Siri Gemini: Does Your Data Go to Google? What Changed

<!-- Hero image concept: a sleek iPhone with the redesigned Siri glow on screen, positioned between an Apple logo and a Google Gemini logo, with a faint padlock motif over the Siri orb to signal the privacy question. Clean, modern, Apple-keynote lighting. -->

Siri Gemini, and the one question that actually matters

The short version: yes, the rebuilt Siri leans on a custom Google Gemini-derived model for its hardest requests, and no, according to Apple your personal data does not go to Google. The Siri Gemini setup runs inside Apple's Private Cloud Compute, the same privacy sandbox Apple uses for its own AI, so the part of Google involved is the underlying model and some of the cloud hardware, not access to what you ask. That is the answer most people are searching for, so it goes first.

Last updated: June 22, 2026

The rest is worth understanding, because the architecture is genuinely unusual and the headlines have been messy. Apple is reportedly paying a fortune for Google's help, the model physically lives on hardware in Google's data centers, and Apple still insists its privacy promise is intact. All three of those things can be true at once. Below is what was announced, how the privacy design works, whether you can switch it off, and what it changes for you.

What Apple actually announced

Apple previewed the next generation of Apple Intelligence and a heavily rebuilt Siri at WWDC on June 8, 2026, during Tim Cook's final keynote as CEO. The new Siri can hold a back-and-forth conversation, see and act on whatever is on your screen, take multi-step actions inside apps, and lives in its own dedicated app that syncs across iPhone, iPad, and Mac.

Under the hood, Apple introduced its third generation of Apple Foundation Models, a family of five models that range from a small on-device model up to a large cloud model. Apple says these were "custom-built in collaboration with Google" and refined using outputs from Google's frontier Gemini models. The most capable one, which Apple calls AFM 3 Cloud Pro, handles the heaviest reasoning and is, in Apple's own words, comparable in quality to Google's frontier Gemini models.

So is Siri using Google AI? The accurate answer is layered. Apple's models were trained and distilled with Google's help, and the top-tier cloud model runs on the same class of hardware Google uses. But Apple drew a hard line on what it is not using. At a post-keynote session, software chief Craig Federighi put it bluntly: "The amount of the Google Assistant we use is none." Apple says it uses none of the Gemini models Google deploys to its own customers, none of Google's client-side code, and none of Google Search infrastructure.

There is an important reported-versus-confirmed split here. The original blueprint came from Bloomberg's Mark Gurman, who reported in November 2025 that Apple would pay Google roughly 1 billion dollars a year for a custom, roughly 1.2-trillion-parameter Gemini model to power Siri. That reporting drove every subsequent headline. Apple and Google later confirmed the partnership but have never confirmed the price or the parameter count, so treat the dollar figure and the model size as widely reported estimates rather than official facts.

Does your data go to Google? How the privacy design works

This is the heart of the Siri AI privacy question, and the design is more subtle than "Google now reads your Siri requests."

Apple routes every request through what it calls a system orchestrator, which decides where each query goes based on how much computing power and personal context it needs. There are effectively three tiers:

• On your device. Simple requests run entirely on the iPhone using on-device models. Nothing leaves the phone.
• Apple's Private Cloud Compute on Apple silicon. Moderately complex requests go to Apple-built servers running Apple chips. These process requests statelessly and never store them.
• AFM 3 Cloud Pro on Nvidia GPUs in Google Cloud. The heaviest reasoning tasks route to the big model, which runs on Nvidia GPUs hosted in Google Cloud.

That last tier is what spooked people, and it is where the nuance lives. The hardware sits inside Google data centers. But Apple says it extended Private Cloud Compute onto those Nvidia GPUs "while maintaining the same guarantees" it makes for its own servers. In practice that means the request is processed in a sealed, stateless environment: it is never stored, is not accessible to Apple or to Google, and is wiped when the task ends. Apple keeps a cryptographically verifiable, append-only ledger of every piece of Google Cloud hardware in the fleet, and it lets outside security researchers inspect the system to check those claims. Apple also says Google is contractually barred from using these requests to train its own models.

The technical enabler is confidential computing on the GPU itself. Apple and Google built the system on Nvidia hardware that encrypts the workload end to end, which Apple says prevents even the cloud operator, Google, from reading the data it is hosting. The analogy Apple uses is that it is renting locked, sealed rooms inside someone else's building. Google owns the building; Apple holds the only keys.

A fair, non-overstated summary: Apple's stated guarantee is that Google does not get your personal data, your identity, or your queries, even though the computation happens on hardware in Google's cloud. That is Apple's claim, backed by a verification scheme that independent researchers can test, rather than something a third party has fully audited as of June 22, 2026. It is a strong design, but it rests on Apple's implementation working as described, so "Apple says" is the honest framing.

If you want a deeper look at how privacy works when AI runs in your browser versus on a company's servers, our guide to whether online tools are safe covers the same client-side versus server-side trade-off in plain language.

Old Siri vs the new Gemini-powered Siri

Can you turn it off or opt out?

Yes, though not as a standalone "Gemini" toggle, because Apple does not surface Gemini as a separate product. The cloud model is part of Apple Intelligence, so you control it the way you control Apple Intelligence as a whole.

• Turn the whole thing off. Open Settings, go to Apple Intelligence and Siri, and switch it off at the top. With it disabled, requests stay on your device or simply do not run, so nothing routes to the cloud model.
• Restrict specific features. Under Settings, Screen Time, Content and Privacy Restrictions, you can allow or block individual pieces such as Writing Tools, Image Creation, and third-party intelligence extensions.
• Keep more on-device. Many everyday requests are handled by the on-device model regardless. The cloud tier only engages for tasks that genuinely need more horsepower.

What you cannot currently do is keep the new Siri's most advanced cloud features while specifically forbidding the Google-hosted tier. It is bundled into Apple Intelligence as one experience. If that bothers you, the realistic options are to leave Apple Intelligence off or to rely on requests simple enough to stay on the phone.

How much is Apple paying, and why pay at all

The reported number is roughly 1 billion dollars per year, attributed to Mark Gurman at Bloomberg. Apple has not confirmed it. For scale, that is a fraction of the reported 20-billion-dollars-plus a year Google pays Apple to remain the default iPhone search engine, so even if accurate, the Siri arrangement is a modest line item for Apple.

The reason is timing. Apple's own large models were not ready to deliver a genuinely conversational, agentic Siri on Apple's timeline, and a much-promised Siri overhaul had already slipped badly. Licensing and building on Google's frontier model technology let Apple ship a competitive assistant now instead of in a year or two, while it keeps developing its own models in parallel. The deal is widely read as a stopgap that buys Apple time, not a permanent surrender of its AI roadmap.

What this means for you

For most people, the practical takeaway is reassuring with an asterisk. You get a dramatically more capable Siri, and Apple's stated privacy model means your requests are not handed to Google, even though Google's technology and hardware are involved. The asterisk is that this is Apple's design and Apple's promise, supported by a verification system rather than something a regulator or auditor has signed off on, so it is reasonable to treat it as a well-engineered claim rather than a guarantee.

A few concrete things to know:

• If you own an iPhone 15 Pro or newer, you will get the new Siri. Older iPhones can run iOS 27 but not the new Siri. Our iOS 27 supported devices guide lists exactly which models make the cut.
• If you are privacy-cautious, you can turn Apple Intelligence off entirely, and Siri falls back to on-device handling.
• In the EU, the new Siri is not launching at first, so the timeline there is different.
• If your main use for an assistant is hands-free reading or turning text into natural speech, you do not need the cloud assistant at all. A focused tool like our free AI text-to-speech generator handles that job in your browser, and you can browse the rest of our AI tools for single-purpose alternatives that do not route anything through a cloud model.

Key takeaways

• The new Siri uses a custom Apple model built with Google's help; its top tier is comparable to frontier Gemini, but Apple says it uses "none" of Google Assistant.
• According to Apple, your data does not go to Google: heavy requests run inside Private Cloud Compute, which Apple extended onto Nvidia GPUs in Google Cloud, processed statelessly and not readable by Google.
• The roughly 1-billion-dollars-a-year price and the 1.2-trillion-parameter model size are reported by Bloomberg's Mark Gurman and not officially confirmed.
• You can opt out by turning off Apple Intelligence in Settings; there is no separate Gemini switch.
• The new Siri needs an iPhone 15 Pro or newer, arrives in beta in 2026 with a wider release around September, and skips the EU at launch.
• Apple's privacy claims are backed by external verification rather than a completed third-party audit, so "Apple says" is the accurate framing as of June 22, 2026.