저자: Rule#1Investing채널진행자 · 외국인
요약 보기
AI 칩이 창고에 쌓여 있고 설치 지연과 세대교체로 감가상각 부담이 커져 빅테크 이익이 과대계상됐다고 주장. 잉여현금흐름 급감, 토큰 소비 과열(테스트성 수요)이 일회성일 수 있다고 경고. 이 기사 같은 AI 거품 붕괴를 예상하며 안전마진과 능력범위 내 투자를 강조. 세미나 홍보 포함.
다듬은 글
Well, let me start this video by giving credit to Michael Bury , who is a true genius at exposing the truth about artificial intelligence. And everything I'm going to tell you, one way or another, came from Michael. So right now, three out of every four chips created, chips that cost more than a car, are sitting in warehouses.
They've been delivered , they've been paid for, but they're not connected and are just waiting for buildings and energy supplies that do n't even exist yet. And here's the main trick: due to one accounting rule, as long as these chips are in the box, they don't depreciate. The fastest-obsolete asset in the history of capitalism ages in storage, and AI accountants say, "No, it hasn't depreciated at all ." Friends, today I want to talk about the AI bubble, but not the way CNBC talks about it.
Because I invested throughout the 1990s, and I remember exactly how it felt. Everyone said the internet would change the world, and the crazy thing is, they were right. He did make a difference, but the Nasdaq index still fell by almost 80%. So being right about the value of technology and being right about the value of investment are two very different things.
So the question is never , is AI real ? Of course, it's real . The question is, what is the value of what we get by buying an AI company at these prices? Remember: price is what you pay, and value is what you get . And when I delve into this question: what is the value here? I find three things that every index investor should understand, and that means almost everyone.
Number one: packaged chips, what's in the warehouse. Nvidia, and now other competitors, create a new, much faster chip almost every year. So, the latest generation of Nvidia chips, which were purchased for billions of dollars. They were shipped, but only about 5% are actually working now. A huge percentage of the rest is still sitting in boxes, waiting for giant data center projects that still have to resolve issues of financing , policy, energy contracts, and construction.
One flagship data center project just announced that it won't even be built until 2029 because a gas pipeline couldn't be agreed upon . Meanwhile, Nvidia operates on a yearly product production cycle. A new, faster generation is already in production. So think about it, friends. Chips are bought today, but it is impossible to install them immediately, and by the time they are connected, they will already be a generation behind, or maybe two or three.
So, why is this a problem? Well, Nvidia's new Blackwell chips have five times the performance of the previous generation Hopper, with 10 times the power efficiency. New developments are not just a little better. This is a whole other level, which is why Nvidia CEO Jensen Huang said that when the new chips come out, no one will need the old ones, even for free.
I mean, seriously. Who among us would buy last year's car if, for the same price, this year's car was five times faster and consumed ten times less fuel, right? He may have exaggerated a bit for sales effect, but the point is clear. Old equipment quickly becomes obsolete. You've seen AI companies try to outperform each other with every generation.
Can you imagine trying to compete with old chips? Unless you are China. So the bottom line is that these old chips will become obsolete before they even get into the data center. This means that they will have to be written off, and we are talking about hundreds of billions of dollars, and this will seriously hit profits.
Not a little, but significantly, and it could quickly collapse the stock price. Why? Ah, because of problem number two. Problem number two is inflated profits. If you need to keep up with your competitors, and they are doing significantly more on this year's fast chips than you are on last year's slow ones, you have to replace those old chips or you are out of the race.
Hmm. Now, if you know this and still tell the accountants to stretch the write-offs over 6 years, then today's reported profits are overstated. Right now, in the largest companies in your index fund. Do you want evidence? Watch your cash. Free cash flow is the money left over after buying new chips.
So, hyperscalers' free cash flow is now plummeting to zero , while reported profits look great because they don't subtract the cost of old chips. If they had kept their books correctly, they would have subtracted the value of the old chips, and their profits would have gone straight to zero along with the cash—and goodbye, AI stock bubble.
Well, Mr. Buffett taught us to look at the owner's return, not just the profits —the owner's return, which is the cash that the owner can actually withdraw. When profits are glorious and cash is zero—be careful. And now number three, and this is what really struck me. The very demand, the optimism about AI stocks, is based on one idea.
We don't care about old chips, we will use them even if they are slow, because the demand for AI is insatiable. Well, that seems to be the case, does n't it? Users are burning available SHI tokens at a record pace, and all types of chips are being used. But look why they burn them. There's a phenomenon happening in corporate America right now that tech insiders call " tokenmaxing." Companies are demanding that employees make the most of AI.
Some companies now even have internal leaderboards for those who consume the most tokens. So, keep doing it, without limits. Why would companies even do this? Because this is the path to your own internal product that will become good enough to no longer need the AI provider you maxed out tokens on, and that provider's growth will disappear, as will its stock price.
Corporate America is currently in a phase of massive, frenzied one-time testing, which is creating a one-time huge demand that won't last long. Here is a recent sequence of events. Company A made the most of a popular coding tool purchased from Company B, while simultaneously training its own, and when it became good enough, Company A revoked all of Company B's licenses and switched everyone to its product.
One of the world's largest technology companies just went through this process in six months, and every CTO in America was watching and taking notes. So, we are not dealing with fake technology, we know that. AI is a revolutionary technology whose most expensive phase of adaptation is being assessed as if it were a constant demand.
Testing is not production, it is not permanent. The "gold rush" of training launches is not permanent. When this hands-on phase is over, my friends, activity will shrink and stock prices will shrink with it, and we've seen this before with Gore Lucent and Nortel, and those companies have gone to zero.
These AI companies won't go to zero. They are, of course, significantly overvalued due to a certain ignorance of investors, similar to what happened to investors in Bernie Madoff's pyramid scheme. I had a friend who lost $60 million to Bernie before everyone found out about the falsification of reports ; Her family was very happy, thinking they had 60 million, which they did n't actually have.
And Bernie certainly lived well on stolen money. So for a while it seemed like everyone involved in the scam got richer. And then reality finally set in. Index investors, you are like Bernie Madoff investors, happy with the way things are going. You have some money, but you are completely unaware of what is inevitably going to happen.
If you invest in market indices now, you think you're doing great. But when accounting reality finally catches up with you, you may not be so happy. So what should we do? Well, find out how much these businesses are really worth, based on the owner's profits, real money, not accounting reports. Require a margin of safety.
And if the price doesn't match—wait. Don't jump out of your circle of competence just because AI is a shiny toy. Study what you understand. Make a watch list of what you understand, because when AI overvalues, a big drop in the entire market will be inevitable, and you'll want to be ready to take advantage of it rather than panic.
The fortunes made in 2000 and 2008 were not made by the same people who predicted these crashes. They were made by those who were ready when Mr. Market finally panicked and put a great business up for sale at half price. So when this bubble bursts, and it will, there's no need to be afraid of it. This is exactly the moment we are preparing for.
Listen, if you want to learn exactly this: how to value a business like Buffett does, how to buy with a margin of safety, we're hosting a three-day live "Rule One" seminar in Atlanta in September. Over 30,000 investors have already completed this training. I'd be happy to see you there. The link is in the description. Thank you for listening. And now—to the point.
Comments 0