He sold the company without a single dollar in revenue. That's Alexander Kardos-Nyheim's story. He founded an AI startup, completed the exit before the business ever generated meaningful revenue, and then moved to the other side of the table — evaluating other AI startups. He's since written up the criteria he drew from that experience.
How do you sell a company with no revenue? His answer: the acquirer wasn't buying the product. They were buying the defensibility of the technology. That distinction leaves founders with one uncomfortable question: Am I layering an app on top of AI, or am I building AI itself?
The Line Investors Draw: App or Infrastructure?
The first question Kardos-Nyheim asks when evaluating an AI startup is blunt: "Is your product a layer on top of an existing AI platform, or does it engage with AI at a deeper level?" The logic behind it is clear. A service built by calling APIs from OpenAI, Google, or Anthropic gets replaced the moment any of those platforms extends its own feature set.
The AI startups he assigns long-term value to operate at a different layer entirely — teams solving hard technical problems in model training, infrastructure design, or novel reasoning architectures. These companies don't wobble when the platform shifts. If anything, the platform eventually needs what they've built.
This distinction became a primary variable in investment evaluation over the past two to three years. As AI has rapidly commoditized, the barrier to "using AI" has nearly vanished. Anyone can call an API and ship a service. Paradoxically, that also means the defensibility of services built on top of those APIs has collapsed in tandem — any competitor can build the same thing the same way. The gap between the value platforms capture from below and the value services can hold from above keeps widening.
The Counterargument: This Logic Only Applies to Well-Funded Teams
There's a hidden assumption in Kardos-Nyheim's framework: enough technical talent, enough capital, and enough runway to survive without revenue. For a solo founder or a lean team, building a company that operates at the model-training or inference-infrastructure layer isn't a realistic starting point. Read "if it's not foundational AI, it has no value" too literally, and most small founders arrive at the conclusion that nothing they do can work.
To be fair, the audience this kind of analysis targets is teams that have raised venture-scale funding or are actively exploring angel rounds. Independent creators and small operators can't simply follow a prescription to "build foundational technology" — and trying to might pull them further from their actual strengths. If every founder must build core AI, this framework has little practical relevance for smaller builders.
But stopping there sells the insight short. Buried in Kardos-Nyheim's evaluation question is a layer that scales down to any business size: "What is irreplaceable about what I'm building?"
Translating Investor-Speak Into Founder Language
Across documented accounts of investment decisions by a dozen venture capitalists, a consistent pattern emerges: investors buy defensibility over scale. Before team size or revenue numbers, they ask whether this service could exist without this specific team. Conversely, when the answer to "Could another team replicate this in three months?" is yes — that business is defensible by nobody's standard, regardless of size.
Bring that question into the context of a solo founder or small operator, and a few useful self-diagnostic checkpoints emerge.
If ChatGPT or Claude added the same functionality my service offers, would my service disappear? A tool built primarily for summarization, translation, or text generation can be made redundant by a single platform update. If the answer is yes, it's worth looking hard at what you're actually accumulating inside the service.
Do the data, relationships, and community I've built create value independently of AI? If so, AI is an efficiency layer on top of something that already stands on its own. That structure survives platform changes. When customers have a reason to return that has nothing to do with AI, the AI's job is to cut costs or accelerate speed — not to serve as the foundation.
Am I operating in a domain AI still can't reliably solve? On-the-ground industry knowledge, context gathered from years of customer interviews, long-cultivated professional relationships — these are things no model can scrape or train on. Anyone generating that kind of proprietary data or maintaining those relationships directly builds defensibility without needing foundational technology at all.
Those three checkpoints compress into a single question: Could this business exist without AI, or does it only exist because of AI? In the first case, AI is a tool that makes things more efficient. In the second, AI is the skeleton of the business. What Kardos-Nyheim looks for as an investor is the team that fits the second description — and that same yardstick translates directly into a founder's self-diagnostic.
Kardos-Nyheim was able to sell a company with zero revenue because he convinced an acquirer that what the company had built couldn't be replicated. Defensibility sold before revenue did. Most founders aren't positioned to build foundational AI, and don't need to be. But the question — "What about my business is genuinely irreplaceable?" — is one anyone can pick up right now, at any scale. As AI services flood the market, the founders who stop to think clearly about which layer they're standing on will be the ones still standing longest.
