원문 보기
Welcome. You've brought us a fascinating stack of sources examining well the biggest financial question of the decade. Is the artificial intelligence boom fundamentally a bubble? It's a question on everyone's mind, right? And we've all seen the news reports. There are some serious fears that some of America's biggest, most established corporations are aggressively propping up a financial scenario that just feels ripe for a massive crash.
And our mission for this deep dive is to try and move beyond the hype, you know, around the technology itself, which is clearly transformative, and apply the hard lessons from financial history. Okay, we're focusing strictly on investor behavior, the uh staggering amounts of capital being deployed and trying to figure out precisely where the financial risk is accumulating.
As one economist noted in your material, history rarely repeats itself exactly, but it often rhymes. Okay, let's unpack this with a critical rhyme from history, then. Our source quotes Derek Thompson suggesting AI could be the railroad of the 21st century. And that powerful historical parallel, it tells us that revolutionary technology captures people's imaginations in predictable almost intoxicating ways.
That analogy is so crucial because it immediately shifts the discussion away from algorithms and you know into investor psychology. Bubbles fundamentally are not caused by technology, right? It's about the people. It's about the people. Mhm. They are caused by the application of excessive optimism or what Alan Greenspan famously dubbed irrational exuberance to these new developments.
When something genuinely revolutionary appears, there's no historical data to restrain the imagination. So the future just feels limitless. Exactly. It feels limitless and that justifies valuations far far beyond any predictable earning power. This distinction between bubble types seems key to really understanding AI.
So how do our sources use this framework? analyze what's going on in investors heads, right? Because not all bubbles are created equal. Not at all. We need to make a critical distinction between two classifications of bubbles as outlined by financial analysts Bern Hobart and Tobias Huber. First, you have the mean reversion bubbles.
Okay, what are those? Think of these as financial fads. The South Sea Company, the subprime mortgage crisis. They offered the promise of returns without risk, but they didn't provide any real underlying progress for mankind. When these burst, the system just kind of reverts to where it was before. They are pure financial folly.
And the other type, on the other hand, you have inflection bubbles. Now, these are tethered to genuine technological progress. Think railroads, electricity, or the early internet. So, things that actually change the world, things that permanently change the world. Investors decide the future will be meaningfully different.
And crucially, after they burst, the world does not revert. the infrastructure remains. It sounds like you're saying this historical pain, it has a hidden societal purpose. Wait, does that mean our sources are arguing there's actually a good kind of bubble? It sounds crazy, doesn't it? That feels completely counterintuitive.
I mean, we usually associate a bubble with widespread financial agony. And it is. But from a societal progress standpoint, the answer is yes. We look to the work of Carlo Perez, who argues these inflection bubbles are not regrettable events, but are actually necessary. Necessary. The sheer speculative mania enables what she terms the installation phase.
Think about it. Building out the transcontinental rail lines or laying fiber optic cable across the world. Okay, these were investments that were necessary for the next age of economic activity. Mhm. But they were not necessarily wise financial decisions for the people who first funded them. bubbles create these massive collective opportunities to deploy capital funding highly risky exploratory projects through what she calls collective delusion and collective vision.
So the mania provides the mechanism for progress if prudent rational investors govern the process building out all the needed data centers and chips and infrastructure that might take decades. It would take forever but the hysteria of a bubble compresses that necessary investment into a very short very intense period.
So the acceleration is the upside. Exactly. The conclusion is that inflection bubbles accelerate technological progress but and this is a huge but they absolutely destroy wealth in the process. They carpet bomb a new area with capital incinerating a lot of money on startups and infrastructure that fail.
So the key for you the investor is simply not to be one of the people whose wealth gets incinerated. You want the benefit of the acceleration without the short-term financial folly. Okay, let's pivot to the AI landscape today and see where we are in that cycle. We know the technology is potentially one of the biggest economic forces of all time, but the concentration of capital is it's startling.
The material we analyze notes that AI stocks account for roughly 80% of S&P 500 profits and 90% of capital expenditure growth. It's an incredible concentration. And the poster child, of course, is Nvidia. From its IPO in 1999 with a market value of just $626 million, it recently became, however briefly, a multi-trillion dollar company.
To put that in perspective for you, that's an appreciation of roughly 40% every single year for over 26 years. No wonder imaginations have been fired. I mean, these are returns that rewrite the rules of compounding. But that excitement, it just obscures the sheer uncertainty of it all. I mean given that pace of change, who can possibly know who the eventual winner will be?
That is the single biggest operational problem for investors. It recalls Warren Buffett's famous auto analogy. Ah, the automobile was revolutionary. It changed everything. But of the 2,000 or so car companies that started up, only three ultimately survived the consolidation. So, will today's leading AI chip makers and model developers prevail, or will they just give way to upstarts?
We're already seeing evidence of this fluid environment. A recent MIT and hugging face study showed that Chinese-made open models recently surpassed US developers in the total share of downloads for the first time. The current leaders are operating under constant threat of displacement. And this overwhelming uncertainty, it fuels what our sources describe as lottery ticket thinking.
The potential payoff is just so massive that it justifies huge investment in a very low probability outcome. Right? Take the example of the AI startup Etched. Its CEO raised $120 million by betting that if AI transformers stick around, they'll be the biggest company of all time, justifying that investment for a potential $1 trillion valuation.
When the potential gain is that high, you only need a fraction of a percent chance of success to compel people to jump in. It's just human nature to chase that astronomical return. And that leads to the crucial profit question. I mean, will AI be a monopoly or a free-for-all commodity or something in between?
It's a huge question because right now it's widely noted that services like Chat, GPT, and Gemini are actually losing money on every query they answer because of the massive computational cost which is unsustainable. Totally. So, we have to ask, is it possible AI will dramatically increase business efficiency and productivity, but without increasing their profitability?
Yeah. just by forcing companies to pass those cost savings on to customers through brutal press wars. That is a fundamental challenge to the entire business model. Now, let's move to where the real red flags from past bubbles start popping up. The financing gymnastics. We're not just seeing equity investment from billionaires.
We're seeing massive debt commitments. The total infrastructure bill for AI, the data centers, the chips, the power is estimated to be a staggering $5 trillion globally. 5 trillion. And yet the biggest spenders Microsoft, Alphabet, Amazon, Meta, Oracle, they only had about 350 billion in the bank collectively.
This delta requires extremely aggressive financing. Think about the risk that poses to the wider economy, to your 401k, and to global financial stability. Right? We see companies like Meta, Oracle, and Alphabet issuing 30-year bonds to finance assets, chips, and data centers. That might become technologically obsolete in 5 years.
So, will those assets maintain their productive value long enough to repay obligations stretching out three decades? It seems unlikely, especially given the explosive rate of AI obsolescence. And talk about truly worrisome behavior. We are seeing what critics call circular deals. Companies are simultaneously giving money to and receiving money from the same small group of AI players.
Yeah, the money just goes around in a circle. Open AAI is set to receive billions from its parent companies, but those same companies received billions back when Open AAI pays them for computing power. Goldman Sachs even estimated that Nvidia, the biggest winner, could make as much as 15% of its sales next year from these deals.
When you see that kind of aggressive roundtrip financing, you know, we are entering speculative finance territory, this is where the Minsky moment comes into play. Okay, wait, let's slow down on that. For listeners who haven't encountered this concept, what exactly is a Minsky moment? Himman Minsky was an economist who described how stability creates instability.
A Minsky moment is that inflection point in a boom where credit expansion exhausts its productive projects and starts financing speculative or even Ponziike schemes. So, it's like realizing your neighbor is no longer borrowing money to renovate his house, but is borrowing money just to pay the interest on his existing loans.
That's a perfect analogy. That shift from financing real growth to financing existing liabilities is when the clock starts ticking toward a crash. So, we see this unhealthy behavior where startups are borrowing money to build data centers for other startups where both are losing cash and lack customer visibility.
It's borrowing to facilitate future borrowing. Precisely. And adding to the concern is the use of special purpose vehicles or SPVS. These are essentially financial shell companies used to take debt and risk off the parent company's main books, hiding the true leverage. Exactly. It makes it very difficult for outside investors to gauge the risk.
This was a critical practice that contributed significantly to the 2000 telecom bust. It's a way of saying this debt isn't really ours. If we connect this to the bigger picture, Bob Oly's formula provides a critical framework for this kind of winner take all competition. Oly's insight is key here. He argues that in an environment where a hundred projects start but only one will be a massive winner, equity is the right way to play, the massive gain from that single winner will compensate for the capital you lost on all the losers.
But using debt is the wrong tool for the job. It's the absolute wrong tool. The lender only gets the small coupon, the interest payment on the winner. That gain is grossly insufficient to compensate for the massive impairments on the debt of the other 99 losers. So this concentration of debt financing in the AI buildout is a major red flag.
It's a huge red flag. It suggests lenders are taking on equity level risk for a debt level reward. It's choice between risk sharing and just betting the farm. One anecdote we found just perfectly captures the sheer exuberance driving this market. The AI startup thinking machines. Oh, this one's amazing.
Helmed by a former Open AI executive, it recently raised a $2 billion seed round at a 10 billion valuation. This is despite having no product and refusing to tell investors what they were building. One investor called it the most absurd pitch meeting. And they still got the money. They still got the money.
It truly seems that when the potential for vast amounts of money is involved, the basic human desire to believe in magic just takes over. It absolutely has the hallmarks of speculative folly. But, and this is important, AI believers have very strong counterarguments for why this time is different, or at least why the technology justifies the massive capital.
Right? Unlike the dot bubble, AI products exist at scale with exploding demand and real revenue. We're seeing companies like Anthropic, whose coding product Claude Code went from 1 million in revenue in 2023 to 100 million in 2024. And it's expected to hit a billion soon. A billion. That's a 100x revenue growth in just two years.
That's a fundamentally different growth pattern than we saw in the late 90s. And crucially, the established players that dominate the spending, Microsoft, Amazon, Alphabet, they have robust diversified revenues, real profits, and massive cash flow. This isn't just funding vaporware. Exactly. And if you look at valuations today, the current price to earnings ratios for the big AI players are significantly lower than the insanity seen during the 1998 2000 internet bubble.
Microsoft's PE today is half of what it was 26 years ago. But surely those lower PE ratios are only lower because the denominator, the profits are concentrated in just a handful of companies like Nvidia and Microsoft. Doesn't that just raise the concentration risk? It means a few company specific failures could have an oversized impact.
That is the perfect counterpoint. The concentration of capital is the risk. If AI is an inflection bubble, the technology will survive and transform the world. But the majority of the current players, especially those relying on aggressive debt financing and circular deals, will not. So what does this all mean for us?
We've heard the extreme arguments for both bubble and technological necessity. As Sam Alman himself stated, are we in a phase where investors as a whole are overexcited about AI? My opinion is yes. Is AI the most important thing to happen in a very long time? My opinion is also yes. And since no one can definitively say whether the current exuberance is purely irrational, the material suggests the best path is financial moderation.
Not going allin, which risks ruin, or all out, which risks missing out. So, a moderate position applied with selectivity and prudence. It's likely the safest and most rational approach. Our sources make it clear that the AI genie is out of the bottle, and this time it is likely different in kind, not just degree, from prior developments.
And that's because it may take over cognition on which humans have historically had a monopoly. And this raises a far more profound question than the financial one, the societal impact. Joe Davis at Vanguard is deeply concerned about joblessness and purposelessness. He points out that for most desk jobs, AI could save about 43% of the time people spend on work tasks,
which could eliminate or profoundly change roles for entrylevel workers, coders, junior lawyers, and drivers, which is still the most populous profession in America. The optimists hope new, currently unimaginable jobs will materialize or that AI will simply make up for the 16 million baby boomers retiring.
But the scale of displacement demands immediate consideration. I mean, if universal basic income, UBI, becomes a necessary national burden, where does the money come from? And maybe even more critically, how do we replace the purpose, the structure, and the self-respect that a job provides to an individual?
That's the real question. So think about the careers our children should prepare for. The source suggests the most defensible jobs are those that involve physical tasks requiring hands-on dexterity like plumbers and electricians or roles requiring hands-on human care and empathy like nurses or perhaps roles that demand unique decadehoned human talent, deep insight and nuanced judgment.
So what stands out to you is that the future of work may not just be about money but about meaning. Precisely. The challenge of the AI age may not be figuring out how to make money, but figuring out what it means to be useful in a world where machines can perform cognitive tasks faster, better, and cheaper than we can.
That question of human utility, that is what we will all be grappling with for decades to
Comments 0