Apple announced the next generation of Apple Intelligence this week, and the most consequential detail was not the features. It was a single clause near the bottom of the release. The new Apple Foundation Models that power everything announced — the rebuilt Siri, the photorealistic image generator, the agentic password tool, the onscreen-aware assistant — are, in Apple’s own words, custom-built in collaboration with Google and its Gemini models. The company that built its brand on doing everything in-house has outsourced the core.
This is the through-line of the announcement, and Apple frames it carefully. The models run on device and on Apple’s servers using Private Cloud Compute, the architecture Apple uses to argue that frontier-scale intelligence can reach the cloud without surrendering user data. The privacy story is intact on paper: when Private Cloud Compute handles a request, Apple says the personal data is neither stored nor accessible to Apple or anyone else, and outside experts can audit the claim. But the intelligence underneath that privacy wrapper now traces back to Mountain View. Apple is selling the container, not the engine.
The second Google fingerprint is harder to miss once you look. Every image generated in the new Image Playground and every photo edited with Apple Intelligence carries a hidden SynthID watermark to mark it as AI-touched. SynthID is Google DeepMind technology. Apple has adopted both its rival’s frontier models and its rival’s provenance standard in the same release. For a company that treats vertical integration as a moral position, this is a quiet reversal, and the press materials do not dwell on it.
Siri AI and the Rebrand of Failure
The headline product is Siri AI, which Apple calls an entirely new version of Siri. The naming is itself a tell. Apple does not append “AI” to product names lightly, and the move reads as an attempt to firewall the new assistant from the reputation of the old one — the personal-context Siri that was promised at the previous developer conference and never shipped on schedule. The capabilities described are the ones competitors have shipped for over a year: conversational back-and-forth, broad world knowledge pulled from the web, onscreen awareness, the ability to search across a user’s messages, emails, and photos, and systemwide app actions. There is a dedicated Siri app that syncs conversation history through iCloud, Visual Intelligence extended to iPad and Mac, and integrated Writing Tools. It is a capable assistant. It is also Apple catching up, with Google’s models doing the lifting.
The most advanced on-device model is gated to the newest silicon — iPhone Air, iPhone 17 Pro and Pro Max, M4 iPads, M3 Macs, and the M5 Vision Pro — which means the expressive voices and high-accuracy dictation Apple is promoting hardest will land on a fraction of the installed base. Everything else routes to Private Cloud Compute, and therefore to the server-class Gemini-derived models, with daily usage limits that Apple is already monetizing through iCloud+ tiers.
The Regulatory Map Tells the Real Story
The availability footnotes are where the strategy becomes legible. Siri AI will not ship initially in the EU on iPhone or iPad, though it will reach Mac, Apple Watch, and Vision Pro there. It will not ship in China at all while Apple works through regulatory requirements. The privacy-first architecture that Apple markets as a competitive moat is also a compliance liability the moment it meets the Digital Markets Act or Chinese data law. An assistant whose entire value proposition is reaching into a user’s private messages, mail, and photos is precisely the kind of system regulators in both jurisdictions are built to scrutinize. Apple’s two largest non-US markets get a degraded product or none of it.
The feature surface is genuine and broad. Spatial Reframing lets users shift a photo’s perspective after capture, Extend fills in cropped edges, and Clean Up improves its infill. Safari gains automatic tab grouping, a Notify Me page-monitoring tool, agentic password upgrades that navigate sites to rotate weak credentials, and natural-language extension generation. Messages, Mail, and Calendar gain context-aware suggestions, with Call Context running entirely on device. Shortcuts can now be assembled from a plain-language description. These are useful. None of them is novel.
The developer betas open today, a public beta follows next month, and users get the release this fall with iOS 27 and its sibling platforms. The features will work, and most will work well. But the announcement that mattered was the org chart. For fifteen years Apple’s pitch was that it controlled the whole stack, and that control was the product. The next generation of Apple Intelligence runs on someone else’s models, watermarked with someone else’s standard, and that someone is the company Apple spent a decade defining itself against.
Leave a Reply