Is There an AI Bubble? The Question That Matters More for Your Business

A client asked us recently whether they should "bet on AI" before it's "too late," in roughly the same breath as asking whether the whole thing might be a bubble about to pop. Both concerns are reasonable. They're also, for a business trying to make an actual decision this quarter, mostly the wrong question.

Whether AI valuations correct is a market question. Analysts and investors are paid to argue about it, and we're not going to add an uninformed opinion to that pile. The question we can actually help answer, and the one that determines what happens to your business regardless of how that market question resolves, is narrower: does the specific thing you're considering building solve a real problem you have today, at a cost and risk you understand?

Two different questions keep getting conflated

"Is AI overhyped as an investment category" and "should we build this specific feature using AI" are unrelated questions with unrelated answers. A market can be overheated while a specific, well-scoped use of the underlying technology is still the right engineering call for your business. A market can also correct hard while genuinely useful capability keeps getting used, quietly, by the businesses that adopted it for a real reason instead of a speculative one. The correction, if it comes, mostly clears out the second category of bet, not the first.

What survives a correction, historically

The dot-com crash is the obvious comparison, and it's instructive for a reason people usually get backwards. A huge number of companies built on "the internet changes everything" evaporated. The infrastructure and the genuinely useful patterns underneath them, broadband, e-commerce, search, did not evaporate; they kept compounding for the next twenty years, boringly, while the speculative layer on top got wiped out. The businesses that came out ahead weren't the ones that avoided the internet. They were the ones that used it for a specific, defensible reason instead of because everyone else was.

We'd expect the same shape here. Narrow AI capability that solves a specific, bounded problem, document extraction, support ticket triage, code scaffolding, doesn't stop working if a handful of richly valued companies correct. What stops working is the strategy of bolting "AI" onto a product with no clear problem it's solving, because that was always closer to a marketing bet than an engineering one.

The actual risk isn't the bubble popping

The risk we actually see clients exposed to isn't macro. It's built into how the feature gets shipped. A chatbot wired directly to one vendor's API with no fallback, at a price point that assumes today's pricing holds forever. A core workflow that quietly depends on a specific model's current behavior, with no plan for what happens when that model gets deprecated or changes its answers on a routine update. A "strategy" with no scoped use case behind it at all, just a slide that says the company is "AI-forward," waiting for someone to explain what that's supposed to mean operationally.

None of that is a market problem. It's a scoping problem, and it's the same problem regardless of which way valuations move.

How we actually scope this to survive either outcome

We build for a specific, bounded use case with a return on investment we can point to now, this quarter, not a speculative one that only pays off if a trend keeps accelerating forever. We avoid single points of vendor failure where the cost of switching later is reasonable, not catastrophic. We keep a human in the loop on anything consequential, which was always good practice and happens to also be the version of "AI-forward" that doesn't collapse if a specific model or vendor has a bad year. None of this requires guessing where the market goes. It's just the same engineering discipline we'd apply to any dependency on any third-party technology, applied honestly instead of set aside because the technology is exciting.

If you're trying to decide whether to build something with AI in it, the market forecast isn't the input that actually helps you. What the feature solves, for whom, at what ongoing cost and risk if the landscape shifts under it, is. Start there, and the bubble question mostly stops being one you need an answer to.