Apple handed every qualifying iOS developer free AI inference at WWDC 2026. The 2M download ceiling is not a size limit. It’s a timer set to run out at exactly the wrong moment.

What Apple actually shipped

At WWDC 2026, Apple opened Private Cloud Compute access to developers enrolled in the App Store Small Business Program whose apps have fewer than two million first-time downloads. No API key, no usage bill, no per-token pricing. The model on the other end is AFM 3 Cloud: 32,000 token context window, three levels of reasoning (light, moderate, deep), and image understanding. One-line switch from on-device to cloud inference.

Fifty thousand prompts per month at standard 2026 API rates runs between $30 and $150, depending on provider. On Apple’s Private Cloud Compute with a qualifying account: $0.

Apple also shipped a unified LanguageModelSession API. Switching between Apple’s on-device model and Private Cloud Compute is one line of code. Third-party providers — Claude, Gemini, others — plug in via the same LanguageModel protocol without touching session code. You write the integration once. You change the provider by swapping a dependency, not by rewriting your stack.

The number you should be looking at

Two million downloads sounds like a ceiling only big apps hit. It isn’t. That threshold is roughly where a consumer iOS product starts generating meaningful revenue — where you have App Store traction, real users, maybe some press. It’s the point at which you’d need to scale your AI usage and have the ability to pay for it.

Apple has not disclosed what the pricing structure looks like once you cross that line. The WWDC documentation describes eligibility for the free tier but says nothing about the paid tier: its cost, its rate limits, or whether access requires a separate agreement. You are making an architecture decision before you know the cost structure that kicks in when it actually matters.

AWS did this with Lambda free tiers. Stripe did it with volume-based pricing. The playbook is consistent: acquire during the zero-cost phase, price after you’re embedded. The question is whether you’ve built the escape hatch before the invoice arrives.

The SiriKit deadline that isn’t actually 2028

SiriKit is formally deprecated as of WWDC 2026. Apple has given it a roughly two-to-three-year support window before it’s removed entirely, which most coverage filed under “2028 problem.” Wrong calendar.

iOS 27 ships this autumn with a rebuilt Siri powered by Gemini, orchestrating cross-app actions via App Intents. If your app is still on SiriKit when iOS 27 lands, it is invisible to that orchestration. Users on iOS 27 get a degraded or missing experience with your app while competitors who have migrated get surfaced.

The migration effort for most apps is measured in hours, not sprints. Xcode 27 has a Convert to App Intent path. This is not a question of whether to migrate. It’s a question of whether you do it before autumn or after you’ve already lost that surface area.

What the build-vs-buy shift actually looks like

Most people read Apple’s WWDC news as “build on Apple” becoming more compelling. They missed the better news: Apple just made the case for “build against a neutral protocol” obvious.

Because LanguageModelSession is a provider-agnostic interface, the right architecture isn’t “use Apple’s free tier.” It’s “write to the protocol, start with Apple’s free tier, and keep the door open.” The lock-in risk isn’t architectural — Apple’s own API makes switching cheap. The risk is the same one it’s always been: building a feature set your users depend on before you know what the infrastructure behind it will cost.

For an iOS founder at pre-seed or seed, the practical position is straightforward. Use the free tier. It’s real money not spent on API calls during the phase where every dollar of burn matters. Build against the LanguageModelSession abstraction so the escape hatch exists when you need it. Get your App Intents migration done before iOS 27 ships — the effort is modest and the cost of skipping it is losing Siri visibility in your most competitive growth period. Watch for when Apple discloses post-threshold pricing, because that disclosure will tell you more about their developer strategy than any keynote slide.

The build-vs-buy question for iOS isn’t “Apple or OpenAI.” It’s “free-until-you-have-traction or pay-as-you-go from day one.” Neither answer is obviously wrong. The error is taking the free tier uncritically and finding out what it costs after you’ve built your users into depending on it.

If you’re building an iOS product and trying to map the right infrastructure decisions before you hit traction, talk to us.

Sources