The AI Bubble: Startups Funded Before Product-Market Fit

Last week, I had a conversation with the founder of a funded AI startup. Not “funded” in the loose aspirational way people sometimes use the word, but actually funded: capital raised, team in motion, narrative assembled, runway secured. They were sharp, technically fluent, and clearly serious. This was not someone who had wandered into AI because the market was hot and they wanted to be near the action. They understood the tools, the language, and the moment…

But as we talked, the conversation drifted toward the only question that really matters for an early-stage company: who is the customer, and what problem are you solving for them that’s painful enough they’ll change their behavior and pay you for it?

The answer was vague... Not incoherent, exactly, but unfocused. There were several possible markets, several possible workflows, several possible buyer personas. There was a lot of language about agents, automation, productivity, knowledge work, and unlocking efficiency. There was much less clarity around the actual user, the budget owner, the current workaround, the urgency of the pain, or why this needed to be a company rather than a feature, a services layer, or a clever internal tool.

That was the moment the conversation crystallized for me. The issue was not that they lacked intelligence or effort. In some ways, that made it more revealing. This was a capable founder with real backing, actively searching for product-market fit after the company had already been financed. And that, more than any valuation chart or breathless funding headline, is what makes me think we are now somewhere near the top of the AI bubble.

The bubble is not that AI is fake. It obviously is not. The technology is remarkable, and anyone who works in strategy, finance, software, research, operations, or content and says they have not felt the productivity shift is either not using the tools well or not paying attention. AI will produce enormous companies. It will change workflows. It will compress labor. It will alter the structure of entire categories.

But “AI is real” and “this AI startup is a good business” are not the same statement. Bubbles form in the gap between those two ideas.

Share

Every serious bubble has a real technology at its center. Railroads were real. The internet was real. Cloud software was real. The mistake in those cycles was not believing in the underlying technological shift. The mistake was suspending normal business judgment because the shift was so obviously important. Investors stopped asking whether a specific company had durable economics, differentiated distribution, customer urgency, or a path to defensibility. The category, itself, became the diligence.

That is where, I think, we are now with a meaningful portion of AI. The market has started rewarding proximity to the platform shift more than evidence of customer pull. (After all, the trendy shoe company Allbirds “pivoted” to become an AI compute company and saw it’s stock skyrocket overnight.) According to this logic, if a company is “AI-native,” it, therefore, must be interesting. A founder has a demo, therefore there must be a product. A product touches a large enterprise function, therefore there must be a venture-scale market. But that chain of reasoning skips the hardest part of company formation: finding a specific, painful, frequent, budgeted problem that someone needs solved badly enough to pay for.

Product-market fit isn’t simply a vibe. It is not investor excitement, conference interest, a design partner, or a pilot with a friendly innovation team. It is not even a beautiful demo. It is not some vague sense that every company will need AI… Product-market fit is much more concrete and much less glamorous. It is a customer with a real problem, a current workaround, a budget, a sense of urgency, and enough dissatisfaction to overcome institutional inertia.

In healthier markets, capital tends to follow some version of that discovery process. A founder finds pain, builds around it, sells into it, learns through rejection, and earns the right to scale. In frothier markets, the sequence reverses. The founder starts with a technology wave, builds a narrative around it, raises capital, and then goes looking for the market. The company exists before the problem has been proven.

That inversion is the tell.

There is, of course, a defensible version of funding exploration. Venture capital exists to finance uncertainty. If everything about a company were obvious, the return profile would collapse. Some of the best startups begin with a technical insight before the market is fully legible... But there is an important difference between uncertainty around execution and uncertainty around existence.

Execution uncertainty sounds like: we know the customer has this problem, but we are not yet sure which widget scales fastest. Existence uncertainty sounds like: we have a powerful AI capability, but we are not yet sure who urgently needs it. The first is normal venture risk. The second is often narrative risk disguised as venture risk.

That distinction matters because access to capital can delay truth. A bootstrapped company with no customers gets humbled very quickly. A funded company with no customers can keep moving for a while… It can generate meetings, advisors, pilots, conference appearances, press mentions, and investor updates. It can confuse activity with validation. It can spend twelve months mistaking curiosity for demand.

And AI is uniquely good at generating curiosity. Every executive wants to know what AI might do to their business, and every department has some manual workflow that looks “automatable” in theory. Every company has some knowledge trapped inside documents, emails, spreadsheets, ticketing systems, CRMs, call transcripts, and Slack threads. A halfway decent AI demo can make almost any serious operator lean forward.

But generating interest is not the same as generating demand. Demand is when someone has a problem painful enough to fight procurement, security review, budget constraints, internal politics, competing priorities, and the general human preference for doing things “the way they’ve always been done.” Demand is when the customer does not merely admire the product, but reorganizes his or her behavior around it.

This is where a lot of AI startups are starting to struggle. They will not fail because the technology is bad. They’ll fail because the customer pain is too diffuse, the buyer is too unclear, the workflow is not owned, the ROI is too hard to measure, or the incumbent system of record can absorb the product as a feature update. They will fail because the market wanted “AI for X,” but the customer needed a specific operational outcome.

“AI for legal” is not a strategy; “AI for healthcare” is not a strategy; “AI for finance,” “AI for sales,” “AI for compliance,” and “AI for enterprise productivity” are not strategies. These are category labels. The strategy lives several layers deeper: whose workflow, which budget, what frequency, what error tolerance, what current workaround, what switching cost, what measurable economic value, what reason to buy, and buy now?

The more broadly available the underlying models become, the more important those questions are. If everyone has access to strong foundation models, then the model is not your moat. If every incumbent can add AI features, then “we use AI” is not differentiation. If the customer can approximate the output with ChatGPT, Claude, Gemini, Copilot, or an internal automation, then the startup needs to be more than a polished interface wrapped around temporary model capability.

It needs workflow ownership, distribution, proprietary context or data access. It needs to become embedded in a process that matters to the end user. And it needs to produce measurable value that survives once all the novelty wears off.

The hard part is that AI has made building look deceptively easy. A small team can now create in weeks what might have required a much larger engineering effort a few years ago. Demos are faster, interfaces are cleaner, prototypes are more impressive, and the gap between idea and artifact has collapsed. That is genuinely exciting. It also means that the bottleneck has shifted from “can you build it?” to “should this even exist?”

The latter is a much harder question. It requires taste, customer intimacy, commercial discipline, and the willingness to abandon attractive abstractions in favor of ugly specifics. The best AI companies I see are not the ones with the grandest horizontal claims. They are the ones with narrow insight into a real, painful workflow. They know the spreadsheet, the email chain, or the compliance bottleneck. They understand the untold analyst hours devoted to a workaround. They know the offshore team doing manual reviews in India. They know the exact point where the current process breaks.

That kind of specificity is not as exciting in a pitch deck, but it is where willingness to pay usually lives.

The companies that interest me the most are not “reinventing enterprise work with AI.” They are reducing denial-management time for a specific class of healthcare provider. They’re automating a reconciliation process where auditability matters. Or turning a messy compliance review into a repeatable workflow with clear ownership and risk thresholds. They are helping a buyer do something expensive, painful, and frequent in a way that is materially better than the status quo ante.

That may sound less ambitious, but it is not. Most large companies begin with a painfully specific wedge issue before a prototype. The grand ambitions come later, after that company earns its right to expand.

The current AI market often wants to skip that first part. It wants the platform before the wedge issue, the valuation before the validation, a cap table before a customer. That is what makes this moment feel late-cycle to me. Not because the technology has stopped being impressive, but because the funding environment has started to underwrite possibility as if it were proof of product-market-fit.

The founder I spoke with may still find the right target market, and I hope he does. Smart, persistent teams can iterate their way to real demand. But the broader signal is harder to ignore. When capable people are raising money first and searching for product-market fit second, the market is no longer merely financing innovation. It’s financing the hope that a sufficiently valuable problem will eventually materialize.

And hope, no matter how intelligent or well-funded, is not real product-market fit.

Alex Randall Kittredge writes the Substack ARK Strategy and is the founder and Managing Director of APR Strategic Consulting. He has advised 9+ CEOs, integrated workforces, and redesigned organizations across hedge funds, startups, and industrial companies. He is a graduate of Columbia University and the University of Cambridge, and mentors entrepreneurs & startup founders through Oxford Entrepreneurs Network, CamEntrepreneurs, and Plug and Play Tech Center. He is a Director of The Oxbridge Method Ltd, and is the author of the forthcoming book, How Your Side Hustles Will Save You: Creating a Durable Career that Transcends the Corporate Ladder.

Compliance Statement: All views and opinions expressed are solely my own and do not reflect those of any current or former employer, client, or organization with which I am or have been affiliated.