Are You the 'Exit Liquidity' for the AI Data-Center Bet?

Updated October 4, 2026

Two words kept showing up under every AI video I watched: exit liquidity. It sounds like trading-desk jargon. It isn't. It's just the person who buys at the top so the early money has somebody to sell to. So I went down the rabbit hole on one uncomfortable question — if the whole AI data-center boom is a giant bet, am I quietly the one on the other side of it? Here's what I could pin down, and the part I couldn't.

So what does 'exit liquidity' actually mean?

Strip the jargon off and it's almost rude how simple it is. Somebody who got in early wants out at a profit. To get out, they need a buyer. You're the buyer. The price they sell at is only real if someone shows up to pay it, and the last people to show up are usually regular folks who heard the thing was a sure winner.

Now here's the part that caught me off guard. I figured you had to go buy Nvidia or some AI stock on purpose to be in this. You don't. If you've got a 401(k) sitting in a plain S&P 500 index fund, you already are. Those seven big tech names — Nvidia, Microsoft, Apple, Amazon, Alphabet, Meta, Tesla — have swelled into roughly a third of that entire index. So out of every dollar in the most boring, set-it-and-forget-it fund in America, about thirty-odd cents now rides on the same handful of companies making the same giant bet. Nobody asked you. The math just did it for you.

Why the money looks like it's chasing its own tail

This is the part people keep calling 'circular,' and once you see it you can't unsee it. Nvidia sells the chips. OpenAI needs cash to buy the chips. So Nvidia turns around and puts money into OpenAI — a commitment that's been floated as high as a hundred billion dollars, though Nvidia's own finance chief admitted late in 2025 they still hadn't signed the final deal. OpenAI then promises to spend staggering sums renting computing power from Oracle and Microsoft, who are also building with Nvidia chips. Round and round it goes.

Picture a pizza-oven company that hands every new customer the down payment to open a pizza shop, on the one condition they only ever buy its ovens. Sales look incredible. The oven company's stock goes to the moon. And it all works great right up until one of those shops can't cover rent — and suddenly everyone asks how many of those 'sales' were ever real hunger for pizza. That's the worry here. Every dollar of revenue Nvidia and Oracle book from OpenAI is, right now, a dollar OpenAI is losing. The bond market noticed, too: the cost to insure Nvidia's debt against default jumped harder than it had since those contracts started trading.

Where did all the risk go? Mostly, off the balance sheet

Here's where I had to read things three times. The companies building these data centers have figured out how to keep a lot of the debt off their own books. Meta teamed up with a big private-investment firm on a data-center campus in Louisiana and pushed around twenty-seven billion dollars of debt into a separate company, so it never lands on Meta's balance sheet. Oracle sold eighteen billion dollars of bonds in a single day to fund its OpenAI buildout. A big chunk of this whole boom is being paid for with private loans and bonds that ordinary financial filings barely show.

And Meta didn't walk away clean. It signed what's called a residual value guarantee — basically a promise to keep paying for that building full of chips even if the AI demand it was built for dries up. For sixteen years. That's longer than most marriages and all of my cars put together. So if the thing cools off, the bill doesn't vanish into thin air. It lands on whoever signed the guarantee, and downstream, on the stock everyone's holding.

Then there's the quietest worry of all, and it's an accounting one. These companies assume their pricey chips stay useful for about six years, and they spread the cost out over that long. The investor Michael Burry — the one from 'The Big Short' — bet real money that the chips are really good for more like two or three years, because Nvidia keeps shipping a faster one every year. If he's right, these companies are making their profits look fatter than they are by pretending old chips are worth more than they are. That wouldn't be a dramatic crash. More like a slow leak you don't notice until the tire's flat.

So am I the exit liquidity? The part I couldn't pin down

Straight answer: I don't know, and neither does anyone yelling about it online. Nobody can call the timing. Circular money looks exactly like healthy growth right up until the second it doesn't, and these things can run for years longer than the skeptics swear they will. Plenty of smart people have gone broke being early.

What I kept seeing in the comments under every one of these videos wasn't really a question about Nvidia. It was the same scared, human question asked a hundred ways: okay, so what do I actually do? And that part I can't answer for you — I'm a person reading the same public filings you can pull up, not your financial advisor. What I can do is keep tugging the thread, because this rabbit hole runs way past the stock price. Into the power grid these data centers are straining. The gas turbines and nuclear plants getting fired back up to feed them. The water they drink. The question of who ends up paying the electric bill when a data center the size of a town moves in next door.

That's the whole reason Byte Bungalow exists — to follow where the money and the power actually go, in plain words, frame by frame. The video is where we dig into this one. Come watch it, and subscribe on YouTube so the next piece of the puzzle lands in front of you.

Common questions

If AI doesn't live up to the hype, will a broad index fund like VOO get hit hard?

Probably, and that's the uncomfortable part. A plain S&P 500 fund now carries roughly a third of its weight in the same handful of AI-tied companies. When a cohort that big drops, the whole index feels it — even the parts of your fund that have nothing to do with AI. People point out that in one past tech blowup, a single collapse dragged the broad market down hard even though that company wasn't in the top ten. The counter-move folks talk about is leaning into value stocks, smaller companies, or international holdings. It softens the blow; it doesn't erase it, because a big AI drop ripples through almost everything.

What does 'circular financing' actually mean in plain terms?

It's when the companies in a deal are mostly funding each other instead of pulling in fresh money from the outside. Nvidia invests in OpenAI, OpenAI spends that kind of money on Nvidia-powered computing, and the sales get booked as growth. The dollars travel in a loop. It can be completely legal and still be fragile, because if any single link stops paying, the whole loop wobbles at once.

Is this basically Enron all over again?

A lot of commenters jump straight there, and the comparison gets stretched too far. The real difference is that Enron hid its losses with outright fraud. What's happening now is mostly out in the open — the circular deals, the off-balance-sheet debt, the chip-lifespan assumptions are all disclosed if you go read the filings. The echo people are hearing is the off-balance-sheet structures and the 'this time it's different' confidence, not proven fraud. But legal and sturdy aren't the same thing.

What happens to all those Nvidia chips if the boom pops?

That's the bull case's best card. The argument goes: worst case, Nvidia or the data-center owners still hold piles of valuable hardware they can repurpose or sell off. The catch is the exact thing Burry is betting on — chips lose value fast when a faster model lands every year, so a warehouse of three-year-old GPUs could be worth far less than the books claim. They're still real physical assets. Just not the lifeboat some people assume they are.

Okay, so what should I actually do about it?

I genuinely can't tell you — I'm reading the same public information you can, not handing out financial advice. What the research does make clear is that 'ignore it and don't look' and 'understand exactly what you own' are two different choices, and only one of them is informed. Knowing that a third of your index fund rides on one shared bet is the sort of thing worth knowing before the bet gets called, not after.

Keep reading

My Electric Bill Doubled and I Can't Move. Here's the First Lever I Pulled.

Retired, bill doubled from the data center boom, stuck in your house? The cheapest real lever isn't using less power. It's a phone call. Here's what I found.

A packed room made the developer walk — so why did the people who won tell everyone not to cheer yet?

Does showing up to a town hall actually kill a data center, or just push the vote to a quieter night? I dug into what really happens after the room clears.

To Escape Data-Center Rate Hikes With Solar — Is Rooftop a Trap and Ground-Mount the Smarter Buy?

Data-center rate hikes push homeowners to solar. Is rooftop a trap and ground-mount smarter? What leaks, insurance, and resale do to the real math.

They Already Voted Yes on the Data Center — So Why Are There Still Two Decisions Left?

The data center vote already passed? It's not the last decision. Here's the air permit and utility rate case you can still weigh in on.

Why Don't Data Centers Cool With Treated Sewage Instead of Your Drinking Water?

A lot of data centers already cool with treated wastewater. The ones that don't usually can't — and the reason comes down to a purple pipe.

If the Ocean's Right There, Why Cool Data Centers With Drinking Water?

The ocean's right there — so why cool data centers with drinking water instead of seawater? A few already do. Here's why it's rarer than you'd think.

Sources

en.wikipedia.org — AI bubble concept since 2025; circular flow of investments inflating valuations; Nvidia's ~$100B pledge to OpenAI's 10-GW expansion boosts Nvidia hardware sales.
outlookbusiness.com — Morgan Stanley: Big Tech ~$3T on AI infra through 2028, own cash flows cover only ~half; if bubble bursts + data center goes dark, Meta owes Blue Owl a multi-billion payment. → use
fortune.com — Over 80% of 401(k)s default into target-date/cap-weighted index funds → automatic exposure to most concentrated part of market. → used for Section 1 ("nobody asked you, the math di
forbes.com — S&P 500 weight in Mag 7 passed 30%.
davefriedman.substack.com — the $176B accounting question framing.

Byte Bungalow

I do the homework out loud.

If this is the kind of thing you want checked against the documents instead of the hype, the video's above — and there's a new one every week.

Subscribe on YouTube →

By Byte Bungalow. Home power and home tech, checked against the documents instead of the hype. Independent commentary; not affiliated with any manufacturer, utility, or builder named here. Not professional electrical advice.