样的丑闻如果发生在小扎身上，董事会连他的皮毛都动不了。 创始人不设AB股，顺风顺水时叫“治理透明的美德”，一旦出事，就是被人一剑封喉的“命门”。 11月8日董事会还在喊“完全信任”，4天后直接6票过半斩草除根，连董事席位都没给他留。 第二个真相：老黄出局，最先断粮的可能是国内的AI算力。 很多人还在猜谁接班（顺便辟个谣：昨天疯传的Jay Puri今年6月就退休了，连8-K文件都没有的假消息别再传了），但对中文圈来说，真正致命的暗雷在华盛顿。 过去几年，顶着巨大的政治压力在华盛顿来回游说、替对华芯片销售撕开裂缝的，主要就是黄仁勋本人。 现在他倒了，国会正在借题发挥，推动在今年的《国防授权法案》（NDAA）里塞进一批极其严苛的先进AI芯片出口限制条款。 老黄留下的这个游说真空，临时顶上的CFO克雷斯补不上，未来的新CEO更不敢碰。我的个人判断：在国安调查的阴影下，英伟达对华特供芯片的窗口，大概率要彻底焊死了。 第三个真相：英伟达正在滑向“2018年的英特尔时刻”。 过去十年，美股有两家巨头因为CEO职场私生活换帅，结局天差地别： ▪️ 2019年麦当劳赶走Easterbrook，股价照样涨——因为卖汉堡不需要技术信仰，谁当CEO都能转； ▪️ 2018年英特尔赶走Krzanich，让CFO临时接盘，结果技术路线一路崩盘，直接错失整个移动与AI时代。 不幸的是，英伟达是比英特尔更极端的“技术独裁”公司。 5天后（11月17日）就是Q3财报夜，这是没有黄仁勋的第一个夜晚。纯财务出身的克雷斯要独自面对华尔街的拷问： 董事会什么时候能选出懂技术的掌舵人？ 失去灵魂人物后，下季度指引会不会下调？ 微软、Meta这些超级大客户有没有动摇？ 有人问老黄会不会像1997年的乔布斯、或者去年的奥特曼一样杀回来？ 别做梦了。 哪怕路透社说绝大多数员工盼他留任，哪怕他还是第四大股东，但这次背着国家安全相关的调查，性质已经彻底变了。 皮衣封存，那个一人决定全球算力走向的时代，真的结束了。 11月17日的财报夜，大家系好安全带。

12%下落

⚠️⚠️ If Nvidia fires Jensen Huang, America loses the AI race. And China wins it. ⚠️⚠️ I'm choosing my words carefully. Nvidia's board is deciding whether to remove its founder. Wall Street wiped out hundreds of billions of dollars in a single evening, and everyone is pretending you can replace a CEO like you change a light bulb. You can't replace Jensen Huang. This isn't about morality. It's about mechanics. Let's start from the beginning. Three engineers meet in a Denny's in San Jose. Jensen is 30. As a teenager, he washed dishes in that same kind of diner. The plan: chips for 3D graphics, in a market that doesn't exist yet. The first product flops. The Sega deal falls apart. Nvidia is weeks away from dying. Jensen goes to Sega, admits his chip won't work, and asks them for money anyway. Sega invests $5 million. That money saves the company. For years, he opens his meetings with the same line: we are 30 days from going out of business. Then comes the GeForce, the first GPU. Then CUDA. He decides to make his chips programmable for any kind of computation, not just video games. Billions burned for a decade on a bet nobody understands. Wall Street hates it. He holds. Then AlexNet, the neural network that kicks off the deep learning revolution, is trained on two Nvidia graphics cards. Jensen understands what it means before anyone else. He pivots the entire company to AI. He personally hand-delivers the first DGX-1 supercomputer to a small research lab called OpenAI. Today Nvidia is worth around $4 trillion. Almost all the AI on the planet runs on its chips. That's not luck. That's 33 years of tryhard. And here's what people don't understand about founders. A founder CEO isn't a job title. It's a nervous system. Jensen has around sixty direct reports. He knows every layer of the stack: transistor physics, chip architecture, software, data centers, TSMC, customers, governments. He sees the next bottleneck before it shows up. That kind of awareness isn't taught in an MBA. It doesn't transfer in a three-month handover. It's built over three decades of brutal decisions, made against everyone's advice. Right now, no one else on Earth has that level of awareness of the system. Even Dan Ives, one of the most followed analysts on Wall Street, says it: Jensen is not replaceable, and no one on the bench can step in seamlessly. The name floated for interim CEO is the CFO. An excellent CFO. But the question isn't who keeps the books. It's who decides the next architecture. We've seen this movie before. Twice. Steve Jobs. Apple's board pushes him out. Serious managers take over. Twelve years later, Apple is months from bankruptcy. Jobs comes back. iMac, iPod, iPhone. Apple becomes the most valuable company in the world. Travis Kalanick. Investors push him out. Uber becomes calmer, more profitable. And walks away from the most important bet of its generation: it sells its self-driving unit. Today Waymo runs robotaxis across American cities, and Uber signs partnerships with the people it could have beaten. A company that manages. No longer a company that conquers. The pattern is always the same. A board optimizes for reputational risk. A founder optimizes for the mission. Replace the second with the first, and the company doesn't die right away. It just stops making the bets that matter. And the further we go into AI, the truer this gets. When everything changes every six months, when a single architecture decision commits $100 billion, when thousands of agents execute what a handful of humans decide, value concentrates in one thing: the judgment of whoever holds the wheel. AI doesn't dilute the founder's role. It multiplies it. Should national security concerns be taken seriously? Of course. An independent investigation is normal. But the board itself said it had no reason to believe Nvidia's secrets or national security were compromised. And we're talking about removing the most strategic person in American tech before the findings are even in. So let's ask one question. If a Chinese intelligence operation really was targeting Nvidia, what would its best possible outcome be? Not a stolen secret. Jensen Huang's exit. Beijing doesn't need to beat Nvidia. The board can do it for them.

実を言うと、私もこれには心を揺さぶられました。 たしか浪人時代、模試で読んだ文章だったと思います。急いで本を探したの覚えています。 当時、「自己犠牲」という言葉の意味を何度も 反芻していたことを覚えています。 同じように、志賀直哉の『孤児』にも模試で出会い、 感動しました。 もっとも、模試の最中に文章に感動しているようだから、浪人したのかもしれません（笑）。