The AI Labs Are Going Public. Don't Bet Your Business on the Winner.

Within the next few months, the two companies at the center of the AI boom are going to become something they have never been: public companies, answerable to shareholders.

Anthropic has confidentially filed for an IPO. OpenAI is expected to file within weeks, with a public debut that could come as soon as September. And the balance sheets they are bringing to Wall Street could hardly be more different. Anthropic is reportedly running at about $30 billion in annualized revenue and projecting positive cash flow by 2027. OpenAI is running at about $24 billion, burning roughly $17 billion in cash this year, with a projected $14 billion loss and a breakeven date that has slipped to 2030.

Same race. Wildly different math. Both asking the public for money.

If your business runs on these tools, the question that matters isn't which lab wins. It's whether you should bet your business on either of them.

The Numbers That Don't Add Up

Start with the figures, because they are genuinely strange. Every number here is a reported or run-rate figure rather than an audited result, but the picture is consistent across the coverage.

OpenAI has somewhere around 900 million users. Roughly 5% of them pay. The company brings in about $2 billion a month and is still on track to burn around $17 billion in cash this year. Its own internal projections reportedly put 2026 losses near $14 billion, with no profit expected until the end of the decade.

Then there is xAI, the third runner. It lost about $2.4 billion in a single quarter on $7.7 billion of capital spending. To keep training, it is renting Anthropic roughly 300 megawatts of compute for about $1.25 billion a month through 2029. Read that twice: one AI lab is paying another roughly $15 billion a year just to keep the lights on.

Notice the circularity. Tens of billions raised, much of it spent on compute bought from each other, from Nvidia, and from the cloud providers. Valuations climbing on revenue that is real but dwarfed by the spending required to produce it.

None of this is fraud. These are growth-at-any-cost numbers, the kind that look brilliant if the growth keeps compounding and catastrophic if it stalls. The only real question is whether the growth outruns the burn before the money gets impatient.

The Moat Is Leaking

There is a second crack in the IPO story, and it is worse than the burn rate.

The pitch to public investors is a premium moat: these labs are worth their staggering valuations because they are far enough ahead that no one can catch them. But the catching up is already happening. Cheaper models, including open-weight ones and a wave of Chinese labs, are matching frontier capability at a fraction of the price, and they are taking share fastest in the enterprise segment the IPO story most depends on. CNBC put it bluntly in a May headline: cheap AI could derail both IPOs.

I mapped this tension a few weeks ago in Open vs. Moat. The short version: the labs are selling Wall Street on being ahead, while a growing share of the market quietly decides that good enough is cheaper. You can charge a premium for a lead. You cannot charge a premium for a lead that is shrinking.

But the Technology Is Not the Bubble

Here is where I have to be careful, because the easy read of everything above is "AI is a bubble, get out." That is the wrong lesson.

Tulips were a bubble. The tulips still bloomed. Railroads were a bubble, the speculators got wiped out, and we still have railroads. The dot-com crash erased Pets.com and a hundred companies like it, and it left Amazon standing. A bubble in the financing of a technology is not the same thing as the technology being fake.

I build software with these tools every single day. I have written about it in Why I Build Software with Claude Code and Two Months of Claude, and the capability is not only real, it is compounding, to the point that AI now writes most of its own creators' code. What might be overvalued is the equity of specific companies. The usefulness of the models is not in question.

So hold two questions apart. "Is this a bubble?" and "Is this real?" are different questions, and right now they have different answers: maybe, and absolutely.

What an IPO Actually Changes for You

Set the bubble question aside for a moment, because even in the rosiest scenario where everyone thrives, going public changes a vendor in ways that reach all the way down to your invoice.

A public company answers to shareholders, by law and by the gravitational pull of the quarterly earnings call. That changes the incentives of the tools you depend on:

• Pricing pressure goes up. Public markets demand growing revenue per user and expanding margins. Expect consumer-pricing experiments and tighter discounts on enterprise deals, often starting in the first full quarter after the bell rings.
• The roadmap follows the revenue. Features that move the numbers Wall Street watches get built first. If your use case is not one of those numbers, it waits.
• The "safe lab" promises meet the stock price. Every governance and safety commitment these companies have made is about to be tested against the pressure of a falling share price. Some will hold. Some will not.
• You inherit the volatility. A missed quarter, a competitor's leapfrog, a cheaper alternative: any of these can swing an AI stock hard, and your roadmap now rides on a company that has to manage its share price as much as its product.

None of this is doom. It is just that the thing you build on is about to get a new and much louder boss: the market.

How to Build So It Doesn't Matter Who Wins

Here is the good news, and it is the whole point of writing this. You do not have to predict any of it. You can build so that the outcome barely touches you.

• Own your core. The software, the data, and the workflows that actually are your business should belong to you, built on top of the models rather than trapped inside one vendor's product. This is the argument I made in Custom Software Is Cheaper Than You Think Now, and the IPO race only sharpens it.
• Stay portable. Design so that swapping one model for another is a configuration change, not a rebuild. Apple just turned the model into a pick-list, letting users choose ChatGPT, Gemini, or Claude for the same features. Your stack should treat it the same way: a thin layer over whichever provider is best this quarter.
• Depend on capability, not a logo. Use the best model for the job today, and assume it will be a different one in a year. Loyalty to a vendor is not a strategy. Results are.
• Don't prepay the hype. A long lock-in contract or a deep, proprietary integration with a single lab is a bet on that one company's survival and its pricing discipline. Keep your exit cheap.

The businesses that get hurt when the financing corrects will not be the ones using AI. They will be the ones that bet the company on a single vendor's stock chart.

Build on Rock

The oldest piece of financial wisdom is not "call the top." It is quieter than that. Before you build the tower, the old teaching goes, sit down and count the cost, and make sure you can finish what you start. Don't anchor your hope to uncertain riches, because riches are the most uncertain thing there is.

You do not need to predict whether this is a bubble. You need a foundation steadier than a valuation. Build the house on rock, not on the sand of whichever lab is winning this particular quarter, and the storm, if it comes, becomes something you watch rather than something you drown in.

So use these tools, gladly. I do. Just don't mistake a lab's share price for your foundation. Own what is yours, stay free to move, and let Wall Street worry about Wall Street.