AI Startups Risk Becoming Victims of Pricing Power
Trending Topics Jakob Steinschaden ● Covered by 2 sources
AI startups can look profitable until the token bill lands. If a model maker controls prices, the app layer can end up paying to serve everyone.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A lot of AI startups are discovering that they don’t really run software businesses. They run reseller businesses with nicer branding. The app feels like SaaS on the outside, but underneath it every answer, every summary and every coaching tip has a running cost attached to it.
The source article uses a fitness app to show the trap. A free app, a 9.99-euro premium tier, 100,000 users, and 5 percent paying sounds respectable. That gets you nearly 50,000 euros in monthly revenue. Then the fees show up: the app store takes about 15 percent, the AI coach costs around 3 euros per premium user in tokens, and even free users need a few answers each week. After that, the leftover margin is about 17 percent. The premium crowd is effectively funding its own service, the free crowd, and the supplier sitting upstream.
That is the big shift. Classic software had near-zero marginal cost, so more users usually meant better margins. AI flips that. Every request burns compute. The article cites a market study saying inference makes up 23 percent of revenue at scaling AI companies on average, with gross margins around 52 percent, far below the 78 to 80 percent common in classic SaaS. No wonder companies like Notion, GitHub Copilot and Zendesk are already moving, at least partly, toward usage-based pricing.
The deeper problem is who sets the bill. OpenAI and Anthropic are not just suppliers; they also build their own apps, coding tools, agents and browsers. That means they can change token prices, context windows, model availability and rate limits while startups sit there pretending switching is easy. It isn’t. The frontier model of one provider can only be matched by the product of another, which is a tidy way of saying the market is not exactly a buyer’s paradise.
There are ways out, but they’re expensive enough to scare off most founders. Cursor is held up as the example of a startup that tried to buy freedom by spending hundreds of millions on its own model infrastructure, before ending up tied to SpaceX and its Colossus cluster with about 200,000 Nvidia GPUs. The article’s point is blunt: real independence means chips, operations and power, plus the stomach to burn billions. For most startups in Vienna, Berlin or Paris, that is fantasy, not strategy. Neoclouds help with data location and compliance, but they don’t change who controls the economics.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI everyone likes to skip: if the model vendor can raise the rent, the app layer is just a nicely designed tenant. Europeans especially keep mistaking data sovereignty for economic sovereignty, which is adorable in the same way a paper lock is adorable. The smart founders are the ones treating token cost like a weapon pointed at their own margins, because that’s exactly what it is.
Read more about this at: Trending Topics
