SpaceX is about to become the biggest IPO in history. $75B raised. A $1.77 trillion valuation. Elon Musk on track to be the world’s first trillionaire.
Everyone’s posting that number today. Almost nobody is looking at the part that actually matters.
@realMaxAvery The loop. If capital keeps flowing between related entities, overvaluation doesn’t matter. The day it stops, structure matters more than narrative
SpaceX is about to become the biggest IPO in history. $75B raised. A $1.77 trillion valuation. Elon Musk on track to be the world’s first trillionaire.
Everyone’s posting that number today. Almost nobody is looking at the part that actually matters.
@firejuggler1aw You’re probably right on valuation. But that’s not the real question. The question is whether the structure holding up that valuation. the loop breaks before the narrative compounds enough to justify it. Overvalued companies survive if capital keeps flowing
@vanwidaAI Fair point and maybe he pulls it off. But the $1.77T isn’t priced for mission. It’s priced for growth trajectory. If growth plateaus, narrative or not, structure matters
@simonwesterlund Right and that’s exactly the point. xAI doesn’t need to beat Anthropic. It needs to keep renting compute to labs long enough to justify the $1.77T story. The loop doesn’t require xAI to innovate. Just to keep spending.
@MuldoonMartin Exactly. And that’s the question: do those returns come from AI/orbital data centers actually working, or from the same dollar circulating between related entities until it stops?
Zoom out and this is the template for the AI era: huge losses, a trillion-dollar number, and a tight circle of related companies funding each other’s demand.
SpaceX won’t be the last. Anthropic already filed. OpenAI is next.
The real question: how much of this demand is real, and how much is just the same money going in circles?
And here’s the catch for anyone buying in.
After the largest IPO ever, Musk walks away controlling 82% of the voting power through special shares.
You won’t be buying a say in SpaceX. You’ll be buying a ticket to watch Elon run it.
Two of the most famous trades ever and neither was stock-picking. Both were macro conviction calls. Honest question for the timeline: in an era of passive flows and everything indexed to a handful of mega-caps, is a single asymmetric macro bet like this even still possible? Or did it die with zero rates?
Wall Street Is Having the Wrong Argument About AI
The fight over whether AI stocks are a bubble misses what's actually fragile underneath them.
Nvidia is worth more than five trillion dollars, the first company that ever has been. The figure is too big to picture, so for scale: that one chipmaker is now valued at almost five percent of the entire world's yearly economic output. This week the S&P 500 closed above 7,600 for the first time while Jensen Huang worked the stage at Computex in Taipei, and Goldman's David Solomon told CNBC that investors had tipped into what he called "greed."
Cue the bubble question. Are we in 1997, with the good part still ahead, or 1999, with the floor about to give out? You'll hear some version of that on financial television every day between now and December, and it's the wrong argument.
The dot-com comparison smuggles in an assumption — that the risk is the price, that AI stocks have simply gotten too expensive and will eventually come back down to where they belong. That isn't the dangerous part. Fixating on it is a way of staring at the scary thing without actually seeing it.
Take the case the bears like best. In 2000, Cisco was the Nvidia of its moment, and Cisco was mostly selling a promise. Nvidia sells actual silicon at margins a luxury-goods house would envy: north of $215 billion in revenue last fiscal year, gross margins above 70 percent, profit margins past 50. Whatever this is, it isn't fumes. The companies carrying the rally are some of the most profitable businesses that have ever existed, and anyone waiting for the trade to implode because the underlying economics don't add up is going to be waiting a very long time. The economics add up fine.
What doesn't add up is the shape of it.
The concentration problem
Six companies now make up close to a third of the entire S&P 500: Nvidia, Microsoft, Alphabet, Amazon, Broadcom, and Meta. The index hasn't been this lopsided in something like half a century. Its cyclically adjusted P/E started the year above 40, which it has done exactly once before in 155 years of records, right ahead of the last tech crash. Palantir, just for color, opened 2026 trading at more than 100 times sales, a multiple that has essentially never held up for anyone who's ever printed it.
The plain translation is that the boring index fund inside tens of millions of American 401(k)s is, beneath the label, a concentrated bet on one technology theme. When six names are a third of the market, "a contained selloff" stops being a thing that can happen. If the AI story stumbles for a quarter, so does the retirement balance of someone who has never once thought about graphics chips.
The part nobody had to engineer before
Concentration isn't the genuinely new feature, though. The new feature is the circle.
Here is the machinery that's been powering a real share of this rally. Nvidia puts money into OpenAI. OpenAI turns around and commits hundreds of billions to cloud companies like Oracle. Oracle borrows tens of billions to buy Nvidia chips so it can meet those very commitments. The cash travels in a ring, and at three or four stops along the way, a single dollar of spending lands as somebody's revenue, somebody else's backlog, and a third party's growth story. Total up the vendor-financing deals and the 2026 estimates run past $800 billion.
At the middle sits OpenAI: a company that still loses money, that has booked roughly $25 billion in annualized revenue against more than a trillion dollars in infrastructure commitments, and that is, improbably, the outfit whose appetite underwrites the forecasts of the most valuable corporations on the planet. When AMD wanted a piece of OpenAI's spending, it didn't simply write up a purchase order. It handed OpenAI warrants on something like a tenth of the company for a penny a share. There is a phrase for paying your customer to remain your customer, and "ordinary sales contract" is not it.
The optimists deserve a fair hearing, because they're not wrong about the history. Capital-hungry industries have always been built this way. The railroads were. Telecom was. Making expensive things means somebody has to bankroll the ugly stretch between "we promised" and "we're profitable," and Janus Henderson would prefer we call the arrangement a "virtuous circle." Fine. It might be one.
The catch is that a virtuous circle and a doom loop are indistinguishable from the outside. The only difference is which direction the money runs, and that can flip without much notice. The thing holds together for exactly as long as every participant keeps feeding it. Let one big participant flinch and it runs in reverse at the same speed. We got a trailer for this over the winter, when it came out that Nvidia's planned $100 billion for OpenAI had stalled. Oracle's stock dropped, and the company rushed out a statement that the news had "zero impact" on its OpenAI relationship. By March, Huang was telling investors the $100 billion everyone had been counting on had never really, in his phrase, been "in the cards." When a company has to publicly reassure you that its most important customer's biggest backer getting cold feet changes nothing, the market has already done the math for you.
The test worth watching
So here is what I'd actually track for the rest of the year, and it has nothing to do with price-to-earnings: how much of this revenue is arriving from outside the circle?
A dollar from a normal business paying for AI because it eliminates a real cost is worth far more than a dollar that left a chipmaker, moved through a lab, paused at a cloud provider, and came home to buy more chips. The first is demand. The second is an echo with an invoice stapled to it. The hopeful version, which again is not a crazy one, is that ordinary companies are now adopting this technology fast enough to grow into all the infrastructure that's been poured for them. The grim version is that the data centers got built for customers who haven't fully arrived, and that an uncomfortable share of the customers who have arrived are the industry buying from itself.
You won't read the answer off a valuation. You'll read it in the quality of the revenue, on the day companies are forced to spell out how much of their AI growth came from genuine outsiders and how much came from the loop.
The dot-com crash was never a verdict on whether the internet was real. The internet was extremely real. It rewired the whole economy. It simply did so on a slower clock than the share prices had promised, and the borrowing stacked on top of the faster clock was the first thing to break. The fiber didn't evaporate. The companies that had leveraged themselves against demand that wasn't there yet did.
AI is real too, and it was always going to be. That was never the question. The question is whether you can run the entire American stock market through one tight circuit of six firms funding one another, and still keep the lights on the day a single one of them decides to stop.