Apple's chief executive changed hands on schedule Sept. 1: John Ternus, the company's senior vice president of hardware engineering for the past five years, became CEO as Tim Cook stepped up to executive chairman, ending a 15-year run at the top. The board approved the succession unanimously back on April 20, so the handoff itself was never in doubt -- the real news is what Ternus now owns: an AI strategy built, more plainly than Apple has ever conceded in public, on someone else's model.
Ternus, 51, has been at Apple for 25 years -- joining the product-design team in 2001, becoming vice president of hardware engineering by 2013 and senior vice president in 2021 -- and led the physical side of AirPods, Apple Watch and Vision Pro, plus the multi-year transition from Intel chips to Apple's own silicon. "Always assume you're as smart as anyone else in the room, but never assume that you know as much as they do," he has said of his own approach to leadership -- a habit of deference that Apple's AI effort, badly behind rivals' timelines for two straight years, has had little choice but to adopt as well.
The team under him has already been rebuilt once. Apple announced in December that John Giannandrea, its AI chief since 2018, would retire in spring 2026, splitting his organization between two other executives and handing Apple's foundation-model and AI-safety work to a new vice president, Amar Subramanya -- who spent 16 years at Google, including a stretch running engineering for Google's own Gemini Assistant, before a brief stop at Microsoft. Siri itself had already been pulled from Giannandrea's group in March and handed to Mike Rockwell, the executive who built Vision Pro. The person now shaping Apple's foundation models, in other words, spent most of his career building the product Apple just agreed to license.
The clearest evidence of that is the deal now sitting at the center of Apple's rebuilt Siri. Rather than build the underlying model itself, Apple agreed on Jan. 12 to pay Google roughly $1 billion a year to license a custom, 1.2-trillion-parameter version of Gemini -- about eight times the size of Apple's own largest in-house model -- to handle cross-app actions and multi-turn conversation that Apple's models reportedly couldn't yet deliver on Apple's own timeline.
The handoff, in short
- Outgoing CEO
- Tim Cook
- Incoming CEO
- John Ternus
- Announced
- April 20, 2026
- Effective
- Sept. 1, 2026
- First keynote as CEO
- Sept. 9, 2026
How that Gemini model actually reaches a user's phone is a routing decision, not a single wholesale swap, and it's the clearest structural fact of the strategy Ternus now runs.
How a rebuilt Siri actually answers a question
- Handled on-device by Apple's own Apple Foundation Models
- Routed to Apple's Private Cloud Compute -- still Apple's own infrastructure
- Handed to the licensed 1.2T-parameter Gemini model
- Delegated to ChatGPT -- now Siri's secondary outside option, not its primary one
Set side by side against the three companies chasing Apple hardest, the shape of that bet gets easier to see.
Two ways to build an AI assistant
| License it -- Apple | Build it -- OpenAI, Google, Meta, xAI | |
|---|---|---|
| Annual AI infrastructure commitment | ~$1B/year, Gemini license fee | $100B+ each in owned data-center capex |
| Underlying model | Licensed, 1.2T-parameter Gemini variant for hard queries | Built and trained in-house |
| Where most processing happens | On-device or Apple's own Private Cloud Compute first | Their own cloud infrastructure throughout |
Cook, on his final earnings call as CEO, was candid about the one number Apple still hasn't nailed down: what running Gemini-scale AI actually costs as usage grows.
“There's also the ability when people use it a lot for them to move up on an iCloud Plan as well, and so what the balance of that is, is a bit uncertain at the moment.” — Tim Cook, then-CEO, Apple
That uncertainty sits inside a broader bet, and it's a bet Apple is making almost alone among its peers: that the capex-light path -- licensing intelligence instead of owning the infrastructure that produces it -- still wins, because owning the chip inside the phone matters more than owning the model answering the question on it. Nothing about that bet requires the licensed model to be great, only cheap enough and fast enough not to embarrass the hardware it runs on.
- About $1 billion a year in licensing revenue and a default position inside the world's largest smartphone install base.
- Its ChatGPT integration, once Siri's main outside option, is now the secondary fallback behind Gemini.
- Inherits an AI strategy built on a rival's model at the exact moment he's judged on whether it finally ships.
- Avoids the $100B+ in data-center capital its rivals are committing -- though Cook himself flagged the compute-cost math as still unresolved.
What doesn't change Sept. 1 is the deadline pressure. Siri's AI overhaul has already slipped twice -- from an original spring-2025 target to a promised iOS 26.4 window this past spring, and now to iOS 27, expected this fall alongside the hardware Ternus unveils at his first keynote as CEO on Sept. 9. That keynote is confirmed to cover the iPhone 18 Pro and Apple's first foldable; whether it also finally puts a working, Gemini-backed Siri in front of a live audience is not yet confirmed by Apple itself. Whichever way it goes, the promise is attached to Ternus's name now, not Cook's.
- John Ternus became Apple's CEO on Sept. 1, succeeding Tim Cook after a 15-year run.
- Cook moves to executive chairman; the board approved the succession back on April 20.
- Ternus inherits Siri's rebuild, which routes hard queries to a licensed Google Gemini model.
- Apple pays Google roughly $1 billion a year for a custom, 1.2-trillion-parameter Gemini model.
- Caveat: Siri's AI overhaul has already slipped twice and now targets iOS 27 this fall.