The WWDC coverage fixated on Siri. The more interesting read comes from Nate B. Jones, who argues the headlines — Siri AI, the Apple–Google Gemini alliance, private cloud compute expanding onto Google Cloud’s NVIDIA GPUs — are three faces of a single question:
When AI starts doing real work for you all day, where does that work physically run?
In a chatbot tab? In a hyperscaler where every serious task burns tokens, GPU time, and data-center power? Or increasingly inside the computer you already paid for — the phone in your hand, the chip in your Mac, with private cloud behind it only when the device isn’t enough?
That third answer is Apple’s whole vision, and it’s why the framing matters beyond Apple. The industry has quietly assumed inference is a cloud business — that the trillion-dollar value pools where the GPUs are. On-device AI is the contrarian bet: that the marginal cost of running an agent all day is a tax users won’t pay forever when the silicon in their pocket can absorb a growing share for free.
Jones frames it as the question that may decide the first AI trillionaire. That’s a creator’s flourish — but strip the hyperbole and a real strategic fault line remains. Apple’s structural advantage is that it sells the device, the chip, the OS, and the apps as one stack, so pushing work on-device costs it nothing and saves the user money. A cloud-first competitor has the opposite incentive: more inference in the data center is the revenue.
The caveat worth holding: “on-device” has hard ceilings. Memory, battery, and model size mean the heaviest reasoning still goes to the cloud, and Apple’s own private-cloud expansion onto NVIDIA GPUs is the tell. The honest shape isn’t device or cloud — it’s where the line sits, and the line moves every time the local chip gets faster. Watch the split, not the slogan.
Source: Nate B. Jones — WWDC Isn’t About Siri. It’s Jensen Huang’s Problem (YouTube). Analysis is his; the skepticism is mine.