✦ Siri AI & the Gemini DealThin moat

The most consequential sentence Apple published this decade was a joint press release: from January 2026, Siri runs on Google's Gemini¹. The terms tell the story — a reported billion dollars a year for a custom 1.2-trillion-parameter model, running in Apple's own Private Cloud Compute data centers so the queries never leave Apple's walls² — and so does the calendar: first Gemini-powered features in iOS 26.4 in spring 2026, a rebuilt assistant unveiled at WWDC, full rollout arriving with iOS 27 in the fall³. Apple spent three years promising an in-house AI Siri and shipping delays instead; renting the frontier was the honest correction, and a shrewd one — better a rival's excellent brain in your pocket than your own mediocre one.

The logic is distribution. An assistant nobody chose becomes an assistant everyone has, on well over a billion daily devices, the day the update lands. The risk is equally plain: when the brain belongs to a rival, the differentiation must live around it — in the privacy architecture, the device integration, the things Google cannot copy without Apple's hardware. Watch two numbers. First, whether Siri usage genuinely moves through the first holiday season with the new assistant everywhere. Second, whether Apple's own foundation models ever graduate from understudy to lead. A bridge is a fine thing; a permanent crutch would be Apple's first strategic dependency since the Intel years — and Intel, recall, was eventually shown the door.