Apple will let developers with fewer than 2 million first-time App Store downloads run its Foundation Models inside Private Cloud Compute with no API cost, the company announced at the Worldwide Developers Conference keynote on Monday. The move zeroes out one of the biggest line items for indie developers experimenting with AI features, and it positions Apple as a cheaper on-ramp than the hyperscaler clouds. The 2 million threshold mirrors the eligibility cutoff Apple uses for its Small Business Program, which lowers App Store commission rates for developers who aren't yet earning millions.
The framework is also opening up. Apple said the Foundation Models framework is expanding in 2026 to accept image input and to plug into third-party server models, meaning developers can route more complex tasks to the cloud provider of their choice through the same API. That's a meaningful concession from a company that historically prefers to keep its stack closed.
The pitch is cost. Frontier-model inference has stopped being something developers can casually burn through, and Apple is betting that a free private-cloud tier will pull a long tail of small studios into its ecosystem before they commit to Anthropic, OpenAI, or Google as a default. For an indie shipping a side-project app, the difference between a $0 monthly API bill and a multi-hundred-dollar one decides whether the feature ships at all.
Key facts
- 01Developers with fewer than 2 million first-time App Store downloads get free access to Apple's Foundation Models running in Private Cloud Compute.
- 02The Foundation Models framework expands in 2026 to include image input and integration with third-party server models.
- 03Uber said it burned through its full 2026 AI budget in four months, a sign of how fast inference costs are escalating.
- 04Meta and Amazon have quietly retired internal AI token-usage leaderboards that once rewarded developers for heavy experimentation.
The broader industry is tightening, not loosening. Meta and Amazon have both discontinued the internal AI token-usage leaderboards that, until recently, gamified how aggressively their engineers experimented with model calls. The leaderboards encouraged spend; the spend got noticed; the leaderboards came down.
Uber said it ran through its entire 2026 AI budget in four months — a four-month burn rate against a twelve-month plan. Whatever Uber forecasted last year, real usage came in at roughly three times that pace. That single data point has rippled through finance teams trying to model AI line items into next year's budget.
At the top of the stack, costs are scaling in the other direction. Google has agreed to pay SpaceX $920M per month for compute, a contract size that underlines how the largest AI consumers are now locking in capacity at the level of national infrastructure deals. The gap between what hyperscalers can absorb and what an indie developer can absorb has never been wider.
That gap is the opening Apple is targeting. By eating the inference cost for smaller developers on-device and in Private Cloud Compute, Apple keeps the lights-on number close to zero for the demographic most likely to abandon AI features when the bill arrives. Apple keeps the data inside its privacy envelope, and it keeps the developer inside its ecosystem when that small app eventually grows.
There are limits worth naming. The free tier caps at 2 million first-time downloads, which means a hit app graduates out of the program and onto a paid arrangement — Apple has not detailed what the post-threshold pricing looks like. The Foundation Models themselves are also Apple's own; the third-party server-model integration handles the more complex tasks, and developers will still pay those vendors directly. The free lane is for Apple-hosted inference, not a universal AI subsidy.
The interesting question is whether this becomes a template. If Apple can credibly underwrite frontier-tier inference for the small-developer tier and still hit its margins, it pressures every cloud-model vendor selling into the indie segment to justify why their token bill should exist at all. Anthropic and OpenAI built their developer businesses on the assumption that infrastructure cost is a fixed feature of the market. Apple just said, for a defined slice of that market, it doesn't have to be.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
