Apple's Siri AI: a redemption arc, running on Google's engine

Apple's 2024 Siri promise — a deeply context-aware assistant that understood your apps, your life, and your screen — became the company's most public AI embarrassment when it slipped, then slipped again. The gap between the demo and the delivery was visible enough that analysts started asking whether Apple had fallen permanently behind on AI. WWDC 2026 was the attempt to answer that question. Siri AI is Apple finally shipping, or at least previewing, the assistant it described two years ago — and the story underneath it is as interesting as the feature itself.

The pitch is a step change in assistant capability, not an incremental update. Siri AI understands personal context — your calendar, your messages, your ongoing projects — without you having to paste it in. It has on-screen awareness, reading what's in front of you at any moment. And it can take multi-step, cross-app actions: add the acts you want to your calendar from a festival poster on screen; have Phone pull useful context from Mail and Messages during a live call; compose a message drawing on emails from earlier in the day. This is the assistant that bridges the gap between 'set a timer' and 'actually help me run my day'.

Apple is shipping all of this inside iOS 27, iPadOS 27 and macOS 27 (the latter named 'Golden Gate'), available this autumn on supported recent hardware — broadly the iPhone 15 Pro and newer, iPhone 16 family, recent iPads and all Apple Silicon Macs. The update arrives alongside a wider Apple Intelligence refresh. The question is not whether Siri AI is a real product now — it looks like one — but what's powering it, and what that means for Apple's story.

Here is the throughline. Reporting from multiple outlets indicates that Siri AI is built on a custom Google Gemini model, with Apple paying Google on the order of $1 billion a year for the arrangement (TechCrunch, 9 June; attributed, not officially confirmed). The company that built its entire brand identity around vertical integration — hardware, software, silicon, and now AI — is, at the core of its flagship intelligence feature, renting the model from its most direct AI competitor. That is a pragmatic and arguably correct call. It is also a strategic tell of considerable significance.

Consider the comparison with how Apple has handled similar dependencies in the past — and how differently it treats them:

The historical pattern is that Apple tolerates high-value external dependencies as a bridge — but builds hard to eliminate them once it has the scale and capability, as it did with Intel. The open question is how long that takes with foundation AI models, which demand compute investment and research depth at a scale even Apple finds hard to replicate quickly. The reported ~$1bn/year, if accurate, is money Apple pays out rather than captures — and the company will be acutely aware of that.

Two geo-restrictions fracture the 'it just works everywhere' promise that is perhaps Apple's most durable brand asset. China is excluded at launch while regulatory review runs its course — and China is not a marginal market for Apple; it is one of its three largest. The EU gets a limited version, largely confined to macOS and visionOS, as Apple navigates Digital Markets Act constraints that complicate how deeply it can integrate an AI assistant across system functions. Neither restriction is Apple's invention — both are imposed from outside — but they reduce the total-addressable-impact of the launch materially, and the company cannot fix either on its own timeline.

Then there is the context of the keynote itself. WWDC 2026 was Tim Cook's final WWDC as CEO, with John Ternus confirmed to take over the chief executive role on 1 September. Apple elected not to make the leadership transition the headline, framing the keynote around products. But the subtext was present throughout: the company wanted this launch, and specifically Siri AI, to be the image of forward momentum it hands to Ternus. The strategic architecture of Apple Intelligence is something the incoming CEO will own for years. What he inherits is a product that looks real, built on a dependency that remains opaque.

Three things matter in the 12 months that follow. First: whether the autumn shipping dates hold. Apple's 2024 promise slipped; a second slip would do lasting damage to the AI credibility it is trying to rebuild here. Second: the China regulatory timeline. Apple's most sensitive market is the one from which Siri AI is absent at launch; the duration of that absence is both a commercial and a geopolitical variable Apple cannot control. Third: how Apple characterises the Gemini relationship at scale — in privacy disclosures, in engineering deep-dives, and in how it responds when competitors or regulators surface the arrangement publicly. The feature is impressive. The dependency is real. Both of those facts will drive the next chapter of this story.