Apple’s WWDC 2026 AI Strategy Hid a Google-Powered Siri

Apple called it the calm, privacy-first alternative to the AI arms race. On June 8 at WWDC 2026, Craig Federighi took the Apple Park stage and accused unnamed rivals of “pursuing AI for the sake of AI, without clear regard for the people, all of us, that it’s ultimately meant to serve.” The new Siri, the new foundation models, and the new OS-level AI features all came wrapped in the same framing: Apple is doing AI differently, on its own terms, with privacy as the load-bearing wall.

Underneath that wall is a different story. The Siri AI Apple shipped this week runs on third-generation Apple Foundation Models, four of which are “custom built for Apple Silicon, trained using proprietary data, and refined using outputs from Gemini frontier models,” per Apple’s own AI lead. The most powerful of those models is served on Nvidia GPUs inside Google’s cloud. The WWDC 2026 deal that put Gemini inside Siri is the underlying story the keynote would rather you not look at too closely. Apple spent two years promising a Siri revival, missed its first deadline, and is now framing the catch-up as a deliberate choice. Tim Cook’s last WWDC as CEO, with the formal handoff to John Ternus set for September 1, is the moment the bill for that delay comes due.

Federighi’s “Non-Racing” Pitch

Federighi opened with the line that ended up defining the day. He framed the WWDC 2026 keynote as Apple’s refusal to join the rest of the industry in chasing bigger, faster, more agentic models. The line was a clear shot at OpenAI, Google, Anthropic, and the rest of the field shipping new coding assistants, AI-native browsers, and chatbot upgrades on a quarterly cadence.

We believe privacy in AI is non-negotiable. Data is only used to execute your request, and outside experts can continue to verify this promise at any time.

Craig Federighi, Apple’s senior vice president of software engineering, said that line during the June 8 keynote. He used the rest of the presentation to make the same point a third way. Apple led with quality-of-life fixes before features, framed a better Siri as one item on a long list of improvements rather than the main event, and built the Apple Intelligence pitch around what it does not do. No coding assistant. No “superhuman” framing. No promise that AI will replace a worker, a writer, or a software engineer.

How Google Became the Brain Behind Siri

The architecture tells a different story than the keynote did. At a post-keynote tech talk the same afternoon, Apple’s AI leadership walked through the third generation of Apple Foundation Models, called AFM. There are five of them, and the lineup is the clearest signal yet that Apple’s “doing it our way” framing sits on top of work the company did not, and could not, do on its own.

• AFM Core: Apple’s next-generation on-device model, a dense architecture shipped on supported iPhones, iPads, and Macs starting in the iOS 27 developer beta.
• AFM Core Advanced: Apple’s first on-device sparse model, natively multimodal, powering expressive voice, visual, and invitation features entirely on the device.
• AFM Cloud: the server-side workhorse, optimized for latency and cost, served from Apple’s Private Cloud Compute.
• AFM Cloud Image: the new image generation and editing model behind Spatial Reframe, Extend, and the updated Cleanup tool in Photos.
• AFM Cloud Pro: Apple’s most capable model, with quality “similar to Gemini frontier models” per Apple AI lead Amar Subramanya, reserved for agentic tool use and complex reasoning.

Four of those five models, Subramanya confirmed, were “refined using outputs from Gemini frontier models.” The fifth, AFM Cloud Pro, runs on Nvidia GPUs inside Google’s cloud, accessed through Apple’s Private Cloud Compute stack. Apple’s post-WWDC tech talk with the press is where the architecture was actually laid out.

We work with both Google and Nvidia to extend our Private Cloud Compute infrastructure to NVIDIA GPUs in Google’s cloud, while maintaining Apple’s unmatched privacy guarantees.

Apple AI vice president Amar Subramanya said that during the post-keynote tech talk on June 8, and the sentence is the spine of the whole announcement. Apple would rather you read past it. Apple is also unusually explicit about what it is not using. Federighi, asked the same afternoon whether Apple had built Siri on the Gemini app, was emphatic: “Of course, we don’t have the Gemini app as our app. In fact, none of that client code is part of how we run on iOS. 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. And then, when it comes to the knowledge base, we of course don’t use Google Search or anything like that as the foundation of our system. So I hope that’s clear. The amount of the Google Assistant we use is none.”

Where the Two Years of AI Delay Showed

The two-year AI delay that preceded WWDC 2026 shows up in the feature list. Apple shipped a credible cross-app Siri, a system orchestrator, a fresh take on Photos, and a one-tap password update flow. iOS 27 is eligible for every device from the iPhone 11 onward, a wider device base than any prior iOS release. Apple also shipped the basics: tab management for Safari, AI-generated Shortcuts from a sentence, dictation that cleans filler words, and a parental controls suite rebuilt for a post-TikTok era. Most of these are parity features that Google, OpenAI, and Anthropic have shipped, in some form, for a year or more. Apple’s full WWDC 2026 announcement is long on polish and short on agentic breakthroughs.

Apple is also the company finally giving users a way to roll back last year’s divisive Liquid Glass design with an opt-in transparency slider, and shipping Health features that have been on the wish list for years. Perimenopause and menopause support in Cycle Tracking, expanded Communication Safety for child accounts, and Time Allowances that parents can set by category all made their debut. TechCrunch’s reading after the keynote was that Apple “led with fixes before features, and framed a better Siri as one item on a long list of improvements rather than the main event.” Apple spent two years watching competitors ship agentic AI, AI-native browsers, and AI coding assistants, and on June 8 it shipped the version of those features that fits inside the Apple sandbox. The “non-racing” line is a description of the lane, not the speed.

Can the Privacy Story Hold on Google’s Cloud?

The privacy pitch is built on three specific claims, and each one is now load-bearing in a way it was not two years ago. The first is the system orchestrator, which Federighi described as “key to the privacy architecture of our entire system.” It coordinates requests against the App Toolbox, the Spotlight Semantic Index, and on-screen context, and decides whether a question can be answered on-device or needs to be routed up to the cloud. The idea is that the most personal queries never leave the device, and the ones that do are routed through a stack Apple controls end to end.

The second is Private Cloud Compute itself, which Apple has pitched since 2024 as an extension of iPhone privacy into the cloud. Requests, in Apple’s telling, “are never stored, they’re never accessible to anyone, including Apple, they’re only processed as part of the request, and nothing can access them. All of those properties are not only built architecturally deep into the system, but are also something that third-party researchers can continuously verify.” That second clause is doing more work than it used to. The cloud in question is no longer Apple’s alone.

The third is the new dependency Apple added this year. To run AFM Cloud Pro, Apple extended its Private Cloud Compute stack to Nvidia GPUs inside Google’s cloud, using a feature called Nvidia Confidential Compute. Apple software VP Sebastien Marineau-Mes said the company wanted Nvidia’s latest chips and “set out to extend private cloud compute to third-party cloud.” The privacy claim now rests on the integrity of confidential-compute silicon from a vendor Apple does not control, running in a data center Apple does not own, hosting weights refined from a frontier model Apple licensed from a direct competitor. The Nvidia partnership for the most capable AFM model is now part of the architecture Apple is asking consumers to trust on faith.

That is not the same shape of privacy promise Apple made when it introduced Private Cloud Compute, and the company did not give journalists much room to interrogate it. The post-keynote tech talk was held to walk through the architecture in detail, but the only outside auditors who can confirm the claims are the ones Apple itself invites in. “Outside experts can continue to verify this promise at any time” is a sentence that requires an active verification program, not a rhetorical one.

Cook’s Exit, Ternus’s September 1 Inheritance

WWDC 2026 is also the last developer conference Tim Cook will open as Apple’s CEO. Apple announced earlier this year that Cook would hand the job to John Ternus, Apple’s senior vice president of hardware engineering, on September 1. Ternus has been at Apple since 2001, has run the hardware organization since 2021, and is best known publicly for leading the Mac transition to Apple Silicon.

The AI strategy he inherits is the one Federighi laid out on June 8, and the September 1 switch gives him roughly three months to ratify, redirect, or quietly reshape it before the iOS 27 public release in the fall. Cook used his closing minutes at WWDC 2026 to frame the moment. “Over the years, you have helped people connect, create, learn, and experience the world in extraordinary new ways, and with the incredible capabilities we introduce today, and so many more still to come, I truly believe the best is still ahead at Apple,” he told the audience. The line was written for the room, and it will be read for years as the last sentence of a CEO who shipped AirPods, the Apple Watch, the Vision Pro, and Apple’s first real AI platform, in that order.

The Federighi Test

The strategic bet is that “calm, private, on your terms” is a more durable consumer position than “most capable, most agentic, first to ship.” Federighi is the one who has to make that case to developers, to regulators, and to a public that has now spent two years getting used to ChatGPT, Claude, and Gemini as default interfaces. Apple previewed its third generation of Apple Foundation Models on June 8, including AFM Cloud Pro, which the company says has quality similar to Gemini frontier models. Apple is offering the opposite. A Siri that does less, knows more about you, and refuses to send your data anywhere the company cannot audit. That is a real product story, and it is a real bet on what consumers will pay for.

It is also a bet that depends on a few things holding. The Gemini-derived foundation models have to match what users get from first-party Gemini in everyday use. The Nvidia-Google cloud stack has to deliver the privacy promise under independent audit. The system orchestrator has to route enough queries on-device to keep the cost and the privacy story intact. And Siri AI has to reach Europe, where Apple has said the Digital Markets Act means the new AI-powered Siri will not launch on iPhones and iPads in the EU, with no public timeline for a workaround. The report that Siri AI will not launch in the EU is the first sign that the “different, not racing” framing is going to be tested in the markets where regulators, developers, and competitors can see the architecture most clearly.

That, more than any benchmark, is the bet Federighi is making on June 8, 2026, the one Ternus will inherit.

Frequently Asked Questions

When did Apple announce the new Siri AI?

Apple previewed Siri AI on June 8, 2026, during the WWDC 2026 keynote at Apple Park. New Siri AI features are available for developer testing starting the same day across iOS 27, iPadOS 27, macOS 27, and visionOS 27, with a public beta next month and a free software update this fall.

Is Siri AI actually powered by Google Gemini?

Apple’s third-generation Apple Foundation Models are co-developed with Google and four of the five models were refined using outputs from Gemini frontier models, Apple AI lead Amar Subramanya confirmed. Apple says the AFM models are custom built for Apple Silicon and trained on proprietary data, and that Apple does not use Gemini as deployed to Google’s own customers, but AFM Cloud Pro does run on Nvidia GPUs inside Google’s cloud, accessed through Apple’s Private Cloud Compute.

Will the new Siri AI launch in the European Union?

Apple has said the new AI-powered Siri will not launch on iPhones and iPads in the European Union because of the Digital Markets Act, the bloc’s competition law, and there is no published timeline for a workaround. The restriction covers the EU specifically, not the United Kingdom or other European markets. Siri AI and the other new Apple Intelligence features will also not initially be available in China while Apple works through regulatory requirements.

Which devices will run iOS 27 and Apple Intelligence?

Apple said iOS 27 is eligible for every device from the iPhone 11 onward, a wider device base than any prior iOS release. The most powerful on-device Siri AI requires an iPhone Air, iPhone 17 Pro, iPhone 17 Pro Max, or iPad with M4 chip or later and at least 12 gigabytes of unified memory, or a Mac with M3 chip or later and at least 12 gigabytes of unified memory. Apple lists English, Danish, Dutch, French, German, Italian, Norwegian, Portuguese, Spanish, Swedish, Turkish, Vietnamese, Simplified Chinese, Traditional Chinese, Japanese, and Korean as supported languages at launch.