Apple has spent more than a decade selling the idea that the best experiences come from controlling every layer, from silicon to software. Siri, the voice assistant it introduced in 2011, is now being rebuilt on technology from Google. Apple has confirmed a multi-year arrangement under which the next generation of its Apple Foundation Models will be based on Google’s Gemini models and cloud technology, and the revamped assistant is the first big beneficiary. For a company that rarely admits it needs outside help with a core product, it is a striking concession.
Siri didn’t lose ground to a startup; it lost it to its own roadmap.
At WWDC in 2024, Apple showed a Siri that would understand personal context, recognize what was on a user’s screen, and take actions across apps. Then the schedule slipped, and Apple acknowledged in 2025 that the more personalized version would take longer than planned. In the meantime, chatbots from OpenAI, Google, and Anthropic became fluent at drafting, summarizing, and reasoning through multi-step requests, which made Siri’s habit of answering with a list of web results feel dated. Apple’s stopgap was to let Siri hand off harder questions to ChatGPT, with the user’s permission. That patch worked as a patch, but it also made the gap obvious: the assistant on a billion-plus devices needed a stronger engine under the hood.
What Apple Is Actually Buying
It helps to separate the assistant from the model behind it. Siri is the interface, the thing that hears a request, decides what to do, and connects to apps and personal data. A foundation model is the large language engine that interprets messy human phrasing, keeps track of a conversation, and plans the steps needed to finish a task. Apple is reportedly paying roughly $1 billion a year for access to Gemini-based technology, a figure that looks modest next to the cost of training a frontier model from scratch. Building one means enormous compute budgets, scarce research talent, and years of iteration. Apple’s strength is hardware, operating systems, and distribution, so renting the engine while it catches up is a defensible trade even if it stings.
The practical payoff is meant to show up in everyday requests. Imagine asking Siri to find the PDF a colleague sent last week, pull the key dates from it, and add them to a calendar, all in one breath and with a follow-up question afterward. That kind of chained request depends on a model strong enough to hold context and plan, which is exactly where the older Siri struggled. Apple has said Apple Intelligence will keep running on its devices and on Private Cloud Compute, the company’s server architecture built around user privacy. How that promise holds up when the underlying models come from Google is the part customers cannot verify from the outside, so the technical details will matter more than the slogan.
There is also a branding question. Apple has not framed this as a Google-powered Siri in the way it framed the ChatGPT option, which appears as a visible, opt-in handoff. The Gemini work sits underneath Apple’s own models and interface, which means most users may never see a Google logo anywhere. That invisibility suits Apple, since the product identity stays intact, but it also raises an uncomfortable point. If the assistant’s quality depends on a rival’s research, the “Apple” in Apple Intelligence becomes more about packaging, privacy architecture, and integration than about the model’s raw intelligence. Those are real strengths, but they are different from the story Apple has told for years.
A Rivalry With a Revenue Line
Apple and Google have been entangled for a long time. Google pays Apple billions of dollars a year to be the default search engine in Safari, an arrangement that surfaced in detail during the Justice Department’s antitrust case. A judge’s remedies ruling in that case let Google keep paying for placement but barred exclusive contracts, and the restriction extended to distribution of Gemini itself. A deal that puts Google’s models inside the iPhone’s assistant will therefore draw attention from regulators, because it adds a second, deeper layer of dependence between two companies that already share a lucrative search relationship. Apple will want to show the arrangement is non-exclusive and that other models can still plug in.
For OpenAI, the deal is a quiet setback. ChatGPT is already wired into Siri as an optional helper, but being a side door is different from being the engine that powers the front door. Samsung’s Galaxy phones already lean on Gemini for many of their AI features, so the two largest premium smartphone brands will now draw on the same model family. That narrows one lever of competition. If the underlying intelligence looks similar on both platforms, the differences shift toward design, on-device speed, how well the assistant reaches into personal data, and how much users trust each company with it. Those are fights Apple is arguably better equipped to win than a pure model race.
The risk for Apple is leverage. A company that depends on a supplier for the heart of a flagship feature negotiates from a weaker position when the contract comes up for renewal. Apple is reportedly continuing to develop its own large models, and the Gemini arrangement may be a bridge rather than a permanent home. Bridges have a way of becoming roads, though, especially when switching means retraining an assistant that millions of people use daily and retesting every privacy guarantee along the way
