Apple's big AI reveal quietly runs on Google's Gemini

Let's give Apple what it's owed first. Siri AI, shown at WWDC on 8 June, looks like the assistant Apple over-promised in 2024 and then couldn't ship — on-screen awareness, personal context, real in-app actions across your apps. Two years late, materially better than what shipped in the interim, arriving this autumn in iOS 27, iPadOS 27 and macOS 27 ('Golden Gate'). That is a real product. The 2024 demo was an embarrassment; the 2026 version looks like it closes the gap.

Now the part Apple didn't put on a slide. Reporting from TechCrunch (9 June) says the brains are a custom Google Gemini model, with Apple paying Google roughly $1 billion a year. The company whose entire AI marketing is 'private, on-device, ours' is leasing its flagship intelligence from the biggest data company on earth. Pragmatic? Absolutely. On-brand? Not even slightly. And the number — a billion dollars a year — is not rounding error. That is a line on Apple's income statement that benefits Google and signals a dependency the company cannot cover up with 'Apple Intelligence' branding forever.

Apple's competitive advantage has always been explained in one word: integration. Hardware, OS, silicon, services — all Apple's, all talking to each other, all controlled. The iPhone's moat wasn't any single feature; it was the stack. Now the most-promoted layer of that stack — the AI layer, the one that is supposed to turn the iPhone into a genuinely intelligent device — is, per reporting, not Apple's. It is a Gemini model with Apple wrappers on top. The wrappers matter. The privacy controls Apple builds around it matter. But the intellectual property and the inference economics of the underlying model belong to Google.

Two more things break the 'it just works' narrative. China — one of Apple's top three markets — gets nothing at launch because regulators haven't approved it. The EU gets a limited version, mostly confined to Mac and Vision Pro, because the Digital Markets Act restricts how deeply Apple can bake a system-level AI into iOS without compliance headaches Apple hasn't solved. Neither is Apple's fault directly. Both reduce the commercial impact of the launch materially. The company shipped Siri AI to a large fraction of its most valuable users, and a meaningful fraction of its user base gets to watch.

And then there is the leadership context the keynote barely acknowledged. This was Tim Cook's last WWDC before handing the CEO role to John Ternus on 1 September. The stage-managed framing — fixes first, Siri as one item on a polished list, no big triumphalist moment — is the product of a team that burned credibility in 2024 and will not do it again. That discipline is correct and probably earned. What Ternus inherits is a Gemini-dependent AI platform with geo-holes, shipped by a company still rebuilding trust after the AI embarrassment of two years ago. That's not a disaster. It's also not the clean handoff the keynote implied.

Three reads going forward. First: does the autumn release hold? Apple said 2024 and slipped. If Siri AI misses again, the credibility hit is larger than the original. Second: when does Apple start replacing Gemini? Apple's track record with dependencies is eventually to eliminate them — it did it with Intel, it has tried with Google Search. The R&D investment required to build a competitive foundation model internally is enormous, but Apple has the resources. Third: how does this read for Google? A $1bn/year deal where Apple distributes Google's model to a billion-plus premium devices is, from Google's perspective, one of the better enterprise agreements on the planet. That alignment of interests may explain why the deal exists at all — and why it's surprisingly durable for two companies that compete in almost every other arena.