By Chris Meredith
Three days ago, Apple shipped iOS 27 as a full public release to 1.8 billion device owners. If you updated, you noticed Siri sounds different. Faster. Smarter. Better at following long, multi-step instructions without bouncing the question back at you.
What Apple did not put in the release notes: the new Siri is Google.
Not metaphorically. Apple signed a multi-year deal worth roughly $1 billion per year, according to Bloomberg’s reporting on the January 2026 announcement, to replace the AI brain inside its most personal product with a custom Gemini build made specifically for Apple. The model on your iPhone 18 reportedly has around 1.2 trillion parameters, roughly eight times larger than Apple’s internal models, per people familiar with the deal terms.
I have been an Apple user since 2007. I remember the “1984” commercial. I remember Steve Jobs calling Google a company that did not respect your data. I remember the privacy billboards, the locked-door WWDC keynotes, Tim Cook explaining year after year that the whole point of Apple was that your data stayed on your device.
That story is still partly true. That partial truth is the most interesting thing about what just happened.
What Apple Actually Did
The deal was announced in January 2026. Apple had been running out of road on its own AI work. Apple Intelligence, launched in 2024, landed flat by any honest measure. The on-device models were too small. The private cloud compute infrastructure Apple built was real and technically serious, but it could not close the gap with OpenAI and Google on the things people actually cared about.
So Apple made a business call. Rather than spend another two or three years and tens of billions trying to close the gap, it licensed the best frontier model available and wrapped its own privacy architecture around it.
The wrapping matters. Siri routes queries through Apple’s Private Cloud Compute layer before anything reaches Google’s servers. Under the agreement, Google cannot access your personal data, and cannot use your Siri conversations to train future Gemini models. You are not feeding the system that is running you.
Apple would call this principled. They built the privacy-preserving infrastructure. They wrote the contract with explicit data restrictions. They kept your photos, your health data, your messages on-device or behind their own servers. Google is providing compute, not surveillance.
Still, you are talking to Google.
The Part Apple Does Not Lead With
Here is what iOS 27 looks like day to day: you open Siri, ask it to book a dinner reservation for Saturday at a place your spouse mentioned last week, pull up the email thread where you discussed it, and text them the confirmation. The new Siri does this in one pass. It holds context across a long exchange without losing the thread.
That is genuinely useful. It is the Siri Apple promised in 2011 and never shipped.
But it runs on a model built by the company Apple spent fifteen years positioning itself against. Google processes the actual language understanding. Google’s training data, Google’s research, Google’s infrastructure shaped what Siri sounds like now. Apple built the layer on top. The engine is Mountain View.
Are you ready to deploy this across your enterprise? You might want to wait and see what the audit trail actually looks like. The privacy contract reads well on paper, but what does “Google cannot use your data” mean in practice when there is a legal subpoena, a policy change, or a renegotiation three years from now? The terms Apple published are the terms today.
Why This Is Bigger Than Siri
Worth noticing: the companies that spent the most time differentiating on AI privacy were, almost uniformly, the ones that fell furthest behind on AI capability.
Apple built a whole marketing identity around keeping your data local. A handful of European startups did the same. Several enterprise vendors sold “private AI” as a premium product. What happened to most of them is what happened to Apple: frontier models trained on massive public datasets, with billions in compute, simply got better faster than anything you could build while restricting your training data.
The gap between top open models and top closed models narrowed. The gap between top closed models and anything you could build locally widened. Apple looked at that and made a rational call.
The awkward implication is that privacy-preserving AI and cutting-edge AI capability may be genuinely at odds, not just as a technical problem but as a business model. If the best models are trained on the most data, and the most data comes from the least restricted data collection, the companies willing to collect more data have a structural advantage that is hard to close with engineering alone.
Apple’s answer is architectural: collect everything, process it behind a privacy layer, pass only the sanitized query to the external model. It may hold up. It may survive regulatory scrutiny. But it requires trusting that the architecture is implemented correctly, that the contract terms stick, and that nothing changes over a multi-year deal.
What Changes for Regular Users
If you are an iPhone user who does not think much about data policy, the practical answer is simple: Siri got dramatically better, and nothing in your daily experience will feel different except the quality of the responses.
Siri in iOS 27 can handle genuinely complex tasks in natural language. The multi-app automation works. Conversational context holds across extended exchanges. Response times on current hardware are noticeably faster, based on early user reports. This is the product Apple has been trying to ship for a decade.
If you are someone who bought Apple specifically because you believed your data never left your device, the answer is harder. Some of it still does not. Your photos, your health data, your contacts, your messages: Apple’s on-device and private-cloud processing handles most of those. But the language understanding layer, the part that figures out what you actually mean when you speak, now runs through Google’s infrastructure.
How much that matters is a values question as much as a technical one.
The Interesting Question Nobody Is Asking
Coverage of this launch has focused on features. Faster Siri. Smarter Siri. All of that is real and worth writing about.
But the more interesting question: what does this deal tell us about where the AI race actually ends?
Apple has more cash than almost any company on the planet. It has world-class engineering talent and a hardware advantage in custom silicon that most AI labs would trade significant equity to have. And it could not build a competitive frontier model on its own. Not in time. Not at a cost that made sense.
If Apple could not do it, the list of who can is very short: OpenAI, Google, Anthropic, Meta, a handful of Chinese labs. Everyone else is licensing or falling behind.
That consolidation is moving faster than most people are tracking. Apple just put it in plain view by signing one of those labs into the product that 1.8 billion people use every day.
Siri sounds better. Google is a little closer to your life than it was last week.
Chris Meredith writes about the intersection of AI, technology, and the choices we make about both.
Chris Meredith writes about AI, technology, and what it actually means for real people. Follow along on Substack: monkeyattack.substack.com
