OpenAI spent roughly $22 billion against $13 billion in revenue in 2025, and doesn’t expect to be cash-flow positive until 2030. It’s also about to go public at a valuation near a trillion dollars, alongside Anthropic and SpaceX, three of the largest private companies ever built, all listing within months of each other.
Most of the coverage reads this as the AI age finally arriving. There’s a different way to read it. Going public is the moment the dream meets the balance sheet, and the dream has run a long way ahead of what AI has actually delivered at ground level. The technology is real. The valuations, and the subsidies propping up cheap access, are the bubble. These IPOs are where that gap gets exposed.
The mistake almost everyone makes is treating “AI” as one thing. There are two. One is financial: the valuations, the circular financing, the below-cost pricing. That one is a bubble. The other is operational: a business using these tools to do real work for real money. That one isn’t. Telling them apart is the whole game.
Cheap AI was never free
You’ve been running on subsidized AI, and most people building on it don’t realize it. The reason you can run a frontier model for the price of a sandwich is that someone else is eating the difference. These companies price access below what it costs to deliver, on purpose, to win the market before they win the margins. The OpenAI numbers above are the tell: those are the economics of buying market share, not selling a product.
We’ve seen this movie. For years a wave of venture-backed companies paid us to use their products. Derek Thompson called it the millennial lifestyle subsidy, the era when an Uber across town cost a few dollars because the company behind the app was losing money on the ride. The moment capital got expensive and those companies had to answer to public shareholders, the discounts ended and prices climbed. Cheap AI is the same trade, and it lasts exactly as long as someone is willing to fund the losses.
Why they’re all going public now
The reason all three are listing at once isn’t triumph. It’s that the private well is nearly dry. These rounds got so large there’s almost no private money left big enough to feed them. OpenAI’s $122 billion round in March was the biggest private raise on record. Anthropic is weighing what’s described as its final private round before an IPO, at a valuation above $900 billion. The capital requirements have outgrown the entire venture universe, so the only pool deep enough left is the public market.
And the day you go public, your masters change. Private investors buy a story, and they’ll tolerate enormous losses to ride it. Public shareholders buy a stock. They want growth they can model and margins they can see, on a quarterly clock. When that incentive flips, the subsidy is the first thing to go. The free tiers shrink, the cheap API creeps up, and the businesses that planned around permanently cheap tokens are the ones who get squeezed.
The growth is real, the ground-level impact isn’t
Give these companies their due, because this is where the bears get lazy. This is not Pets.com. OpenAI is doing roughly $2 billion a month in revenue with more than 900 million weekly users. Anthropic’s run-rate went from about $9 billion at the end of 2025 to past $30 billion months later, driven largely by enterprise demand. The demand is real.
But the valuations are priced for a remaking of the economy that, on the ground, has barely started. Walk into actual companies and most have a ChatGPT subscription and a vague plan. Goldman Sachs found that 76% of small businesses say they use AI, while only 14% have embedded it into core operations. What that 14% looks like in practice is the whole point of our case studies. The price assumes the destination. The reality is still the on-ramp.
It rhymes with 2000
The dot-com crash is the closest map we have, and the popular version of the story is wrong. People remember it as “the internet was overhyped and it crashed.” What actually happened was a convergence, and every piece of it is forming again.
Capital got expensive. Cash burn became undeniable once the filings were public. The companies were finally judged on results instead of vision. And insiders headed for the exits when their lockups expired. The NASDAQ fell more than 76% over the next year and a half.
But the IPOs never caused it. They were the stage where the dream met the numbers and insiders found a way out. Today’s three listings don’t have to cause a correction to be the first domino. They’re the moment the market gets to see the math and decide whether it still believes.
Operational AI is not the bubble
Here’s the part that matters if you run a business. The internet didn’t disappear in 2001. The companies pretending to be the internet did. The same split is coming for AI.
Financial AI, the valuations and subsidies and circular financing, is what these IPOs put at risk. Operational AI, the real work of using these tools to do more with less, is what survives the correction. It’s the kind of system we build through AI engineering: early, under-hyped relative to the frenzy, and indifferent to what the NASDAQ does.
Build something worth paying full price for
So the move isn’t to wait for the crash or to ignore the risk. It’s to build value that holds up at full price.
If your AI advantage only works because tokens are artificially cheap, you don’t have an advantage. You have a subsidy on a clock. The custom systems that come out the other side of a correction are the ones that would still pencil out if inference cost what it actually costs.
That’s the work we do. The fastest payback right now isn’t customer-facing, it’s the repetitive, multi-step internal process nobody enjoys owning. You can see one in production in our 4AM Media build, where agents handle 90% of support tickets and recovered $200K a month, or across the rest of our e-commerce work. The numbers, and what broke along the way, are in the case studies.
The bubble bursting is a financial event. Whether it helps you or hurts you depends on whether you built something real before it happened. The question was never whether you believe in AI. It’s whether what you’ve built survives the moment the subsidy ends.
