Apple Outsources Siri Cloud Workloads to Nvidia Blackwell: macOS 27 Release Notes Reveal Usage Caps Driven by Compute Costs
Apple's landmark September 2026 launch of its revamped Siri and Apple Intelligence suite marks a fundamental shift in how the company manages consumer AI workloads. Rather than processing all server-side queries on its own infrastructure, Apple is actively routing heavier Siri queries through Nvidia Blackwell B200 chips leased on Google Cloud.12 However, the immense operational cost of leasing this high-end external compute capacity has forced Apple to impose strict usage limits on its marquee AI features.
Official macOS 27 Release Notes Disclose Server-Side Caps
On September 9, 2026, five days prior to the official September 14 public rollout of macOS 27 Golden Gate and iOS 27, Apple quietly updated its release notes to reveal that its headline Apple Intelligence capabilities will be metered.
The disclosure, placed at the bottom of the release documentation, states:
"Certain Apple Intelligence features that rely on server-side models are subject to daily usage limits, including but not limited to Siri AI, Intelligent Photo Editing Tools, Image Playground, and AFM 3 Cloud models in Shortcuts.3"
These limits affect Apple's most heavily promoted features, including Siri AI (which pulls personal context and operates cross-app), Intelligent Photo Editing Tools (such as Spatial Reframing and Extend), and Shortcuts utilizing Apple's AFM 3 Cloud models.
The Blackwell-Google Cloud Connection and Inference Economics
The decision to meter these consumer features represents a direct response to the economics of frontier AI inference. Initially, Apple had committed to executing cloud workloads strictly on its proprietary, highly secure Private Cloud Compute (PCC) servers. However, following performance bottlenecks with modified Google Gemini models running on Apple's internal silicon, the company pivoted to leasing external Nvidia Blackwell B200 systems hosted on Google Cloud.
Because inference capacity at this scale is leased on a per-query basis and carries substantial marginal costs, Apple has a direct financial incentive to ration usage. This creates an ongoing operational challenge for Apple as it balances the promotion of its premium "Apple Intelligence" subscription/hardware ecosystem against the high, recurring infrastructure costs paid to external cloud providers.
-
An instance of Scaling metered agentic workloads demands strict budget guardrails and outsourced cloud infrastructure. — The richest consumer platform responds to metered AI costs exactly as the law predicts — usage caps plus outsourced cloud infrastructure. ↩︎
-
An instance of Rapid hardware depreciation turns proprietary data centers into a toxic liability for consumer platforms. — A consumer giant rents frontier compute rather than owning GPU infrastructure whose economic life decays within a few years. ↩︎
-
An instance of Per-query compute costs now ration the consumer AI features that run on them. — Apple rations its marquee consumer AI features because leased Blackwell inference carries substantial marginal cost per query — per-query compute, not capability, now sets the usage policy. ↩︎