When Will AI Companies Actually Turn a Profit?

In this era of AI semiconductors hitting new highs one after another, investors can't help but wonder: how long can this rally possibly run? We can say with confidence that Jensen Huang, AI's high priest, is making money. But training new models on his chips is what AI is actually about. There's an old Chinese proverb about a buyer who purchased an ornate box containing a pearl and returned the pearl, keeping only the box — and right now, the company selling the fancy box is making money, while the company selling the pearl still isn't. Take OpenAI, the most successful AI company today: this year it may spend $8.5 billion but only bring in $3.5 billion in revenue, putting it on track to lose roughly $5 billion.

Nobody Knows Who Wins Yet — But Nobody Can Afford to Fold Now

That's why David Cahn, a partner at top-tier venture firm Sequoia Capital, recently wrote a sharp essay arguing that the AI industry as a whole will need to generate $600 billion in future revenue to justify what's being spent on it. This piece had a predecessor from last year, asking the same question — back then, the number was that the AI industry needed to prove out $200 billion in revenue to make the math work. In less than a year, that figure has tripled.

I think it's a good question — the kind of question even a fool could ask. But if Sequoia counts as fools, there aren't many smart people left in the world. When a smart person asks a dumb question, it usually turns out to be a good one. Sure enough, during this year's earnings season, every major tech CEO got asked some version of it on their quarterly or half-year calls: you keep spending on chips and talent — so when does this actually turn a profit?

Think about it from the CEO's chair. Faced with that question, you really only have two answers: double down firmly on AI investment, or admit you don't see the payoff and plan to pull back. If a CEO chose the second answer, investor confidence would collapse instantly — because while nobody knows who's going to win this race yet, everyone would immediately know who just lost it. Every partner and investor would head for the exits. Unsurprisingly, every single CEO said the same thing: we're doubling down on AI. Even Meta's Mark Zuckerberg published an open letter to make his commitment known.

Analysts recently pointed out that Microsoft plans to pour another $63 billion into AI. Even those same analysts admit they have no idea how Microsoft turns that into short-term profit — yet the capital markets are still granting Microsoft a solid price-to-earnings multiple. Because doubling down always beats folding.

So, having posed the brain-melting question of "when will AI start making money," that same David Cahn at Sequoia turned around and wrote another essay explaining why going all-in on this spending spree is actually rational (the man clearly loves asking himself questions and then answering them). His argument: cloud providers are locked in a brutally tight oligopoly, and if you don't keep pace right now, you fall behind — possibly hopelessly so. That dynamic is forcing Amazon, Microsoft, and Google to race to upgrade their AI infrastructure at all costs. If you don't match the raise, your models fall behind, and you're out of the game entirely. These three companies alone carry a combined market cap of $7 trillion — they can more than afford to play a game of chicken.

Whoever Can't Raise Money — or Won't Spend It — Is Out

One tier below the pure muscle-flexing at the very top of AI, there's a second circle of startups doing reasonably well — France's pride, Mistral, which just raised 600 million euros, and Anthropic, which took $4 billion from Amazon. Outside of the three giants and this handful of well-funded names, everyone else is getting caught in the churn and struggling badly. Companies that still can't raise serious money look headed for elimination, and companies that aren't willing to spend big may be headed out the door too — or at minimum, they'll need to change course.

Why are the big tech companies spending so recklessly? And what actually happens to the companies that get knocked out?

Past experience tells us this: for any category of software that's a universal, global-scale utility used by ordinary consumers, the winners end up being vanishingly few. Mobile operating systems? Just two. The one-time third-place contender, BlackBerry, has vanished entirely. Social networks, if we're being honest, come down to a handful of companies you can count on one hand — fewer than five — and both Google and Meta each effectively own two massive social platforms apiece, so one company counts as two. Search engines, strictly speaking, really come down to just one: Google, with 92% market share.

The current AI trio of Google, Amazon, and Microsoft — plus Meta, king of open-source AI — all clawed their way to the top through exactly this playbook: knock out every rival until you own a monopoly, or at least end up in a tight oligopoly with a small handful of survivors. This is a Squid Game they know intimately well. The logic is simple: nobody wants to use the sixth-best social network, and advertisers certainly don't want to buy ads on it either. You wouldn't even want to use the third-best mobile operating system, would you? So the fourth-best large AI model may genuinely end up with no users at all — and the trouble is, the world has a lot more than four first-rate tech companies competing for those top spots.

But the top two or three large models — because everyone will effectively be forced to use one of them (aren't you, after all, forced to use one of the top two mobile operating systems already?) — will eventually be used by nearly all 8 billion people on Earth. Charge everyone just $1 and that's $8 billion; charge $100 and that's $800 billion. That's more than enough to cover the kind of investment figure David Cahn is talking about. And if one year's worth of revenue isn't enough, they'll just collect it again the next year — exactly like how phone subscriptions work today. And since AI algorithms and infrastructure won't only serve humans — they'll also run robots and all sorts of devices — the eventual deployment scale will run well past the tens of billions.

So I — and David Cahn — remain bullish for the long run. But for a good long while in the near term, the AI industry is still going to struggle to answer the question of whether all this investment actually pays off. One more thing: Jensen Huang sits on the other side of this equation — he's not the one who has to prove this math works. The man didn't earn the nickname "AI's high priest" for nothing. Of course, chips are just the tool for running algorithms, models, and software applications — they're not the end product. So over a longer horizon, I still believe the value chain will eventually shift toward software. It's just going to take a while.