Bubbles end with someone losing money, and which someone matters. The AI boom has already found a way to put the public on the hook before a single IPO lands.
The AI “bubble.” A tale as old as time. By now, most of us have seen it show up in our feeds. Depending on who you ask, it’s either the media’s usual fear-mongering to farm views, or we’re sleepwalking into another 2008 (read: fraud).
Before I became a software engineer, I put on a suit, sold my soul, and worked on Wall Street. So these days I find myself more interested in the macro picture around AI than in whatever model update of the week is being peddled on Hacker News.
It helps to start with a baseline. The textbook definition goes something like this: a bubble is a market cycle where asset prices (stocks, real estate, whatever) surge fast on speculation and investor optimism, until they sit far above what the underlying fundamentals can justify.
There are two key ingredients in there: the run-up is unsustainable, and it’s fueled by speculation (aka hype). I’d like to propose a third: whose money is on the line.
The scariest part of any bubble is the ending, when people lose their money. But which people lose matters a lot. Broadly, there are three buckets of money that can get stuck holding the bag.
Private money. Picture a company funded entirely with private capital, meaning investments that regular retail investors can’t touch. Stripe is the classic example. You can’t log into your favorite trading app and buy shares of Stripe, and you can’t call up a financial advisor and get a stake. But institutions, hedge funds, and accredited investors can. So purely from an invested-dollars perspective, if Stripe went bankrupt tomorrow, who’s at risk? The rich folks and the institutions. And is that so bad? Gatekeeping these deals to people who can afford to lose the money is the whole point of the accredited investor rules.
Retail money. The moment those opportunities open up to everyday investors through our 401(k)s, IRAs, and brokerage accounts, we’ve entered an entirely new realm of risk. Now it’s regular people’s retirements on the line.
And this isn’t hypothetical. AI debt already makes up roughly 10% of the biggest investment-grade bond ETFs, and financial advisors are funneling retail money into private credit funds that lend directly into the buildout (Forbes).
Taxpayer money. Then there’s the worst bucket of all. When losses get socialized through a bailout or a government backstop, everyone pays, whether they invested or not. Remember this one. We’ll come back to it.
As of this writing, neither Anthropic nor OpenAI is publicly traded. And both have raised an absolute fxck-ton of money: roughly $180 billion for OpenAI, including a single $122 billion round in March, and around $130 billion for Anthropic, including a $65 billion round in May at a $965 billion valuation.
So put yourself in their shoes. You’ve tapped just about every pool of private money on the planet. Regulations stop you from going direct to consumers. And you need more money, fast, to keep this train on the tracks. You need public money. How do you get it? Go public.
That’s exactly where this is headed. Both companies confidentially filed for IPOs back in June. Anthropic is reportedly targeting a listing before Thanksgiving, while OpenAI has pushed its own toward 2027.
Going public isn’t easy, though. Set aside the heightened scrutiny and regulation: an IPO means opening the books, and right now no one really knows the math of these companies. I’m not convinced the companies know it either, because the math is no longer math-ing. For valuations this big to pay off, they’d basically need to solve quantum computing within the next year. Ha, good luck.
So while the IPO machinery grinds along, is there a way to get at public money in the meantime? It turns out there’s a back door, and it runs through money that’s ultimately backed by the government. Wait…what?
Like many grand economic heists, this one starts with private equity.
Private equity has been scooping up life insurance companies. PE firms now own at least $1.5 trillion in life insurer assets. Apollo bought Athene, and KKR bought Global Atlantic. Why does that matter?
Because life insurers sit on mountains of cash: policyholder premiums that don’t need to be paid out until the end of people’s lives. PE owners take that cash and pour it into private credit, which is deeply tangled up in AI, including loans to build data centers. According to research by Andrew Granato and Pranjal Drall, nearly half of new PE-owned insurers’ 2024 investments went into privately placed instruments, versus about 14% for other insurers.
So what happens if the insurer goes insolvent? Isn’t it bad to gamble with money that’s owed to people at the end of their lives? This is the crux of it all. An insolvent life insurer doesn’t go through bankruptcy. Instead, policyholder claims are paid by state guaranty funds, which then bill the surviving insurers in that state. And in 44 states, those insurers can claim a tax credit for 100% of that bill over time.
Read that again. The losses end up on state taxpayers, and the PE firm that took the risk walks away. As David Dayen put it in The AI Bailout Could Be Baked Into the AI Bubble:
It’s a familiar story of privatizing the profits and socializing the risks.
AI and its hype are now propping up a big chunk of the US economy. ING estimates AI investment accounts for roughly a third of US GDP growth this year.
The concentration risk is wild, too. Microsoft disclosed that it booked $24.1 billion from OpenAI in fiscal 2026 (July 2025 to June 2026). That’s about 7% of Microsoft’s total revenue, and by Bloomberg’s estimate around 70% of its entire AI business. That’s a TON of power, leverage, and risk in the bucket of a single customer, and Microsoft isn’t the only company in that position.
Now layer on inflation, which climbed back to 3.7% in September, driven largely by energy prices. A bailout is a lot harder to pull off when inflation is already running hot. The usual playbook of cutting rates and flooding the system with money gets riskier, and that’s how you end up with a trap door: things drop straight down and spiral out of control.
So where does this leave us? The industry has already found a way to slide “public” money underneath the AI boom, through PE-owned life insurers and the state guaranty funds behind them. Our bond funds and retirement accounts are already exposed, and once the IPOs land, that exposure only grows.
If (read: when) this bubble pops, it’s more than likely you and I left holding the bag, while the big institutions count their money until their fingers bleed. And even in the lucky scenario where the institutions do feel the pain, we’re still screwed. AI has become too big to fail.
We’re cooked, chat.