Alexander Kardos-Nyheim, an angel investor who sold his own AI startup before hitting revenue, argues that the real wealth creation in artificial intelligence flows toward infrastructure and model-layer companies, not the crowded application layer built atop existing platforms like OpenAI's GPT.
His thesis challenges the startup gold rush toward ChatGPT wrappers and consumer-facing AI tools. Most founders chase quick market entry by layering features onto existing large language models. Investors chase them for the same reason. Speed beats depth. But Kardos-Nyheim contends this race misses where genuine defensibility and long-term value live: solving fundamental technical problems at the model and infrastructure level.
Kardos-Nyheim shares his investment framework for evaluating AI startups. He focuses on technical depth. Can the founding team actually solve hard problems in model architecture, training efficiency, or infrastructure? Or are they assembling shallow products that any competent engineer can build in weeks?
The distinction matters for exit potential and investor returns. Application-layer startups face brutal competition and razor-thin margins. Infrastructure and model-layer companies control the foundational technology that powers dozens of applications downstream. They command pricing power and defensibility.
This perspective carries weight. The investor sold his own AI startup before generating revenue, suggesting he recognized the vulnerability of his position relative to better-capitalized competitors with deeper technical moats. The exit came from understanding where value actually lives, not from chasing vanity metrics like user growth or ARR.
For founders, Kardos-Nyheim's argument translates to a hard question: Are you building a feature or a platform? Are you solving a problem that only your team can solve, or something any competent engineer with API access can replicate? Investors increasingly ask the same thing. The hype cycle for application-layer AI startups has already peaked. Funding flows