Apple's New CEO Is a Hardware Guy. His First Job Is AI.

Apple made its succession plan official back in April and executed it on schedule: Tim Cook moved to executive chairman on September 1, 2026, and John Ternus — Apple's senior vice president of hardware engineering since 2021, and an Apple engineer since 2001 — became CEO. It's the calmest possible version of a CEO transition at a company this size. The harder problem lands in Ternus's lap the same week: Apple's AI strategy is no longer entirely Apple's.

A hardware chief taking over during Apple's least hardware-shaped problem

Ternus built his reputation shipping iPhones, Macs, and AirPods on a predictable annual cadence — the opposite discipline from the one Siri's rebuild has needed. Apple's in-house effort to give Siri a more conversational, personalized LLM backbone missed its original 2025 target and kept slipping through internal testing. The company's answer, confirmed by CNBC in January, wasn't to keep pushing the in-house model — it was to license one.

The $1B-a-year Gemini deal Ternus now owns

Apple struck a multi-year agreement to run a custom 1.2-trillion-parameter Gemini model from Google inside the next generation of Apple Intelligence and Siri, at a reported cost of roughly $1 billion a year. It's a striking move for a company that has spent a decade marketing on-device processing and data minimalism as a core differentiator — and it's now the load-bearing decision behind Apple's most visible AI product, made before its new CEO had the job.

The build-vs-buy lesson underneath the leadership story

Strip away the succession drama and there's a plain engineering-management lesson here, and it's the one worth sitting with if you're shipping AI features on a team a thousandth of Apple's size: capability and capital don't guarantee you can out-build a frontier lab on your own timeline. Apple has more engineers, more cash, and more incentive to keep everything in-house than almost any company on earth, and it still concluded that licensing Gemini beat waiting on its own model.

• Set a real deadline for your in-house model track, and pre-decide what "it's not converging" looks like before you're emotionally invested in it.
• Licensing a frontier model isn't a failure state — it's a normal architecture decision, made explicit and reversible rather than avoided out of pride.
• Whatever you don't build yourself, you still own the integration, the latency, the failure modes, and the trust story to your users — Apple can't outsource that part to Google.

What actually rests on Ternus

None of this means Ternus is a bystander. The Gemini deal buys Apple a competitive Siri; it doesn't buy Apple a roadmap. Whether Apple's next real product — something beyond a chat-capable assistant — comes out of Ternus's hardware instincts (he's spent his career on the teams building AirPods, wearables, and other device categories) is the actual open question, and it's one a licensing deal can't answer for him.