Table of Contents
It’s the strangest sentence in tech this year, and it’s true: the new Siri’s brain comes from Google. In January, Apple and Google announced a multi-year deal under which the next generation of Apple’s foundation models will be based on Google’s Gemini models and cloud technology, powering Apple Intelligence and the long-promised smarter Siri.
Cue the reasonable questions: does my iPhone talk to Google now? Does Google see what I ask Siri? And after fifteen years of Siri jokes, does this finally make it good? I’ve gone through the joint statement, Apple’s earnings-call comments, and the reporting since, and the real picture is more interesting than both the “Apple surrendered” takes and Apple’s own soothing framing. Here’s the user-level version.
The Deal, in Plain Language
Apple evaluated the frontier AI labs and concluded Google’s technology was the most capable foundation for its models. The widely reported terms, which Apple hasn’t confirmed: roughly $1 billion a year for a custom-built Gemini model of about 1.2 trillion parameters, purpose-tuned for the things Siri does most, like summarization and planning. For scale, Apple’s own best cloud model reportedly topped out around 150 billion parameters. This wasn’t a tweak; it was an eight-fold jump Apple couldn’t build in time itself.
And that’s the honest subtext everyone in the industry read instantly: the deal is an admission that Apple’s in-house models weren’t ready to power a modern assistant on the timeline customers expected. Apple frames it as a bridge, and the reporting backs that up, with Apple developing its own AI server chips and data centers targeted at the next couple of years while Gemini carries the load. Non-exclusive, multi-year, and Apple keeps building underneath it.
What Gemini Powers vs What Stays Apple
Your iPhone didn’t become a Google phone, and the division of labor matters for everything else in this article.
Staying on your device, running Apple’s own models: the quick, personal stuff. Expressive Siri voices, advanced dictation, on-screen awareness, and lookups involving your personal context. This is the same on-device approach we’ve covered before, and it never leaves the phone.
Going to the big Gemini model: the heavy thinking. World-knowledge questions, complex multi-step requests, the conversational back-and-forth that made ChatGPT famous. These run on the custom Gemini model, but, and this is the architectural heart of the whole deal, that model runs inside Apple’s Private Cloud Compute servers, not on Google’s cloud.
Siri also remains a front door rather than a locked room: the existing ChatGPT hand-off stays, and iOS 27 goes further by letting you set third-party AI models as providers for some Apple Intelligence features. Gemini is the default brain, not the only one allowed in the building.
The Privacy Question, Answered Honestly
Here’s the question that matters most, so let’s do it properly rather than reassuringly.
Does Google see your Siri requests? No. The Gemini model is licensed to run on Apple’s own Private Cloud Compute infrastructure: Apple-controlled servers where requests are encrypted, processed in hardware-isolated enclaves, not stored afterward, not used for training, inaccessible to employees, and verifiable against publicly auditable code through cryptographic attestation. Google supplied the model the way an engine maker supplies an engine; it doesn’t ride along in your car. Tim Cook put it flatly on the earnings call: the privacy rules aren’t changing.
Now the honest asterisk. The architecture is genuinely strong, and it’s also true that with a more capable Siri, more of your requests will travel to a server than before, simply because the assistant can finally handle harder questions worth asking. Security researchers have made the fair point that a privacy guarantee is only as strong as its weakest link, and Apple’s answer, publishing the system for outside verification, is the right one, but “verifiable” and “verified forever” are different tenses. My practical read: for the overwhelming majority of Siri use, this is as privacy-protective as cloud AI currently gets, and meaningfully more protective than using the Gemini or ChatGPT apps directly, which run on their makers’ clouds under their makers’ policies. If something is sensitive enough that you wouldn’t email it, don’t dictate it to any assistant, this one included.
Which iPhones Get the New Siri
The hardware line is the same one we mapped in our iOS 27 coverage: the rebuilt Siri requires Apple Intelligence-class hardware, meaning an iPhone 15 Pro or 15 Pro Max, any iPhone 16, or any iPhone 17 family model, with the 8GB of RAM those chips carry being the true gatekeeper. Older iPhones run iOS 27 but keep the old Siri.
On timing: the new Siri is in beta right now behind a waitlist you’ll find in the Apple Intelligence settings, with the public release expected this September alongside iOS 27’s launch. If your iPhone qualifies and you’re on the beta, join the waitlist and be patient; if you’re not a beta person, September is close.
Does It Actually Make Siri Good?
The honest answer: the ingredients finally exist, and the verdict doesn’t yet.
What Apple is betting on isn’t beating ChatGPT at trivia; it’s context. ChatGPT and the Gemini app are powerful but live outside your operating system, while Siri natively reaches your Mail, Messages, Calendar, and Photos with system-level permission. A frontier-class model plus that access, wrapped in the privacy architecture above, is something neither pure-cloud assistant can offer. That’s the pitch, and early beta impressions suggest the conversational leap is real.
The risk is equally honest: Apple has a two-year record of missed Siri deadlines, and “good enough and built in” has to beat “noticeably smarter in a separate app” for the people who already live in ChatGPT. September’s public release, on hundreds of millions of capable iPhones at once, is where that question gets answered for real. I’ll be testing it the week it ships.
Quick Answers
- Is Siri just Google Gemini with an Apple logo?
No. A custom Gemini model handles the heavy reasoning, but it runs on Apple’s servers under Apple’s privacy rules, Apple’s on-device models handle personal tasks, and Apple controls the entire experience. - Does Google get my data from Siri requests?
No. Requests are processed on Apple’s Private Cloud Compute, with no storage, no training use, and no Google access. Google licensed the model; it doesn’t operate the service. - Why did Apple need Google’s AI?
Its in-house models weren’t competitive in time. The reported deal, about $1 billion a year for a model roughly eight times larger than Apple’s best, bridges the gap while Apple builds its own infrastructure. - Which iPhones get the new Siri?
iPhone 15 Pro and newer, meaning any model with Apple Intelligence support. Older iPhones get iOS 27 with the existing Siri. - When can I use it?
It’s in beta behind a waitlist now, with public release expected in September 2026 alongside iOS 27.
The Bottom Line
Apple buying Siri’s brain from Google is a genuine humbling and, for users, probably a genuine win: a frontier-class model, fenced inside the most privacy-protective cloud architecture in consumer AI, wired into the personal context only an operating system has. Google gets a reported billion a year and zero access to your questions; Apple gets time to build its own models; you get, finally, a Siri with a fighting chance of being useful. Whether the execution lands is a September question. The architecture, at least, deserves more credit than the headline suggests.
