There is a line in Apple’s fine print that most iPhone owners will never read, and it changes how you should think about everything Siri AI does for you.
Some Apple Intelligence features are metered. Not throttled when the network is busy, not degraded on older hardware. Metered, with a daily allowance, a timeout when you hit it, and eventually a price for going past it. Apple confirmed all of that in writing. What it has not confirmed is the only part anyone actually needs: how big the allowance is.
The short version
- Apple Intelligence features that rely on server-side models are subject to daily usage limits
- Affected features include Siri AI, Image Playground, cloud photo editing tools such as Clean Up, Extend and Spatial Reframing, and AFM 3 Cloud models in Shortcuts
- Hitting a limit does not break the phone. The feature times out and returns later
- Apple says increased access will be available for a fee, with the next generation arriving in the 2027 software releases
- Apple has published no limit size, no timeout length and no price
- At WWDC the company indicated higher iCloud+ tiers would include expanded access, which is a different answer from a standalone fee
What Apple actually wrote
The language appeared in Apple’s own release material alongside the September announcements, and it is unusually direct for a company that normally leaves this sort of thing to a support page nobody finds.
Apple lists the features that lean on server-side models: Siri AI, intelligent photo editing tools including Clean Up, Extend and Spatial Reframing, Image Playground, the AFM 3 Cloud and AFM 3 Cloud Pro models in Shortcuts, and developer apps that call Apple Foundation Models through Private Cloud Compute. The list is explicitly not exhaustive.
Then the mechanism: “In the future, Apple will implement usage limits on these and other Apple Intelligence and Siri AI features to ensure a high-quality, reliable experience for everyone. If you reach a limit in a particular feature, that feature will become available again after a time out period. Limits may vary by feature, request complexity, system demand, system policies, and other factors.”
And then the sentence that got the attention: “Increased access to such features will be available for a fee.”
Where the meter sits
The important distinction is between the model running on your phone and the model running in Apple’s data centers. Only one of them is being counted.
That split matters in practice. Summarizing a notification or running a quick text suggestion tends to stay on the phone. Generating an image, reframing a photo spatially or asking Siri AI something that needs real world knowledge does not. If you are a light user who mostly gets tidied-up notifications, you may never see a limit at all. If you spend an afternoon in Image Playground, you are the user this policy was written for.
What is confirmed, and what is guesswork
It is worth separating the two, because coverage of this has tended to blur them together and produce a scarier picture than the documents support.
The iCloud+ question is the genuinely interesting ambiguity. At WWDC, Apple indicated that higher iCloud+ tiers would come with expanded access to these features. The September fine print says increased access “will be available for a fee” without mentioning iCloud+ at all. Those are two different products. One is a storage subscription that quietly grows an AI allowance. The other is a line item you buy because you ran out of image generations.
Apple has form here for bundling rather than charging separately, and the recent pattern supports it. When Apple folded TV and Arcade into its cheapest iCloud plan, it was pretty clearly building iCloud+ into the place where every extra Apple service eventually lands. An AI allowance fits that shape.
Which features are metered
| Feature | Where it runs | Counts against the limit |
|---|---|---|
| Siri AI | Mostly Private Cloud Compute for knowledge queries | Yes |
| Image Playground | Server-side generation | Yes |
| Clean Up, Extend, Spatial Reframing | Cloud photo editing models | Yes |
| AFM 3 Cloud and Cloud Pro in Shortcuts | Private Cloud Compute | Yes |
| Third-party apps using Apple Foundation Models | Private Cloud Compute | Yes |
| On-device writing, summaries, suggestions | Local model on your iPhone | No |
Apple’s phrasing, “including but not limited to,” means this list can grow without a new announcement. That is the part worth watching. A limit you never reach is a non-event, but a list that quietly expands into the features you use daily is a different product than the one you bought.
The privacy side has a receipt
Buried in the same set of releases is a feature almost nobody has switched on: the Apple Intelligence Report. It shows which requests your device has sent to Private Cloud Compute, which is effectively a log of every time the meter ticked.
Apple’s position on what happens to that data is unusually specific. “Private Cloud Compute uses larger, server-based models,” the company says. “The data being processed in Private Cloud Compute is not stored or made accessible to Apple. It is processed only to fulfill your request, after which point the results are returned securely to your device and are not retained by Private Cloud Compute. And independent privacy and security researchers can verify this privacy promise at any time.”
The verifiability claim is the substantive one, and it is the part that distinguishes this from a marketing page. It also creates a small tension worth noticing: Apple is counting your requests in order to enforce a limit, while stating that the content of those requests is not retained. Counting and retaining are different operations, and nothing in the wording contradicts itself, but it does mean there is a usage ledger of some kind attached to your account.
How this compares to everyone else
Rate limits on AI features are not new. They are the normal condition of every competing assistant, most of which have been open about their tiers from the start. What is unusual is the sequencing. Other companies published the number and then charged for more. Apple has published the fee and withheld the number.
That ordering is probably deliberate rather than sinister. Apple has spent two years managing expectations about Siri, and the new assistant arrived behind a waitlist rather than landing on every eligible device at once. If you are still gating access by waitlist, you do not yet know what steady-state demand looks like, and a published limit becomes a promise you might have to break.
What this means for you right now
- Nothing is being billed today. The fee language is forward looking, pointed at the 2027 software releases
- Check the Apple Intelligence Report. It is the only view you have of how often your phone is actually calling the cloud
- Heavy Image Playground use is the likeliest way to find a wall. Generation is the most expensive thing in the list
- A timeout is not a lockout. One feature pauses, the rest of Apple Intelligence keeps working
- Watch your iCloud+ tier, not the App Store. That is the most likely place this shows up as a charge
What to take from this
- The limits are confirmed, the numbers are not. Apple committed to metering in writing and left every quantity blank
- Only cloud requests count. On-device Apple Intelligence work is unaffected
- The feature list can expand. Apple’s own wording is “including but not limited to”
- The fee has two possible shapes. A standalone charge, or expanded access folded into higher iCloud+ tiers, and Apple has gestured at both
- Timeline points at 2027. The next generation of Apple Intelligence is where this becomes a thing you pay for
The honest read is that Apple is doing what everyone running inference at this scale has to do, and being more candid about it than most in advance. The awkwardness is purely in the gap. Telling customers there is a meter, telling them it will eventually cost money, and declining to say what it measures leaves people to imagine the worst possible number. Apple will publish it eventually. Until then, the safest assumption is that the free allowance is sized for someone who uses Siri a few times a day, not for someone who discovered Image Playground on a Sunday afternoon.

