원문 보기
Here's a question. Why would the head of Citadel call AI a bubble in the very same week that his firm is using it to reduce research projects from eight weeks to three hours? You see, there's a headline about Ken Griffin warning of an AI collapse, and you probably think he means the technology is fake.
But what he really means is the opposite, and he's blatantly saying that the technology works so well that it's about to bankrupt those who create it. What exactly am I saying ? Look, if you're new here and wondering what a cartoon cat on YouTube could possibly offer, well, in short, today, to understand whether the AI business can still survive, we'll be looking at productivity data to see who's actually making money with AI.
And if you 're still skeptical, the only conclusion you need now is that the real winners in the stock market are no longer the companies that sell AI, but the boring companies that use it discreetly. Now, on your screen, let's put some context first. Those are the combined capital expenditures , which only mean money spent on physical things like servers and buildings, of Microsoft, Alphabet, Amazon, and Meta.
Last year , they spent nearly $410 billion . By July, their average forecast for this year jumped 79% to $732.5 billion . Do you see that dashed red line at $700 billion ? Goldman Sachs calculated that $ 700 billion is the exact level needed to match the peak of the telecom bubble of the late 1990s.
We surpassed that mark in four months. Morgan Stanley believes that spending on data centers will reach $850 billion this year and $1.3 trillion next year. Now, look at the small bar on the right. That is the total projected revenue for specialized AI providers at this time. It is below 35 billion dollars.
Now, if they have shares in Amazon, Google, or any broad index fund, they probably assume they're spending more because they're building more data centers for paying customers. Drumroll, please. Ta-da! They're not doing it. And worse still, the money they 're spending isn't even coming from their own bank accounts.
That change masks a massive transfer of profits that only appears when you cross-reference corporate debt statements with federal labor statistics. Let's break down that increase in spending first. When a company increases its capital expenditure budget, Wall Street is usually pleased because it signals booming demand, and I'm not saying that's wrong, except that this time a large part of that new spending isn't even buying an extra server.
Look at the red segment that fits at the end of the Amazon bar. Amazon increased its 2026 spending plan from approximately $200 billion to $220 billion. Management admitted that the extra $20 billion was mainly due to computer memory prices rising without any change to the construction plan. Below that, Microsoft estimates that the increase in component prices alone added about $25 billion to its account .
When your expenses rise only because suppliers are pushing up memory prices, your future returns aren't growing. Their costs are simply inflating. And how are they paying these inflated bills? Well, for the past decade, Big Tech has paid for everything with free cash flow, which is the purest measure of financial health.
You may have already heard about this. It's as if, as long as our "MAG 7" have a solid balance, the AI bubble will remain intact. And I don't oppose that thesis, but right now we must admit that things have started to change. Look at the first row of this table. Alphabet just reported a negative free cash flow of $5.9 billion.
This is the first time Google has burned cash since its initial public offering in 2004. To continue building, Alphabet's debt jumped from $16 billion to nearly $100 billion, and they announced an $ 80 billion stock sale. Look at the second row for Meta. They earned $31.9 billion in operating cash, spent almost all of it, and ended up with only $784 million of free cash flow.
They took on $24.9 billion in new debt and stopped buying back their own shares altogether. In the third row, Amazon's free cash flow for the last 12 months is now negative at $7.6 billion . Only Microsoft, in the end, remains in positive territory with $19.6 billion . And let's be clear, booms rarely end because the technology stops working.
They end when the money to finance them becomes expensive. When cash-rich monopolies start issuing debt and diluting shareholders to buy overpriced memory chips, the cycle has entered its late, finance-dependent phase . Prior to this, I also spent time researching stocks for one of the largest financial channels.
And the pattern that appeared at each historical peak of the sector was that of cash flow kings suddenly becoming bond market beggars. And there's an accounting trap waiting in 2027. When a company buys a server, it doesn't immediately deduct the full cost from profits. Distribute that cost over several years using an accounting rule called depreciation.
If 60% of the $730 billion spent this year goes to servers that last 5 to 6 years, look where those red bars fall on the timeline. That single year of spending adds approximately $80 billion a year in depreciation expenses starting next year. Some market researchers also point out that most of this cost will not affect reported profits until next year.
Today's technology profits seem artificially high because the bill has n't yet reached the income statement. If I spilled a glass of water off a table and told you it wouldn't count as a disaster until the next fiscal year, you'd still reach for a towel. Now, if you own shares in Nvidia, Microsoft, or Amazon, you might disagree on this point.
You could point out that cloud growth is exploding . And you would be right. AWS grew 37 % year-over-year, its fastest pace in 18 quarters. Google Cloud grew by 82% and Microsoft's AI business grew by 123%. Microsoft even has an $80 billion Azure order backlog locked up because they literally can't get enough electricity to power the servers.
And this is the part that sometimes worries me. You see, on the one hand, the demand is completely real. Nobody is falsifying those cloud figures. But on the other hand, what makes me cautious is realizing that " demand exceeds supply" is exactly what every telecom executive was saying in 1999 . Internet traffic was actually doubling every few months back then .
The fatal error was not predicting that people would use the Internet. The mistake was assuming that every dollar spent on laying fiber optic cable would yield a high return for the company digging the trench. And this brings us back to the headlines. Ken Griffin hit the nail on the head about why this happens.
He noted that while AI is real, companies have a very uneven ability to extract value from it , and corporate finance teams are now auditing their Anthropic and OpenAI software invoices line by line. But it's also natural to wonder, if it's so obvious, why do tech CEOs keep investing money? Are they idiots?
Why don't we just replace them with cats? Because their incentives simply don't align with yours. Amazon's management made it clear that they are not going to be conservative. If a technology chief spends too much, shareholders lose some return on capital. If they spend too little, they risk losing their monopoly platform for the next decade.
Their bet is about corporate survival. Your bet is about the return on the stock price you pay today. In 2000, Cisco dominated networking equipment and continued to grow its business. However , investors who bought at the peak waited decades just to recoup their investment. Even worse, Griffin warned that AI is bulldozing the competitive pits of older software companies.
Now, if you're a regular viewer, you might be frowning at your screen wondering, is the cat finally going crazy? Is he finally accepting Michael Burry's whole AI bubble thesis? Please bear with me and let me ask, if the developers are burning through cash and borrowing money, where is the real financial reward hiding?
Which brings us to the second half of Ken Griffin's quote , the part that everyone ignored. He said that companies on the other side of the bubble become market leaders for decades, adding that he wished he had kept his shares after the dot-com crash. Remember, Amazon didn't install the fiber optic cables in the 1990s.
Amazon used the cheap bandwidth paid for by telecom companies that went bankrupt. Even better, we don't have to guess if that's happening again. It's right there in the BLS second quarter report . Look first at the left side of this graph. Despite more than $1 trillion in cumulative technology spending, overall US productivity grew by only 1.4% annualized in the second quarter.
And total factor productivity grew by 0.8% last year. Since the end of 2019, productivity has averaged 2.1% per year. Notice how it remains completely flat on the gray line representing the historical average since 1947. It's nowhere near the golden 3% line of the late- 90s boom. And part of today's gross domestic product is just the construction of the data centers themselves, which counts as economic output regardless of whether those servers generate a penny or not.
Now, look at the right side of the screen. This is the most important graphic in this entire video. While overall productivity seems totally average, look at the red curve falling off a cliff at the bottom. The portion of economic output allocated to workers fell to 52.8%. That's the lowest labor force participation rate on record since the government began tracking it in 1947.
Now, look at the green line shooting straight up above it. Unit profits in non-financial corporations jumped 43% year-on-year in the second quarter and 17.8% year-on-year. Now, take a moment to think about what the crossing of those two lines means for your money. Within Citadel, AI turns an 8-week analyst job into a 3-hour test.
Across the economy, ordinary non-financial businesses are using automation to limit hiring and expand their profit margins at an annualized rate of 43%. Yes, frankly speaking, wealth isn't going to the workers, and it's not staying with the cloud providers who are burning cash on memory chips either .
It's directly impacting the profit margins of ordinary companies that use the tools. Excuse me while I choke on how much debt Google took on just so banks and retailers could increase their margins by 43%. Meanwhile, cracks are already beginning to appear on the margins of the infrastructure sector .
Incidentally, in recent days, investors have begun to withdraw from smaller "neo-cloud" data center operators. If you 've seen my video about CoreWeave, you'll already know. An Australian data center operator, NextDC, has just cut its outlook after a pause in power connections. Barclays warned of price pressure for electrical equipment manufacturers.
In addition, 71% of American adults now oppose the construction of an AI data center in their neighborhood. And the window for initial public offerings is wide open. Chipmaker Cerebras opened approximately 89% above its offering price. And Astera Labs' newly listed shares fell . Now, Anthropic is preparing one of the biggest IPOs in history, while reportedly planning $750 billion in computing spending.
When the most highly touted private experts rush to sell shares to the public , you're usually closer to the peak of the spending cycle than to the bottom. So what do you actually do with your wallet? It's easy. Again, cut back or avoid bets on debt-financed "neo-cloud" data centers, unprofitable small businesses, leveraged ETFs, and don't chase Anthropic's IPO on day one.
On the other hand, you could support the infrastructure giants that still pay their bills with their own cash flow. Microsoft clearly stands out here, but review your overall exposure. If your index funds plus your individual tech stocks put more than 35 to 40% of your portfolio into AI hardware, you are not diversified.
You're making a huge bet on the price of servers. And finally , take advantage of market downturns to buy non-financial companies— yes, non-financial—that are actually using automation to reduce costs and expand their unit profits. He continues his search for a 43 % profit. Now, I understand that all this isn't easy to digest, but if you value facts over advertising, even when they come from a cat that always has a skeptical tone, hit the "like " button and subscribe to keep him well fed.
Now, tell me in the comments where I'm wrong with my calculations. Specifically, what single figure in the upcoming third-quarter cash flow statements would convince you to buy data center builders right now ? See you in the next video.
Comments 0