Skip to main content

The World’s Most Valuable Hardware Company Just Rented Its AI Brain: Inside the New Gemini-Powered Siri

A composite graphic of an iPhone at night, with a glowing brain made of energy. Text includes 'Apple + Google AI Integration', 'OpenArt', 'Siri AI Integration' on the phone screen, and 'Google Gemini'.

Siri Just Turned 15. Its Brain Belongs to Google.
Siri turned 15 this year, and for most of that decade and a half it was less an assistant and more a punchline. "Sorry, I didn't get that." We all heard it. We all mocked it.
Well, the new Siri the one Apple teased at WWDC 2024 and finally shipped with iOS 27 is real. And here's the detail the headlines rushed past: Apple didn't build the intelligence. Google did. Running in Google's cloud. On Nvidia's chips.
Apple. The company that built an entire brand on "your data stays on your device." Its flagship AI feature runs on a rival's model, inside a rival's cloud, on a third company's hardware. Worth sitting with.
What actually shipped
This isn't a coat of paint on the old Siri it's a rebuild. Twenty-plus conversational turns in a row. Multi-step commands that hop between native apps. A dedicated Siri app with conversation history that syncs privately across your devices.
Under the hood, requests split in two. Simple stuff timers, weather, toggling settings stays on-device, running on Apple's own silicon. Anything complex, like summarizing a document or orchestrating across apps, routes to the cloud.
That cloud layer is a customized Gemini, reportedly around 1.2 trillion parameters bigger than anything Apple runs on its own hardware. The reported price tag for access: roughly $1 billion a year.
Why license instead of build?
Because Apple tried the other way and it didn't work. The on-device Siri overhaul teased in 2024 slipped quietly through 2025 amid internal reshuffles. By WWDC 2026, Apple made the pragmatic call.
The engineering logic is hard to argue with. Training and serving a trillion-parameter model takes compute infrastructure Apple simply never built its whole hardware philosophy was optimized for efficient edge processing, not warehouse-scale clouds. Google spent the last decade constructing exactly that. Licensing beats building from scratch when you're years behind and the market won't wait.
The privacy plumbing
To keep this from torching its privacy reputation, Apple built a proxy layer. Queries go through Private Cloud Compute rather than hitting Google's public APIs. The Gemini weights run in Apple-controlled, hardware-isolated environments with end-to-end encryption no raw user data handed to Google, nothing retained after execution.
The servers themselves? Nvidia Blackwell B200 GPUs with confidential computing enabled.
So a single Siri query now passes through Apple, Google, and Nvidia three giants, none of whom can see the whole picture. Whether regulators and security researchers buy that arrangement is a question the next few update cycles will answer.
The philosophical shift nobody's pricing in
iOS 27 will eventually let developers plug third-party models from the App Store straight into Siri's orchestration layer.
That's not just a developer feature. For decades, Apple's moat was vertical integration silicon, OS, apps, all owned end to end. Opening the core assistant to outside models is a quiet admission: you don't need to own the frontier model to own the customer. Apple's crown jewels are the device, the interface, and the ecosystem wrapped around whatever brain is inside.
The timing is not a coincidence
Tim Cook handed the CEO title to John Ternus on September 1, and reporting ties the long Siri delays partly to that transition. The Gemini deal lands as the defining move of Apple's new era.
It also landed days after a federal court declined to force Google to divest Chrome a ruling that kept Google's legal room to do deals exactly this size. The stars aligned, legally speaking.
What it means
For developers, Apple's Foundation Models framework means small teams get frontier-scale capabilities without negotiating their own licenses. Convenient and quietly anchoring a growing slice of the iOS ecosystem to Google's infrastructure.
The bigger story: when training frontier models costs this much, nobody builds full-stack anymore. Not even Apple.
One caveat for the global audience: the new Siri isn't available in the EU or China yet, pending local regulatory and data-sovereignty reviews. The future, but not everywhere, not yet.

Comments

Popular posts from this blog

Next-Gen EV Solid-State Battery Breakthroughs in 2026: Commercial Scaling, Material Science & Market Integration

How to Build a Full-Stack AI Tools Directory App: The Complete Developer’s Guide (Next.js + Supabase)

Prompt to Production: The Technical Architecture of Autonomous Full-Stack AI Generation