Growth is not durability: reading an AI company in a hype cycle
While the feed argues about whether AI is a bubble, the useful question for an investor is narrower and older, is this company's advantage durable. Here is how to read one when valuations and noise are both high.
TL;DR
The bubble debate is loud and mostly unactionable. What an investor can act on is durability, whether a company gets stronger as models improve, owns something a competitor cannot copy, survives its provider dependency, and holds margin at scale. In a hype cycle, disciplined diligence is the edge, because the market is pricing stories and durability is what remains when the story ends.
Open any feed and the AI conversation has collapsed into one argument: bubble or not. A famous investor is shorting it. An analyst says it is seventeen times the dot-com frenzy. A central banker says this time there are real earnings. Trillion-dollar valuations arrive faster than anyone can absorb them, and capex has grown large enough to move national economic statistics.
It is a genuinely interesting debate. It is also, for someone deciding whether to write a specific check, almost useless. Whether “AI” in aggregate is overvalued tells you nothing about whether this company in front of you is a good investment. Markets can be frothy and still contain durable winners, exactly as the dot-com bubble did. The job is not to time the weather. It is to tell the durable companies from the ones that only look good while the sun is out.
That job has a name, and it is older than this cycle: diligence. Here is how to do it when both valuations and noise are running high.
Separate the market question from the company question
The bubble debate is a market question. It is about aggregate sentiment, capital flows, and macro capex, and it is mostly out of your control. The company question is narrower and answerable: does this specific business have a durable advantage. Conflating the two is how good investors talk themselves out of durable companies in a correction and into fragile ones in a boom.
Discipline here is simple to state. Decide the company question on its own evidence, then let your view of the market set your price and your pace, not your yes or no.
The four durability tests
Durability in an AI company comes down to four questions. None of them is about this quarter’s growth, because growth is the thing hype produces most easily and correction destroys fastest.
Does it get stronger as the models improve? The single best signal. Durable companies ride the model wave: each foundation-model release makes their product better, because they have built something that captures new capability and applies it to a problem they own. Fragile companies get swallowed by the same wave, because the capability they resold is now native to the platform. Ask the founder directly what a much stronger base model does to their business. The durable ones have thought about it constantly.
Does it own something a competitor cannot copy quickly? Proprietary data that compounds with use, distribution that is expensive to displace, workflow depth that took years to build. If the only asset is a clever prompt and a nice interface, both are visible and copyable, and there is no moat under the growth.
Does it survive its dependency? Most AI companies depend on one model provider. Awareness and optionality matter more than independence: do they know their exposure, have they tested alternatives, would a price or policy change be painful but survivable, or fatal.
Do the unit economics hold at scale? Many AI products are demos that cannot afford their own success. At low volume, inference cost is a rounding error; at scale it can eat the entire margin. Model it at ten and a hundred times current volume before the growth story and the profit story quietly diverge.
This is the Wrapper Test, and it is built for exactly this environment.
Why the noise makes diligence more valuable, not less
There is a temptation, when everyone is shouting bubble, to either freeze or to dismiss the whole category. Both are expensive. The correct response to loud, low-information noise is to raise the value you place on quiet, high-information work, which is what real diligence is.
When the market prices stories, the edge belongs to whoever can read past the story to the durable substance underneath. That is not a market call. It is a company-by-company reading of architecture, data, dependency, and margin, done by someone who can evaluate the technology as fluently as the spreadsheet. In a calm market that reading is worth something. In a hype cycle it is worth much more, because the gap between price and durability is where both the risk and the opportunity concentrate.
What this looks like in practice
Before the check: run the four tests, write the honest memo, and let the answer, not the mood, decide. Price the deal against your read of the market, but do not let the market’s mood decide the company question for you.
After the check, if you invest: the same four questions become the things you monitor. Dependency risk, margin at scale, and defensibility against the next model are not one-time diligence items. They are the health metrics of an AI company through a cycle.
The bubble will resolve itself the way these things always do, unevenly, with durable companies surviving and fragile ones not. You do not need to predict which way the aggregate goes. You need to know, company by company, which side of that line your money is on.
If you have a deal that needs this kind of read, that is what our AI diligence engagement is for.