The business model is well-understood – the same as any casino or street corner hustler: get the user addicted, then raise the price. But AI is at such a different scale, what with their estimated seven trillion dollar infrastructure needs and – until now – dime bag pricing. So how are they going to pay for all that?

But maybe that’s the wrong question. The better question is, now that the software engineering industry is starting to rethink usage due to price increases, what are people going to use AI for? If going full Gas Town or Dark Factory is going to get you half billion dollar monthly token bills without a clear link to useful consumer features, then what’s AI good for?

Even I will admit that there are some situations in which AI appears to be genuinely useful. Code reviews. Security reviews. Explaining legacy code. Critiquing technical designs. Anything where a correct answer is valuable and an incorrect answer doesn’t cost too much. But this is far from the grand vision we were sold—at one point, people were straightfacedly arguing that we wouldn’t even have applications anymore, we’d just ask AI to build whatever we wanted in the moment, de novo, every time. This was laughable even then, and more so now that the true cost is being passed down to users.

Meanwhile, the hucksters and wannabe influencers are still posting on LinkedIn about how their new startups are managing clouds of agentic bots inside carefully constructed guardrails with test harnesses and meticulously constructed (by the bots, of course) unit tests and acceptance criteria, and about how this is clearly better in every way… But their drip campaigns already seem hopelessly behind the times, as Amazon and Microsoft start controlling costs, and as subreddits fill with horror stories of monthly token limits being wiped out in hours after usage-based price increases.

So I ask again – if even the wealthiest companies are cutting back, and if the back-and-forth iterative development process is punitively expensive, then what are people going to use AI for? Tokenmaxxing was always a dumb idea, and the companies that incentivized ever-increasing token usage are either going to learn (very quickly) or die. Likewise, a generation of startups built around the old pricing and fake-it-til-you-make-it / no-engineers-only-makers hustle will pivot or die as their business models run up against astronomical costs. The companies that survive will be the ones that go on a diet, figure out use cases with acceptable ROI, and rigorously control token usage.

What this means for you, today, is that if you aren’t already managing token usage, then you’ve given every employee at your company a blank check and a mandate to spend infinite money. You won’t even need a disgruntled engineer to set up fleets of bots to argue about the best way to implement Doom on a smart toaster. All it will take is one intern to get excited about your company’s six month-old AI strategy, start experimenting, and suddenly you’ll have burnt through gazillions of dollars. It’s just not that hard. You need to lock that stuff down, and fast.

But—and I can’t emphasize this enough—this doesn’t address the more fundamental problem. To protect your company, you’re going to put caps in place, and you won’t be the only one. The software industry is at the end of an all-you-can-drink bender, and—with prices rising dramatically—is starting to tighten the belt and focus on ROI and actual generated value. It’s going to converge on less extravagant use cases, which aren’t going to be able to sustain the annual revenue needs of the AI companies.

The budget for AI isn’t infinitely elastic. The AI companies are never going to be profitable. You know that. You’ve always known that. Maybe you got caught up in the hype, but in the same way you knew that NFTs were a scam, you’ve always known that the math for these companies just couldn’t work. You might have wanted to believe, but in your heart of hearts you knew, and here we are. The use cases which are both valuable and affordable can’t support an industry that needs trillions of dollars in annual revenue to survive.

To be clear, I have no idea how long it’s going to take. Hope is a stubborn thing. But as Stein’s Law says, if something can’t go on forever, it will eventually stop—and just like the 2000 dot-com crash and the “new economy,” we’re headed for a cliff.

---

All em-dashes in this post were spitefully generated using alt-shift-hyphen.