When Tim Cook took the stage at Apple Park on Monday to announce that Siri’s cloud backend will now run on Google’s Gemini model, the applause in the room was polite — developers know a surrender when they see one. The deal, which Bloomberg’s Mark Gurman has been tracking since January, puts a custom ~1.2-trillion-parameter Gemini model behind every Siri query that can’t be handled on-device. Apple didn’t name a dollar figure on stage. They never do. But the strategic cost is the real number worth staring at.
This isn’t a partnership of equals. It’s a managed retreat.
The Moat Wasn’t the Chip; It Was the Margin
For two decades, Apple’s argument for vertical integration was elegantly simple: we control the silicon, the OS, the apps, and the services, so we can deliver experiences nobody else can replicate — and we can charge accordingly. The iPhone’s gross margins, hovering around 45-50% for most of the last five years, are a testament to that thesis. The Vision Pro launch in early 2024 was supposed to be its apotheosis.
But Apple Intelligence, introduced at WWDC 2024 with the kind of breathless fanfare Apple reserves for things it’s nervous about, has been a slow-motion disappointment — limited on-device capabilities, anemic adoption, and a Siri that still couldn’t answer basic follow-ups. The company spent two years and reportedly several billion dollars trying to catch up to OpenAI and Google. Monday’s keynote is the white flag: they couldn’t. Gemini will now do the heavy lifting Apple couldn’t do itself, and the integration will be deep enough that most users won’t know whose model they’re talking to.
“It’s the quietest $10 billion-a-year licensing deal in tech history,” one Cupertino engineer told me in a Slack DM during the keynote. “And we’re the ones paying.”
What Google Gets Is Worth More Than the Check
The predictable right-of-center take here is “Apple caved to Google, privacy is dead, walled garden breached.” That’s half-right and half-useful. The more interesting thing is what Google doesn’t need to do now. Google doesn’t need to win the premium hardware game. It doesn’t need Pixel to outsell iPhone. It just needs its models to become infrastructure — invisible, indispensable, the thing developers build against because it’s already there.
Apple just handed Google distribution into the most valuable user base on Earth: roughly 1.5 billion active Apple devices worldwide, a user base that skews higher-income and higher-spending than Android’s. Every Siri query that touches Gemini is training data Google can use to improve its models broadly — not the raw audio, which Apple will still process on-device for privacy, but the semantic patterns, the question structures, the intent signals that make a language model smarter over time. Privacy-preserving or not, that’s a feedback loop Android OEMs can’t match.
This is the revenge of the services layer. For years, Apple’s argument was that hardware is the durable competitive advantage — you can copy software features, but you can’t copy the A-series chip or the tight integration. Google is now proving that the opposite can be true: you can buy all the silicon you want, but if you can’t ship a competitive AI model, you’re renting one from Mountain View and hoping nobody notices.
The Developer Lock-In That Wasn’t
The most under-covered part of Monday’s announcements isn’t the Siri demo. It’s the set of APIs Apple quietly opened — or was forced to open — to make Gemini integration work across third-party apps. Developers who’ve spent years building against Apple’s proprietary frameworks now have a credible path to building against a cross-platform model that works on iOS, Android, and the web.
One iOS developer at a midsize fintech startup, texting me from the conference floor, put it bluntly: “Why would I optimize for Apple Intelligence when Gemini gives me the same thing plus Android reach?”
That’s the crack in the foundation. Apple’s developer ecosystem has always been its second moat — the millions of apps that make switching to Android feel like a downgrade. If the AI layer underneath those apps becomes platform-agnostic, the switching costs drop. Not to zero. But lower than they’ve been since the smartphone wars of the early 2010s.
The Prestige Trap
Apple has spent 15 years convincing itself — and its shareholders — that it doesn’t compete on price and it doesn’t compete on features it can’t control. The Gemini deal breaks both rules, quietly. It competes on an AI it can’t control and it will almost certainly pass the licensing costs to users through higher iCloud+ tiers or a bundled Apple One price hike by this time next year.
The irony is that this was probably the right call. Apple’s homegrown models weren’t getting there. The LLM race rewards scale and data access, and Apple’s principled privacy stance — which is real, not just marketing — genuinely constrained what its models could learn. Better to rent Google’s model and preserve the privacy narrative on-device than ship a broken assistant and watch loyalty erode. But “the right call” and “a strong position” are not the same thing.
When the history of the AI era is written, Monday’s keynote will mark the moment Apple stopped pretending it could win on AI on its own terms and started negotiating the terms of its dependence. That’s not a failure of execution. It’s a recognition that the game changed, and the old playbook — own the stack, own the margin, own the customer — doesn’t cover the new field.
Nobody in Cupertino will say that out loud. But the Gemini logo isn’t going to appear on stage by accident.