On September 10, 2026, the question of whether Siri runs on Google Gemini is back in tech headlines, with outlets including The Mac Observer and TechRadar revisiting a partnership Apple confirmed at the start of the year. The short answer, based on Apple’s own public statements, is not a simple yes or no. Apple and Google have entered a multi-year collaboration that puts Google’s Gemini models and cloud technology underneath the next generation of Apple Foundation Models, the system expected to power a more personalized Siri later this year. Apple has also been explicit, more than once, that this arrangement is not the same Gemini that shows up in the consumer Gemini app. Sorting out what “does Siri use Google Gemini” actually means requires walking through what each company has said, on the record, since January.
Apple Confirms the Gemini Deal Behind Siri’s Next Chapter
The collaboration first became public in a statement reported on January 12, 2026. Apple said, “After careful evaluation, we determined that Google’s technology provides the most capable foundation for Apple Foundation Models and we’re excited about the innovative new experiences it will unlock for our users,” a line that CNBC and several other outlets carried in full. That single sentence did most of the work of confirming what had been rumored for months: Apple would not keep building its foundation models entirely in-house. Instead, the next generation of Apple Foundation Models, referred to internally and externally as AFM, would draw on Google’s Gemini models and cloud infrastructure as a base layer.
Apple followed up in June 2026 with a newsroom post describing the same generation of models as “custom-built in collaboration with Google and its Gemini models” for “deeply integrated Apple Intelligence experiences.” That framing matters. Apple did not describe this as licensing an off-the-shelf Google product. It described a custom build process, with Google’s technology as the substrate and Apple doing additional engineering on top. Whether that distinction holds up under scrutiny is exactly what has kept the “does Siri use Google Gemini” question alive through the summer and into September.
What Apple and Google Actually Announced
The clearest single document on the arrangement is the joint statement itself. It reads, “Apple and Google have entered into a multi-year collaboration under which the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology.” That is a joint statement, not a leak or an analyst estimate, which is why it carries weight in a story that has otherwise generated a lot of speculation.
The same statement ties the deal directly to Siri: “These models will help power future Apple Intelligence features, including a more personalized Siri coming this year,” according to the excerpt published by Six Colors. Two things stand out. First, this is a multi-year deal, not a one-off licensing arrangement tied to a single Siri release. Second, Apple tied a specific product promise, a more personalized Siri, to the Gemini-based foundation, which is why so much of the public conversation has collapsed “Apple Foundation Models” and “Siri” into the same question.
Craig Federighi Draws the Line on Whose Gemini This Is
The most direct pushback on the “Siri runs on Gemini” framing came from Craig Federighi, Apple’s senior vice president of software engineering. In comments reported by Computerworld, Federighi said, “Of course, we don’t have the Gemini app as our app.” He went further, adding: “For these models, we use none of the models that Google deploys to their customers, nor do we use the infrastructure and means by which they deploy models to their customers.”
That is a specific, falsifiable claim, and it is the crux of why “does Siri use Google Gemini” resists a one-word answer. Apple is not saying it built its own model from scratch. It is saying it took Google’s underlying technology, the research and infrastructure approach behind Gemini, and built a separate, Apple-controlled model and deployment stack on top of it. In Federighi’s telling, the consumer-facing Gemini app and Google’s public model-serving infrastructure play no role in what reaches an iPhone user through Siri.
Inside the Third-Generation Apple Foundation Models
Federighi also gave the clearest technical breakdown of what Apple now ships. “Now, in that deployment model, we have a family, our third generation, of Apple Foundation Models, from our AFM Cloud and AFM Cloud Pro models to our AFM Fusion model and image model,” he said. That is the first time Apple’s leadership has publicly numbered the generations this way and named the specific model tiers: AFM Cloud, AFM Cloud Pro, AFM Fusion, and a separate image model, placing Apple’s stack alongside the broader wave of AI model announcements from rival platforms this year. It confirms Apple still ships a family of distinct models rather than a single rebadged Gemini instance, even if Google’s technology sits underneath the Cloud and Cloud Pro tiers.
How a More Personalized Siri Fits Into the Deal
Apple has spent roughly two years promising a smarter, more context-aware Siri, since the original Apple Intelligence rollout in 2024 fell short of the more ambitious personal-context features Apple had demoed at its September hardware events. The January and June 2026 statements are the clearest sign yet that Apple decided the fastest path to that smarter Siri ran through Google’s technology rather than a purely internal buildout. The joint statement’s language, a more personalized Siri coming this year, sets an expectation that has not yet been matched with a specific ship date in Apple’s public materials referenced here.
That gap between “coming this year” in January and no confirmed release date by September is a big part of why coverage from outlets like The Mac Observer and TechRadar keeps circling back to this story. Eight months is a long runway for a feature that was framed as imminent, and every Apple event since has renewed speculation about exactly when, and in what form, the Gemini-based Siri actually reaches users.
That timeline finally moved just after this story’s early-September snapshot. On September 14, 2026, Apple shipped iOS 27 with the rebuilt Siri included, though not switched on for everyone: the feature launched as an English-only gated public beta with a waitlist, limited to iPhone 15 Pro and newer and held back at launch in the European Union and China. Installing iOS 27 on its own does not enable it, which keeps the careful distinction Apple has drawn all year intact: the Gemini-derived foundation reaches users through an Apple-controlled, staged rollout rather than a single switch being flipped. In other words, the “coming this year” promise landed on schedule, but as a limited first release rather than a broad launch.
Why Apple Turned to Google Instead of Building Alone
Apple’s own explanation is unusually candid for a company that rarely admits a competitor did something better. The “after careful evaluation” statement amounts to Apple conceding that its in-house foundation model work, the first two generations of AFM, did not clear the bar Apple set for itself on Siri’s next phase. A separate remark attributed to an Apple representative and reported by MacRumors put it plainly: “We basically determined that Google’s AI technology would provide the most capable foundation for AFM (Apple Foundation Models), and we believe that we can unlock a lot of experiences and innovate in a key way due to the collaboration.”
Building frontier-scale models is expensive, and training infrastructure at Google’s scale is not something Apple assembled overnight. Renting the foundation, in Apple’s words, while keeping the deployment layer in-house, let Apple ship faster without fully outsourcing the product experience. It is the same logic that has pushed other hardware-first companies toward cloud AI partnerships rather than isolated, ground-up model development.
On-Device, Private Cloud Compute, and Where Gemini Actually Sits
Reported coverage of the deal has consistently said the resulting models still run on Apple devices and on Apple’s Private Cloud Compute, the server architecture Apple introduced in 2024 specifically to extend on-device privacy guarantees into the cloud. That detail is the backbone of Apple’s public defense of the deal: even if Google supplied the technology, Apple insists the request handling, routing, and data protections remain under Apple’s control rather than Google’s.
The general shape of that routing, as described across Apple’s own Apple Intelligence materials and echoed in coverage of the Gemini deal, looks something like a tiered decision process. The block below is an illustrative sketch of that flow, not an official Apple diagram, meant to show how the pieces likely fit together based on what Apple has publicly described.
Siri request
-> On-device AFM model (fast, private, limited context)
-> if task exceeds on-device capability:
-> Apple Private Cloud Compute
-> AFM Cloud / AFM Cloud Pro (Gemini-derived foundation, Apple-run stack)
-> response returned to device, no retention per Apple's stated policy
Nothing in that flow routes a request to Google’s own servers or to the public Gemini app’s backend, according to Apple’s statements. The dividing line Apple draws is between using Google’s model research and infrastructure approach as a foundation versus using Google’s live, customer-facing service. That is a nuanced distinction, and it is one regulators and privacy researchers are likely to keep testing rather than take at face value.
What Remains Unconfirmed About the Siri-Gemini Relationship
Two things are not confirmed in the record available as of September 10, 2026. First, it is not confirmed that Siri is “using Gemini” in the sense of running the consumer Gemini app or Google’s public customer-facing model stack directly. Apple’s quoted statements describe a collaboration that uses Google Gemini models as an input to Apple Foundation Models, with Apple stressing its own device and cloud deployment on top. Second, specific internal model names beyond AFM Cloud, AFM Cloud Pro, AFM Fusion, and the image model, along with parameter counts or a detailed technical architecture, have not been laid out by either company in the sources reviewed here.
That leaves plenty of room for the headline version of this story, “Siri uses Gemini,” to overstate what has actually been confirmed. A more accurate framing, based strictly on what Apple and Google have said, is that Siri’s next generation is built on Gemini-derived technology inside an Apple-run system, not that Siri has quietly become a Gemini client.
Market Impact: A Trust Signal for Google, a Question for Apple
Google’s Cloud and Gemini Business Gets a Halo Effect
For Google, being named as the technology foundation behind the next Siri is a meaningful credibility marker, independent of the deal’s exact dollar terms, which have not been disclosed in the sources reviewed for this piece. Google has spent years trying to convince enterprise and consumer partners that Gemini and its underlying cloud infrastructure can compete with in-house alternatives at the largest companies in tech. Landing Apple, a company famous for building rather than buying critical infrastructure, is a different kind of proof point than another cloud customer signing up for API access.
Apple’s Silicon and Data Center Bets Still Matter
For Apple, the deal is a hedge rather than a retreat. Apple still controls the deployment layer, the on-device models, and Private Cloud Compute, all areas where its custom silicon and privacy engineering remain differentiators regardless of whose foundation model sits underneath AFM Cloud and AFM Cloud Pro. That approach also lines up with Apple’s broader AI infrastructure strategy, which leans on owned hardware even when it borrows model technology. The open question for investors and developers is whether this arrangement is a permanent architecture or a multi-year bridge while Apple’s own foundation model research catches up. Apple has not said which, and the multi-year framing of the deal suggests Apple itself may not have fully decided.
Competitive Comparison: How Apple’s Approach Stacks Up
Apple is not the only device or platform company weighing whether to build foundation models in-house or lean on an outside partner. The table below lines up the broad strategic posture of Apple against the other major players shaping this cycle of consumer AI, based on each company’s own public statements about its approach.
| Company | Foundation model approach | Deployment layer | Consumer AI product tied to it |
|---|---|---|---|
| Apple | Gemini-derived foundation (3rd-gen AFM), custom-built with Google | On-device plus Apple-run Private Cloud Compute | Siri, Apple Intelligence |
| Gemini models built and trained fully in-house | Google Cloud and Google-run consumer infrastructure | Gemini app, Google Assistant successor features | |
| OpenAI | GPT-family models built in-house, deployed via own and partner infrastructure | OpenAI infrastructure plus Microsoft Azure | ChatGPT and API-based integrations |
| Samsung | Mix of in-house Galaxy AI models and third-party partnerships, including Gemini integration in its own apps | Device plus partner cloud, split by feature | Galaxy AI, Bixby-adjacent features |
| Anthropic | Claude models built in-house, offered to partners rather than borrowing outside foundations | AWS and Google Cloud infrastructure for serving | Claude apps and enterprise integrations |
The pattern that stands out is that Apple is the only company on this list combining someone else’s foundation model technology with its own full deployment stack for a flagship consumer assistant. Samsung already blends its own models with Gemini in specific app integrations, but Siri’s role as the operating-system-level assistant across a billion-plus devices makes Apple’s version of this trade-off higher stakes. By contrast, Google, OpenAI, and Anthropic have all kept their foundation model layer and their primary consumer product under one roof.
Historical Context: From Siri’s 2011 Debut to a Third Generation
Siri launched in 2011 as a rules-based assistant built on technology Apple acquired rather than built from scratch, so a Siri that leans on outside AI research is not without precedent. What has changed is the scale and visibility of the dependency. Apple Intelligence arrived in 2024 as Apple’s first serious large language model push, built on the first generation of Apple Foundation Models, running a mix of small on-device models and larger cloud models under Apple’s own research. A second generation followed with incremental improvements, but reports throughout early 2026 pointed to Apple falling behind Google and OpenAI on the kind of complex, multi-step reasoning that a truly personalized Siri would require.
| Generation | Approximate period | Foundation source | Public status as of Sept. 10, 2026 |
|---|---|---|---|
| 1st generation AFM | 2024 launch with Apple Intelligence | Built in-house by Apple | Superseded |
| 2nd generation AFM | 2025 update cycle | Built in-house by Apple | Superseded |
| 3rd generation AFM (Cloud, Cloud Pro, Fusion, image model) | Announced January 2026, detailed June 2026 | Custom-built with Google’s Gemini models and cloud technology | Current, tied to “more personalized Siri” promise |
Read against that history, the Gemini deal looks less like a sudden reversal and more like the third attempt at solving the same problem: Apple wants a Siri that can reason across a user’s messages, calendar, and apps without sending raw data off the device, a bar that even Google’s own retirement of the original Google Assistant shows the whole industry is still chasing. The first two generations of AFM did not get Apple there on Apple’s own timeline. The third generation is the one built with Google’s help.
The Regulatory and Trust Questions This Raises
Apple and Google already share one of the most scrutinized business relationships in tech, the search default deal that has drawn years of antitrust attention in the United States. Adding a foundation model partnership on top of that existing relationship gives regulators and privacy advocates a new angle to examine, particularly around how much visibility Google has into the data flowing through Apple’s Private Cloud Compute even if Google’s live infrastructure is not directly handling it. Apple’s public position, that it uses none of the models or infrastructure Google deploys to its own customers, is the company’s first line of defense against those concerns, but it has not stopped commentators from asking for more technical detail than either company has published.
Privacy-focused users and enterprise IT teams evaluating Apple devices for regulated environments are the audience most likely to push for that detail. Until Apple publishes something closer to a technical white paper on how the Gemini-derived foundation is isolated from Google’s production systems, the assurance remains a statement rather than an audit.
What Developers and Enterprise Customers Should Watch
Developers building on Apple’s on-device and cloud AI APIs have a practical reason to care about this story beyond the headline. If the underlying foundation model shifts generations again, behavior, latency, and output quality across Apple Intelligence-linked features can shift with it, even if the developer-facing API surface stays stable. Enterprise customers evaluating Apple devices for regulated workloads should watch for any updated documentation on Private Cloud Compute that addresses the Gemini-derived foundation specifically, since existing PCC transparency reports predate this collaboration.
The practical guidance for now is to treat the current AFM Cloud and AFM Cloud Pro tiers as a new baseline rather than an incremental patch, given that Apple itself is calling this a distinct, third generation of its foundation models.
Predictions: Where the Apple-Google AI Partnership Goes Next
- Apple will ship the “more personalized Siri” feature referenced in the joint statement before the end of 2026, most likely tied to a fall software update, given the deal’s own timeline promise.
- Apple will keep emphasizing on-device processing and Private Cloud Compute in its marketing to separate its privacy story from Google’s role as a technology supplier.
- Expect continued pushback and clarification from Apple executives, following Federighi’s pattern, each time press coverage compresses the story into a simple “Siri runs on Gemini” headline.
- Regulatory attention on the Apple-Google relationship will expand beyond the existing search default scrutiny to include this foundation model partnership, particularly in the European Union.
- Apple will likely continue in-house foundation model research in parallel, keeping the option to reduce its reliance on Google’s technology in a future generation once its own models close the gap.
Frequently Asked Questions
Does Siri use Google Gemini directly?
Not in the sense of running the consumer Gemini app. Apple says the next generation of Apple Foundation Models is built using Google’s Gemini models and cloud technology as a foundation, but deployed through Apple’s own systems, including on-device processing and Private Cloud Compute.
When did Apple confirm the Google Gemini partnership?
Apple’s statement was reported on January 12, 2026, and Apple detailed the arrangement further in a June 2026 newsroom post describing the models as custom-built in collaboration with Google.
What did Craig Federighi say about Siri and Gemini?
Federighi said Apple uses none of the models Google deploys to its own customers, nor the infrastructure Google uses to serve those customers, distinguishing Apple’s Gemini-derived foundation from Google’s public-facing Gemini product.
What are AFM Cloud, AFM Cloud Pro, and AFM Fusion?
According to Federighi, these are the model tiers in Apple’s third-generation Apple Foundation Models family, alongside a separate image model, representing Apple’s current lineup of cloud and on-device models.
When is the more personalized Siri launching?
The joint Apple-Google statement said the feature was “coming this year,” referring to 2026, and Apple delivered on that window on September 14, 2026, shipping the rebuilt Siri with iOS 27. It arrived as a limited first release, an English-only gated public beta with a waitlist, restricted to iPhone 15 Pro and newer and not available at launch in the European Union or China.
Does this deal mean Apple stopped building its own AI models?
No. Apple’s statements describe a custom-built foundation using Google’s technology, not a wholesale replacement of Apple’s model development. Apple continues to control deployment, on-device models, and Private Cloud Compute.
Is this the same as Samsung’s Gemini integration on Galaxy phones?
No. Samsung integrates Google’s Gemini app and services directly into some Galaxy features. Apple’s arrangement, per its own statements, keeps deployment in Apple’s hands and does not put the consumer Gemini app on iPhones.
Could Apple move away from Google’s technology in a future generation?
Apple has not said. The deal is described as multi-year, and Apple’s continued in-house research suggests the company is keeping options open for later generations, but no public timeline for a shift back to fully in-house models has been announced.
