John Ternus Is Apple's Next CEO. The Real Story Is AI.

On January 12, 2026, Apple and Google put out a joint statement that neither company seemed excited to celebrate. Apple would pay Google roughly $1 billion a year to run a custom 1.2-trillion-parameter Gemini model (roughly eight times the size of Apple's own largest in-house cloud model) on Apple's Private Cloud Compute servers (first-party data centers running Apple silicon behind Apple's privacy architecture) to power a revamped Siri.^[4] Three months later, on April 20, 2026, Apple announced that John Ternus, a hardware engineer who joined the company in 2001, will become its next chief executive officer on September 1.^[1] The press treated the two announcements as separate stories. They are the same story.

Tim Cook's 15 years as CEO took Apple from a market capitalization (the total value of a company's outstanding shares) of roughly $350 billion in August 2011 to $4 trillion by October 2025.^[11] Services revenue (subscriptions, the App Store, iCloud, Apple Music, AppleCare) crossed $109 billion in fiscal 2025, up from under $20 billion when Cook started.^[12] Apple Silicon (the in-house-designed M-series chips that replaced Intel processors on the Mac) shipped on time and completed the transition by 2023.^[13] The case for Cook as a great operator is airtight. The case for Cook as the right CEO for the decade that started in November 2022 is not.

That is the month ChatGPT shipped.^[14]Apple Intelligence, the company's branded answer to generative artificial intelligence, was announced at the Worldwide Developers Conference (WWDC, Apple's annual developer keynote) in June 2024.^[3] The flagship feature, a personalized Siri that could take actions across your apps using knowledge of your personal context, was promised for spring 2025. As of April 2026 it still has not shipped.^[5] The Gemini deal is Apple's admission that the company was never going to ship it on its own model. Ternus is the chief executive you pick to finish that admission cleanly.

The reset started three months ago, not this week

Mark Gurman, the Bloomberg reporter whose Apple coverage is the closest thing to an insider newsletter, broke the Gemini story on November 5, 2025.^[4]A multi-year agreement. Roughly $1 billion a year. A 1.2-trillion-parameter custom Gemini model, eight times larger than Apple's own internal cloud foundation model (the company's biggest in-house large language model, or LLM, the class of model that powers chat assistants).^[4]The thing nobody at Apple wanted to say out loud was in the story's next paragraph. Anthropic's model had been rated better in Apple's internal bake-off (the competitive evaluation where candidate models are tested side-by-side against a common set of tasks). Google won on financial grounds, because Google and Apple already had a multi-billion-dollar search-distribution relationship and the Gemini deal slotted into it.^[15]

The best model did not win. The best business relationship did. A company whose whole brand is taste, integration, and picking the best technology in service of the product picked the second-best model because the first-best came from a company with no financial hooks into Apple. I have been watching Apple's AI answer get later for two years and this was the sentence that finally made it make sense.

OpenAI, the company that already supplies ChatGPT as an opt-in external assistant inside Siri, had been asked to become the custom model provider and said no. The Financial Times reported in January that OpenAI had taken “a conscious decision to not become the custom model provider for Apple” the previous autumn, because it was focused on building its own AI-native hardware device with Jony Ive.^[6] The iPhone maker had asked the two best AI labs in the world to power its flagship assistant. One said no. The other said yes for a fee.

Apple picked the executive furthest from the model layer

The succession shortlist, as reconstructed from Bloomberg and New York Times reporting, had three real names on it.^[19] Craig Federighi, senior vice president of software engineering and the public face of every Apple Intelligence stumble. John Giannandrea, who owned the machine-learning organization Cook is reportedly said to have considered replacing leadership of as early as 2024.^[20] John Ternus, senior vice president of hardware engineering. Jeff Williams, the chief operating officer long treated as the presumptive heir, stepped down from operational responsibilities in July 2025 and was out of the race by the time the board started picking.^[21]

Federighi runs the software organization that ships Apple Intelligence and is the public face of every stumble. Giannandrea ran the machine-learning organization that produced them. Cook picked the executive furthest from both. The succession is not engineering continuity. It is a statement about where Apple thinks its edge actually lives.

Ternus joined Apple in 2001 on the product design team, first assigned to the Apple Cinema Display.^[22] Before Apple his only job was at a small firm called Virtual Research Systems designing virtual-reality headsets, which reads in hindsight like a note-to-self about Vision Pro.^[22]At Penn he was a varsity swimmer, and he nearly broke the engineering school's CNC (computer numerical control) mill so often that his classmates took to calling him “Crash.” His senior project was a mechanical feeding arm for quadriplegic patients, operated by head movements.^[23] Every Apple product line launched since 2013 has his fingerprints on the hardware side, including AirPods, iPad Pro, every generation of iPhone after 2020, and the Mac as it transitioned from Intel chips to Apple Silicon.^[24]

Read his public quotes about AI and the strategy falls into shape. In an interview with Tom's Guide conducted in the weeks before the CEO announcement, Ternus said Apple does not “think about shipping a technology” but about “how can we leverage technology to ship amazing products.”^[25]On Apple Intelligence specifically: “If we're doing it right, people won't even really notice or think about it.”^[25] That is not a frontier-model builder talking. That is a distribution executive describing the AI layer as a hardware feature that should disappear into the device.

Apple's AI problem is distribution, not modeling

Draw Apple's bet as a diagram and the argument is obvious. On the left are the assets no competitor can replicate. In the middle is the pitch. On the right is what ships to the user, most of which is code written by someone else.

The capex (capital expenditure, the money a company spends building physical infrastructure) numbers make the same argument, in one glance.

The Ruoming Pang departure sharpens it. Pang ran Apple's roughly 100-person Apple Foundation Models team, the group whose work powers email summaries, priority notifications, and Genmoji (Apple's user-created emoji feature). Meta paid over $200 million over multiple years to poach him in July 2025. Apple did not try to match.^[7] If Apple genuinely believed its internal foundation-model program was a winning bet, the reported package Meta offered is exactly the number a company writes to keep the leader in place. Apple wrote nothing. Four months later the Gemini deal was on the front page.

The three decisions Ternus has to make in year one

Forget the hundred-day plan. Ternus has three calls to make before the end of 2027.

1. Ship the Gemini-backed Siri on time.The personalized Siri has been promised and slipped three times. It is now targeted for iOS 27, shipping in the fall of 2026, which means roughly Ternus's second month as chief executive. A fourth slip would break the public narrative that the Gemini deal fixed the problem. Shipping on time with Gemini under the hood is the table-stakes decision that lets him make every other decision.
2. Name a new head of AI with real scope and real budget. Giannandrea is gone. The Siri revamp reports to Rockwell, who reports to Federighi. That is a structure that made sense when Apple was treating Siri as a software problem to be patched. It is not a structure for an AI era. Whoever Ternus names (internal hire, external hire, a combined chief AI officer with the engineering budget to hire back the Ruoming Pangs of the world) will tell you whether the company is trying to catch up at the model layer or has written that off.
3. Decide what Vision Pro is.This is the third AI decision wearing a hardware costume, which is exactly the kind of call a hardware CEO gets paid to make. Vision Pro shipped in February 2024 at $3,499. IDC estimated roughly 390,000 units sold in the launch year and 80,000 to 90,000 in 2025, with Luxshare reportedly halting production early in 2025 and Apple cutting the product's US and UK ad spend by more than 95%.^[28]Ternus told Tom's Guide in April that Apple is “still very much in the early innings of spatial computing.”^[25] The honest answer is that Vision Pro either becomes a Mac Pro-shaped niche product that Apple keeps alive at low volume and high margin, or gets the full iPhone-scale investment to become a mass-market headset by 2028. Both are defensible. Pretending neither call has been made is not.

What would change my read

I have written this piece as if the strategy is settled. It mostly is. But three things would tell me I read the Ternus pick wrong.

• A named frontier-model leader hired at a package north of $200 million, on the record, within 90 days of September 1. That is not a bolt-on. That is a company saying it is still trying to build its own model, and treating the Gemini deal as a two-year bridge rather than a destination.
• A Ternus all-hands, a shareholder letter, or a WWDC 2027 keynote that explicitly commits new foundation-model capex at hyperscaler scale. The tell is the dollar figure being said out loud. Cook never said it. Ternus saying it would rewrite the thesis.
• A successor to Giannandrea who reports directly to the CEO rather than to Federighi, and who comes in with a real compute budget.

Absent all three, the bet is already made. Anthropic and OpenAI will keep training frontier models. Google will keep renting Apple a version of Gemini it tunes specifically for Siri. Apple will keep owning the device, the silicon, the ecosystem, and the customer relationship. Three years ago everyone assumed Apple would build the best AI on the planet. What actually shipped is a billion dollars a year to a competitor and a chief executive whose career was built shipping hardware. Ternus inherits a company that already made its choice. His job is not to reverse it. His job is to make it look deliberate.

The hardware was never Apple's problem. That is exactly why a hardware engineer is the person Apple picked to run the rest of the decade.