Apple's Free AI APIs Expire at 2 Million Downloads

Apple announced at WWDC that developers with fewer than 2 million first-time App Store downloads can use its cloud AI APIs without paying. On the surface, this looks like a gift to indie builders. Apple is effectively saying: do not worry about the compute bill, just build.

But the math is more interesting than the marketing. Two million downloads is a large number for most indie apps, but it is also a hard cliff. The moment you cross that threshold, you start paying full price for the AI infrastructure that your users now expect. Apple is not giving you free tools forever. It is giving you free tools until you are too big to leave.

The 2 Million Download Trap

This is classic platform economics. Get developers hooked on a subsidized stack, then flip the pricing switch once switching costs are high. The 2 million download cap is generous enough to feel safe, but low enough to catch most successful apps before they have the cash reserves to absorb a sudden AI bill.

The real problem is not the cliff itself. It is the architecture you build while the ride is free. If your app relies on Apple's cloud AI for core features, you are baking in a dependency that gets expensive exactly when your app starts working. That is a terrible time to re-engineer your backend.

We have seen this movie before. Startups build on subsidized Firebase credits, then hit the usage tier and discover their burn rate doubles overnight. They build on Stripe's free tier, then scale and realize payment processing is their second-biggest expense. Free infrastructure is never free. It is a loan against future revenue.

Build for the Bill Before It Arrives

The smart move is to treat Apple's free tier as a prototyping bonus, not a foundation. Build your first version fast, validate the idea, and get to market. But if you see traction coming, start planning your exit from the subsidized stack before you need it.

This is where your choice of backend matters more than your choice of front-end framework. A reactive database with built-in vector search, real-time queries, and serverless execution gives you AI-native infrastructure without the per-request tax of a closed cloud API. You want a backend that scales with your user count, not your API call volume.

Botflow runs on Convex for exactly this reason. The database is reactive, workflows are durable, and vector search is built in. You are not pinging an external AI cloud for every user action and praying the rate limiter is kind. Your app runs on infrastructure you control, with costs that grow linearly instead of exponentially.

That matters when you go from ten thousand users to ten million. You do not want to discover at user 2,000,001 that your AI feature costs two cents per call and your app makes one cent per user.

Read the Fine Print

Apple's move is a signal, not a strategy. Big tech knows that AI experimentation is getting expensive, and they are racing to own the developers who are still small enough to catch. If you are building something real, take the free credits, but keep your architecture portable. The builders who survive the next phase will be the ones who built for the bill before it arrived.