Apple’s Siri Reboot Puts Google’s Gemini Inside 2.5 Billion Devices

Bloomberg’s leaked renders of Apple’s redesigned Siri landed on Thursday with a tidy hook: a dedicated chatbot app, a Dynamic Island that morphs into a card-style result panel, a swipe-down search bar that talks back. The publication described illustrations produced from what reporters saw and learned from sources, eleven days before Apple’s Worldwide Developers Conference opens at Apple Park on June 8.

Tucked into the same coverage is the part that matters more than the icon. The rebuilt Siri runs on a custom version of Google’s Gemini, under a multi-year deal Apple and Google jointly disclosed on January 12. ChatGPT now reaches 900 million weekly active users. Apple’s install base sits at 2.5 billion devices. The new Siri app puts a Google-trained model inside the larger number.

Bloomberg’s Renders Show a Dynamic Island Siri

The leaked images depict three distinct surfaces. A standalone Siri app with persistent chat history, voice mode, and uploads for documents and photos. A Dynamic Island animation that expands from the iPhone’s pill-shaped notch when the side button is pressed, returning quick answers in formatted cards. A reworked Spotlight Search that still opens with a swipe down from the home screen, but routes the query through the new AI model instead of the local index.

From those three surfaces, iPhone users can ask about the weather, schedule a calendar entry, launch an app, dictate a message, search Notes, or trigger an app shortcut. Results render as text cards rather than the older Siri’s terse spoken replies. The standalone app’s interface, separately reported by 9to5Mac, will ship with a beta label and an auto-deleting chat-history setting tied to a privacy preference of 30 days, one year, or never.

Bloomberg’s report attributed the visuals to its own illustrators working from descriptions, not to screenshots pulled from a build. That sourcing matters: the renders are interpretation, not the shipping UI. Apple has yet to publicly confirm any of it. The official reveal arrives at the keynote, scheduled for 10 a.m. Pacific on June 8 at Apple Park.

The Gemini Layer Apple Will Not Headline

Apple and Google announced their tie-up in a joint corporate statement on the Apple-Google Gemini collaboration that ran four short paragraphs and avoided every number. The deal is multi-year. The next generation of Apple Foundation Models will be based on Gemini models and cloud technology. Apple Intelligence will keep running on devices and inside Private Cloud Compute, the company’s encrypted server enclave.

What the statement left out, Bloomberg’s Mark Gurman filled in. Apple is paying Google roughly $1 billion a year to license a custom Gemini variant with 1.2 trillion parameters, eight times the size of Apple’s own 150-billion-parameter cloud model. The custom build uses a mixture-of-experts (MoE, an architecture that activates only the relevant slice of the network per query) tuned for summarization, planning, and natural-language understanding.

That is the spine of the second-order story. Apple is not also building a chatbot. Apple is putting a Google model into the consumer surface where most people will first meet generative AI. The Siri app’s interface is the wrapper. Gemini is the brain. Apple’s foundation models stay on the device, for short and private queries; the heavier reasoning lifts out to the Gemini cluster.

• $1 billion per year, Gurman’s estimate for the Gemini license, never officially confirmed by either company.
• 1.2 trillion parameters in the custom model, against Apple’s own 150 billion in its cloud tier.
• Eight times the size of Apple’s prior in-house cloud model, by parameter count.
• Spring 2026 is when Apple has told staff the long-delayed personalized Siri features should ship, after roughly fifteen months of slipped dates.

2.5 Billion Devices Against 900 Million Weekly Users

Apple disclosed the 2.5-billion active-devices figure in its January 29 earnings call, up from 2.35 billion a year earlier. The count includes iPhones, iPads, Macs, Apple Watches, AirPods, Apple TVs, and HomePods in regular use. ChatGPT’s 900 million weekly active users were announced by OpenAI on February 27, with 50 million paying subscribers and a trajectory pointing at a billion before year-end.

The numbers measure different things. Apple counts hardware in someone’s pocket. OpenAI counts active product use in a seven-day window. But the ratio is the strategic point.

Even discounted by a generous engagement haircut, the new Siri’s reach dwarfs every dedicated chatbot in the consumer market. The iPhone owner does not download Siri. It is already there. That distribution moat is what makes Apple’s late arrival to generative AI a different competitive story from, say, Samsung’s Galaxy AI push or Microsoft’s Copilot rollout.

The Search Deal Apple Already Ran With Google

Apple has run this play before. Google has been the default search engine on iPhone since 2002 under an arrangement that, by 2022, was paying Apple an estimated $20 billion a year, equivalent to 36% of Safari search advertising revenue. The terms emerged publicly during United States v. Google, the antitrust trial Judge Amit Mehta ruled on in August 2024. The remedies order, issued in 2025, barred exclusive default agreements but left preferential placement payments intact.

That is the template Apple is reusing. Building a frontier search index from scratch was never on the company’s roadmap, so it let Google index the web and took a revenue cut. Building a frontier foundation model from scratch is the same calculus today: too expensive, too compute-heavy, too far behind. The Gemini deal hands Apple a model that is already better than what it can ship alone, while Apple’s own engineers keep refining the on-device tier.

The optics are inverted this time. In search, Google paid Apple. In AI, Apple is paying Google. The flow tells you who has the leverage in each market. For prior coverage on how the WWDC 2026 calendar is shaping up around this hand-off, see Budgy’s Apple WWDC June 8 Siri reckoning preview.

Beyond Gemini, the Routing Layer Stays Crowded

The leaked Siri app, according to multiple reports on the same Bloomberg sourcing, will let users route specific queries to other models. The drop-down menu pictured in the renders includes options to send a question to ChatGPT, Gemini, or Anthropic’s Claude directly. Apple already integrated ChatGPT into iOS 18.2 in December 2024 as an opt-in fallback for harder queries.

The routing surface gives Apple something Google’s Pixel and Microsoft’s Copilot devices do not: model-agnostic positioning. The pitch to a privacy-skeptical user is that Apple is the broker, not the brand of the brain.

What that menu looks like in practice:

• Default tier handles short, private queries on the device using Apple Foundation Models.
• Cloud tier escalates harder requests to the custom Gemini variant inside Private Cloud Compute.
• ChatGPT fallback remains available for queries the user explicitly routes to OpenAI’s model, free of charge for the basic tier.
• Claude routing, newly visible in the leaked menu, points to a planned Anthropic integration that has not been formally announced.
• App intents let third-party apps surface their own AI-powered actions inside Siri’s card results.

For prior reporting on the chat-history controls and how the privacy framing co-exists with a Google model, Budgy covered the Siri auto-delete setting tied to the Gemini deal.

The Risks Sitting Inside a Partner-and-Build Model

Three problems live in this strategy, and Apple knows each one well.

The first is execution risk Apple has already paid for once. The personalized Siri features previewed at WWDC 2024 slipped from a March 2025 launch into spring 2026 after the original architecture failed quality testing. Bloomberg reported in March 2025 that engineers had to swap to a stronger underlying model, which is the work the Gemini deal is meant to finish. A second public slip would hand competitors another year.

The second is regulatory exposure. The same $20 billion search arrangement that Apple is implicitly mirroring is now the subject of an active Department of Justice appeal, and Judge Mehta’s remedies order specifically forbade Google’s exclusive-default agreements. A Gemini default inside iOS could attract similar scrutiny, particularly if Apple structures the deal in a way that disadvantages OpenAI’s or Anthropic’s access to the routing menu. Google’s own search-monopoly liability also limits how much it can afford to pay Apple in any new arrangement.

The third is brand risk. Apple sold Apple Intelligence on a privacy promise. The on-device tier delivers that. The cloud tier, running a Google-built model inside Apple’s Private Cloud Compute enclaves, is harder to explain to the marginal customer who reads the press release and asks why Siri now has a Gemini sticker on the back. Apple’s pitch will be that the data still does not leave its enclaves. Whether that holds up to a year of media coverage is the open question.

The June 8 Keynote and What Ships After

Susan Prescott, Apple’s vice president of worldwide developer relations, framed WWDC in the company’s March announcement as a week celebrating “technology, innovation, and collaboration.” The collaboration word is doing more lifting than usual this year.

The Monday keynote will set the public framing. Expect an on-device demo first, a Private Cloud Compute architecture slide second, and a careful naming of Google as a partner somewhere in the second half of the segment. The standalone Siri app will likely ship to developer beta the same week, with a public beta in July and a general release alongside iOS 27 in September.

If the demo lands cleanly and the standalone app holds together under early reviewer scrutiny, Apple’s late entry becomes a distribution story rather than a catch-up story. If the personalized features slip again, or if the Gemini layer surfaces a privacy embarrassment in the first month of public use, the strategy that looks elegant on paper this week starts paying its bill in the fall.