One App Store fee waiver is changing where small apps get to start their AI experiments.

Uber burned through its entire 2026 AI budget in four months. Meta and Amazon have shut down internal leaderboards that tracked employee AI token usage. Even as the biggest companies in the world struggle to control the cost of AI experimentation, Apple quietly pointed in a different direction with an announcement in early June aimed at small-scale app developers: if your app's cumulative new downloads haven't crossed two million, you won't pay a cent for cloud API costs when running AI on Apple's servers. The cost calculator that indie developers would reach for first whenever they wanted to add an AI feature — for now, that spreadsheet disappears. It's worth examining exactly what this new threshold changes, and what it doesn't.

Under 2 Million New Users? No Server-Side AI Costs

At WWDC this year, Apple formally unveiled the Foundation Models framework. Through it, developers can tap two distinct paths: on-device AI models running locally, and server-side Private Cloud Compute. On-device inference already came at no extra cost. What's new is the server-side path. When a device lacks the horsepower for a given task or heavier computation is needed, requests get routed to Apple's servers — and it's the cost of that server-side route that Apple is now waiving for small developers. The framework also debuted image input support, expanded server-model options, and broader integration with third-party cloud providers of the developer's choosing.

The eligibility threshold is simple. The cumulative number of first-time downloads across a developer's apps must stay under two million. Re-downloads and updates don't count. Among indie developers on the Korean App Store, most apps attract somewhere between tens of thousands and a few hundred thousand new users. By that measure, virtually every indie developer and small team in Korea falls within the benefit.

There are several ways to read the timing. As the experience gap between AI-enabled and AI-free apps grows increasingly visible, if small developers keep delaying AI adoption because of cost, the quality distribution across the App Store starts to look lopsided. Apple's interest in keeping developers inside its own ecosystem is also hard to separate from the announcement. The company itself drew a parallel to its Small Business Program, suggesting that two goals — sustaining ecosystem quality and attracting developers — are working in tandem here.

Remove the Cost and AI Moves Inside the MVP

Adding AI to an app has always meant running cost calculations at every step: which model to use, how often to call it, what the per-token rate would be. Only after that math could development actually begin. For an early-stage app that isn't yet generating revenue, that calculation becomes a reason to push the experiment back. Many indie developers have chosen to delay AI to a post-MVP phase, citing the unpredictability of API costs.

Remove the cost, and that sequence reverses. Ship the feature first, watch how users respond, then revisit cost design at monetization time — that becomes a viable path. It means launching a product with AI built in from day one. The difference shows up not just in time-to-market but in the quality of feedback. When user data starts accumulating with AI already present, the features worth strengthening in the next version become far more concrete. An app that launches without AI and adds it later is forced to graft new functionality onto habits users have already formed.

The server-side path also compensates for the limitations of on-device models. Device performance varies, and on older hardware AI features can run slowly or not at all. With server-side access available at no cost, a developer can deliver a consistent AI experience regardless of the user's device generation — and do it with minimal upfront resources. The era where the vintage of a user's phone determines the quality of the app experience starts to shrink.

When the Free Period Ends, Platform Lock-In Stays

That said, this benefit is not an unqualified windfall.

The Foundation Models framework only works fully within Apple's ecosystem. Developers who also maintain Android apps or web services still have to absorb separate AI API costs on those platforms. Many small Korean developers run an iOS app alongside an Android version, or operate web-based services in parallel. For any team pursuing a multi-platform strategy, this benefit applies to only half the picture.

How long the waiver will last is also unclear. The two-million threshold could be revised, or the policy could shift to charging above a certain usage level. Developers who build an app deeply dependent on AI during the free period could face significant switching costs if pricing kicks in later — and the more AI logic is woven through an app's architecture, the harder it is to migrate.

At a broader level, critics have pointed out that this policy may simply be a mechanism to bind developers more tightly to Apple's platform. The lower the entry cost, the higher the platform dependency. Starting is easier in the short run; leaving — or building an independent AI infrastructure — gets harder over time. If Uber's four-month budget burn was a product of unchecked experimentation, then rushing into unlimited AI experimentation simply because it's free, and waking up to deep lock-in later, isn't so different. The fact that a zero-cost condition can return later in the form of platform dependency is something small developers in particular need to think through from the start.

What Indie Developers Should Check First

There are practical things a solo developer or small team can do to assess what this change actually means for them.

The App Units metric under the Analytics tab in App Store Connect shows new download counts. For most indie developers, two million is a distant number — meaning current and prospective small-scale developers are very likely to qualify. Even so, verifying the exact exemption conditions in Apple's official API documentation comes first. The specifics of a policy condition can shift between a public announcement and the actual documentation.

It's also worth determining in advance whether the AI features you want to build fall within what the Foundation Models framework actually supports. It handles text summarization, classification, tagging, and short-form generation well — but complex reasoning or high-quality language generation may still require pairing it with an external API. Assuming this framework can cover every AI use case is a mistake worth avoiding early.

Designing for platform dependency management from the start is also worth the effort. Keeping AI call logic in a layer separate from core app logic reduces the cost of switching later, whether because Apple's policy changes or a different API becomes preferable. Use the benefit — just don't get welded to a single API.

Developers who have kept building under constrained resources tend to work this way already. They don't wait for perfect conditions — they use what's available now and move forward incrementally. Whatever the waiver's lifespan, the user relationships and feedback accumulated during that time stay with the developer. That's what Apple can't take back when the terms change.