One App Store fee waiver is shifting where small apps begin experimenting with AI.
Uber burned through its entire 2026 AI budget in four months. Meta and Amazon scrapped internal leaderboards that tracked employees' AI token usage. Even at companies that size, controlling the cost of AI experimentation has proven difficult — which makes the move Apple made for small app developers in early June point in the opposite direction. Developers whose apps haven't crossed 2 million cumulative new downloads on the App Store no longer have to pay for the cloud API costs of running AI on Apple's servers. For a solo developer weighing whether to add an AI feature, the cost spreadsheet that used to be the first thing they opened just disappeared, at least for now. It's worth looking at exactly what this threshold changes — and what it doesn't.
Under 2 Million New Users, Server-Side AI Is Free
At this year's WWDC, Apple officially unveiled its Foundation Models framework. It gives developers access to two paths: on-device AI models and a server-side option called Private Cloud Compute. Processing that happens on-device was already free. What changed is the server-side path. When a device doesn't have enough horsepower, or a task requires heavier computation, requests get routed to Apple's servers — and the core of this announcement is that Apple will now waive the costs on that path for small developers. The framework also picked up image-input support and expanded server models this round, and Apple widened the door for developers to plug in outside cloud providers of their choosing.
The threshold is simple: as long as a developer's app hasn't been downloaded for the first time by more than 2 million cumulative users, the exemption applies. Re-downloads and updates don't count. Most indie apps on Korea's App Store see new-user numbers in the tens or low hundreds of thousands, so applying this threshold as written, nearly every indie developer and small team in Korea falls inside the benefit.
There's more than one way to read the timing. The user-experience gap between apps with AI and apps without it is widening visibly, and if small developers keep delaying AI adoption purely over cost and get left behind, quality across the App Store as a whole becomes uneven. It's also hard to separate this from Apple's interest in keeping its developer ecosystem tied to its own platform. Apple has framed the move as being in the same spirit as its Small Business Program, which suggests two goals are running in parallel: maintaining ecosystem quality and keeping developers engaged.
Remove the Cost, and Experimentation Moves Into the MVP
Bolting AI onto an app has always meant running the numbers at every step: which model to use, how often to call it, what the per-token rate comes out to. Only after that math was done could development actually start. For an early-stage app with no revenue yet, that math is exactly what pushes experimentation down the road. Plenty of indie developers have pushed AI past the MVP stage for one reason: they wanted the feature, but couldn't predict what the API bill would look like.
Remove the cost, and that order flips. Now it's possible to ship the feature first, watch how users respond, and only redesign the cost structure once the app starts making money. In other words, a product can launch with AI baked in from day one. The difference shows up not just in how fast something ships, but in the quality of the feedback. Once user data starts accumulating with AI already in the mix, exactly which features to strengthen in the next version becomes far more concrete. An app that launches without AI and bolts it on later has to force AI into user habits that have already set.
The server-side path also covers for the limits of on-device models. Because performance varies device to device, AI features can run slowly or not at all on older hardware. With the server-side path free to use, a small team can build an app that delivers a consistent AI experience regardless of device specs, without spending much on upfront infrastructure. The generation of hardware a user happens to own matters less for app quality than it used to.
When the Free Ride Ends, Platform Dependence Remains
Still, it would be a stretch to call this an unqualified win.
The Foundation Models framework only works fully within Apple's own ecosystem. Developers who also run an Android app or a web service still have to pay separate AI API costs on those other platforms. Plenty of small Korean developers maintain an Android version alongside their iOS app, or run a web-based service in parallel. For a team with a multi-platform strategy, this benefit really only covers about half the picture.
How long the exemption will last is also unclear. Apple could always move the 2 million threshold or start charging past some usage level. If a developer builds an app that leans heavily on AI during the free period and costs kick in later, switching to a different API is no small task — and the more deeply AI is woven into the app's logic, the higher that switching cost gets.
Zooming out, there's also criticism that this policy is really a tool for tying developers more tightly to Apple's ecosystem. The lower the cost of entry, the higher the platform dependence tends to run. Getting started gets easier in the short run, but building independent AI infrastructure or migrating to another platform down the line could get harder. If Uber's four-month AI budget burn was the result of experimenting without discipline, then running unlimited experiments purely because they're free — and ending up locked in later — isn't so different. What looks like zero cost now can come back as platform lock-in later, and that's exactly the kind of thing smaller developers need to weigh from the start.
What Korean Indie Developers Should Check First
It's worth spelling out what solo Korean developers and small teams can actually take away from this change.
New-download counts are visible under App Units in the Analytics tab of an App Store developer account. For most indie developers, 2 million is still a distant number, so anyone running an app now, or launching a small one, is likely to qualify. That said, the first step should be checking exactly how the exemption is spelled out in the official API documentation — the fine print sometimes ends up differing from the public announcement.
It's also worth checking upfront whether the AI feature you want to build actually falls within what the Foundation Models framework supports. It's plenty for tasks like text summarization, classification, tagging, or generating short sentences, but features that need complex reasoning or high-quality language generation may still require pairing it with an outside API. Don't assume this one framework can handle every AI use case.
It also helps to design around platform dependence from the start. Keeping AI calls in a layer separate from the app's core logic lowers the cost of switching later, whether that's because the policy changes or you need to move to a different API. The idea is to take advantage of the benefit now while designing so you're not locked into any one API.
People who've kept building steadily even without much in the way of resources tend to work this way: rather than waiting for perfect conditions, they use whatever's available now and move forward incrementally. However long the fee waiver lasts, the experience and feedback gathered with users during that window stays with the developer. That's the part Apple can't take back, no matter how the terms change.
