Apple announced at its annual developer conference that small app developers can now use its cloud-based AI models for free. The condition: you must be in Apple's Small Business Program, and your app must have fewer than 2 million total first-time downloads.
The on-device part has already been free since launch. What changed is the cloud side. When a task is too complex for a phone to handle alone, it gets routed to Apple's private cloud servers. That used to mean a bill. Now, for qualifying developers, it does not.
To qualify for the program, a developer also needs to earn under $1 million per year from their apps. That covers the vast majority of people building on Apple's platforms.
The timing is not accidental. Right now, AI running costs have become a real problem even at companies with enormous budgets. Uber burned through its entire 2026 AI budget in just four months, after encouraging staff to use AI tools as heavily as possible. Its COO said it was "very hard to draw a line" between rising AI costs and useful features actually reaching customers.
Amazon ran into the same wall. Its internal leaderboard ranked employees by how many AI tasks they completed. Workers responded by instructing AI tools to run pointless jobs just to improve their scores, a practice that spiked costs with no real output. Amazon shut the leaderboard down. Meta killed a similar system that had tracked token usage across 85,000 employees.
This pattern matters for any business operator watching AI adoption. The lesson from large companies is that measuring AI use by volume, how much you spend, how many tasks you run, leads to waste. The useful question is what actually got done.
Apple's move acknowledges that cost itself is a barrier to useful experimentation. A solo developer or a small team building an app cannot absorb a variable cloud bill that grows with every user interaction. By removing that cost, Apple gives them a real reason to try things that might fail, which is how most useful software gets built.
The models themselves work on text tasks: summaries, content extraction, writing assistance, simple dialogue. They run on Apple devices and, for more complex requests, on Apple's private cloud servers, where Apple says user data is not stored or shared, not even with Apple itself.
The framework also now supports images alongside text, and it connects to other AI providers, including Claude and Gemini, when a developer needs something more powerful. Apple absorbs the free tier; developers pay only if they choose to plug in a third-party model for heavier work.
For business operators, the relevant takeaway is not technical. It is that AI running costs are becoming a serious management problem across industries, even for the most resourced companies in the world. The companies hitting budget walls are the ones that pushed usage without first defining what success looks like. Apple's free tier for small developers is a product decision. But the industry-wide budget crunch it sits inside is something any operator with AI tools in their business needs to plan around.