On September 1, 2026, John Ternus stepped into Apple's corner office, taking charge of the world's most valuable technology company along with a strategic question Tim Cook had already settled for him. Would Apple build a frontier AI model? Cook said no. Now Ternus has to make that bet pay off
Nearly every company mentioned alongside Apple has a model of its own. Google has Gemini. Microsoft holds a deep financial stake in OpenAI and runs Copilot across its products. Meta releases Llama to the open-source community. Amazon offers Titan and Bedrock. Anthropic has Claude, and xAI has Grok. Even the challengers in the frontier race now appear on this list.
Apple doesn't. This is a company at or near the top of the global market capitalisation rankings, with a platform of roughly one billion active users and a device in more than one in five pockets worldwide. And it has deliberately chosen to stay out of the race for frontier AI capability. In January 2026, Apple agreed to pay Google roughly one billion dollars a year for a custom version of Gemini to power a rebuilt Siri, with ChatGPT picking up overflow queries through an existing integration. Apple's most important AI features now run on models it didn't build.
Ternus inherited that arrangement on September 1, along with everything else. His first keynote as CEO, an event Apple called "Surprise and Shine" on September 9, offered the clearest early hint of what his tenure will look like. It wasn't a break from Cook's strategy. It was a defence of it.
What changed on September 1
Apple announced the succession on April 20, 2026, four days after a board vote. The transition then ran for four months, an unusually long runway meant to signal stability. Every press release between April and September still listed Cook as CEO. The September 1 start date was timed to fall about two weeks before Apple's usual fall iPhone launch window, so Ternus's first keynote would come just eight days into his tenure.
Cook, 65, steps into the newly created role of executive chairman. He keeps a seat at the table, focused on policy engagement with governments worldwide, a meaningful job given the antitrust scrutiny Apple faces in Washington and Brussels at the same time. The company described the handover as the product of a "thoughtful, long-term succession planning process." Cook called it "the greatest privilege of my life."
Ternus, 51, was Apple's senior vice president of hardware engineering and has spent more than two decades at the company, having joined in 2001. He was the executive behind Apple Silicon, the M-series chips that moved the Mac off Intel processors and gave Apple control over its own performance and power envelope. He also led the engineering teams behind several generations of MacBook redesigns, the Mac Pro, and the Vision Pro headset. Before the announcement he was seen as a technical builder rather than a public face, although Bloomberg's Mark Gurman had named him the likeliest successor for years, and his public appearances had grown steadily as the succession process matured.
Who John Ternus is
The label that has stuck to Ternus since the announcement is a fair one: he is an engineer's engineer. He talks about products in terms of what they do and how they work, rather than what they symbolise or how they fit a narrative. His predecessors include Steve Jobs, a product visionary, and Tim Cook, a master of operations and logistics. Ternus is neither, at least in the brand-archetype sense. He is someone who understands deeply how things are made and what it takes to make them well.
That background matters for the AI question. The most candid version of Apple's AI strategy, which Ternus has already described in public, is really a hardware argument. The iPhone is the best AI device because of what it can do on-device: the Apple Silicon Neural Engine running models without sending data to a server, the camera and microphone array that makes multimodal input practical, and the deep operating-system integration that a web-based assistant can't replicate. At his first keynote, Ternus described the ideal AI device as an "intelligent personal hub" with broad understanding of a user's personal context, always present, capable of on-device inference, with seamless integration across apps and services. "In other words," he said, "you would arrive at something remarkably familiar, because there's no product in the world better designed to be your intelligent personal hub than iPhone."
That is the case Apple is making, and it is a product argument rather than a model argument. Whether the distinction holds up is the central question of the Ternus era.
The AI situation he walks into
Apple Intelligence, the umbrella brand for Apple's generative AI features, was unveiled at WWDC 2024 with the promise of a transformed Siri that could act across apps, hold multi-turn conversations, and understand personal context. The 2025 rollout moved more slowly than announced, and the features that shipped were thinner than what had appeared on the slides. Through its first year, press coverage and users alike kept repeating the same verdict: it hadn't delivered on its promise.
The January 2026 Gemini licensing deal was Apple's practical admission of that gap. It handed Apple a capable large language model to put behind Siri without training one itself. It also created a structural dependency that every Apple analyst has flagged: a Siri built mainly on licensed technology only succeeds commercially if it still feels distinctly like an Apple product, and not merely a wrapper around Gemini with an Apple logo attached.
The September 9 keynote reported a record quarter, with $109.4 billion in revenue, up 16% year over year. Apple's stock still fell 6% on the announcement. Investors weren't after revenue figures. They wanted an AI business model, and Apple hasn't yet laid one out that explains where the next decade of growth comes from. The hardware upgrades are real, and the iPhone Ultra foldable, priced from nearly $2,000, pushes the lineup upmarket. But premium hardware pricing is not an AI strategy.
Apple's deliberate choice not to build a frontier model
It's worth being precise about what Apple has and hasn't decided. The company has chosen not to compete in the race to build and operate a frontier general-purpose large language model. That is a specific decision with specific reasons, not an oversight or a gap in capability.
As of September 2026, the frontier race is expensive, capital-intensive, and skewed toward winners. The three leading labs in frontier capability, OpenAI, Google DeepMind, and Anthropic, run on compute investments measured in billions each year. Their returns come mainly from API access and enterprise licensing, not from consumer hardware margins, and Apple's business model doesn't naturally plug into that revenue structure.
On-device inference, which Apple has genuinely advanced through Apple Silicon, is a real and defensible technical moat. A model running locally on a Neural Engine fits Apple's privacy story in a way that sending every query to a cloud endpoint doesn't. It cuts latency, works without a network connection, and protects the user data Apple has spent a decade marketing itself as not monetising. Those are not trivial advantages.
The honest limitation is that on-device models, bound by power and memory budgets, are not frontier models. They are useful, fast, and private, but they aren't the tools that draft a legal brief, design a product spec, or autonomously run a multi-step agentic workflow. For those jobs, Apple's users are now routed through Gemini, which means routing through Google, simultaneously Apple's biggest rival in mobile operating systems and its new AI supplier.
Why the bet is harder to defend in 2026 than in 2024
When Apple Intelligence was announced, the case for Apple's approach was reasonable on three counts: the frontier race was still unsettled, on-device capability was improving fast enough to catch up on many tasks, and privacy was a real differentiator with genuine consumer value. All three pillars are under more pressure today than they were two years ago.
The frontier race has narrowed to three labs that are clearly ahead of everyone else. The gap between frontier and on-device models hasn't closed. On the tasks that matter most to enterprise and power users, it has widened. And the privacy argument, though still valid, has been complicated by the Gemini deal itself: Apple's privacy story now has to explain how a billion-dollar licensing arrangement with Google protects user data as well as an on-device model would.
Before Ternus took over, Gene Munster of Deepwater Asset Management set the bar plainly: the rebuilt Siri needed to be a "10 out of 10" when it arrived. That standard became Ternus's first performance review before he had run a single meeting as CEO.
"The challenge is not simply catching up to OpenAI or Google on raw model capability. It is proving that Apple can build products people trust and prefer around someone else's AI model, without losing the product experience, cost control or differentiation that outside dependency usually threatens."
PYMNTS, Apple's New CEO Inherits an AI Strategy Built on Outsiders, September 2026
The three paths available to Ternus
From where Ternus stands, no version of Apple's AI story is comfortable. Every direction has a real cost.
Stay the course with the hardware-first strategy. Deepen Apple Silicon, expand Private Cloud Compute, and keep licensing frontier capability from Google and OpenAI for the tasks on-device models can't handle. This path causes the least near-term disruption, and Ternus has already hinted it's his preference. The risk is that it relies on competitors staying willing to license their best technology to Apple on acceptable terms, and on Apple's hardware premium surviving in a world where AI assistance becomes a baseline expectation instead of a differentiator.
Build or acquire frontier model capability. This is the road Apple has conspicuously avoided. Training a competitive frontier model from scratch would demand heavy capital investment, a research culture Apple has partly built but not fully, and a change in the company's relationship with its own technology stack. Acquisition is faster, but the companies with competitive frontier models today aren't available at prices or on terms that fit Apple's structure.
Reframe the AI question around agentic hardware. Camera-equipped AirPods, smart glasses, a robotic home display, the Apple Watch, the iPhone as an agentic action centre. If the next AI era is defined by devices that sense, act, and integrate with the physical world, rather than by models that reason in a chat window, Apple's hardware distribution and device network become a stronger asset. This is the speculative version of the hardware-first bet, and the one that leans most directly on Ternus's engineering background.
What kind of CEO this moment actually needs
Every Apple CEO transition has raised the question of whether the successor can measure up to the predecessor, but the questions about Ternus are different in kind from those that followed Cook in 2011. When Cook replaced Jobs, people asked whether he could sustain a culture of product vision. His answer was fifteen years of consistent execution, with Apple's market capitalisation growing from $350 billion to roughly $3.5 trillion under his leadership.
With Ternus, nobody is asking whether he can maintain Apple's product quality or operational efficiency. His record on Apple Silicon settles the first point. The real question is whether an engineer at the top of the world's most valuable company can define a compelling vision for Apple's AI era when its AI runs on competitors' models.
"We make the best hardware describes the past, not a strategy for the next decade."
PYMNTS, Apple's New CEO Has a Hardware Event and an AI Problem, September 2026
That's a harsh line, but it captures the real tension. Nobody disputes Apple's hardware quality. What remains open is whether Apple can turn that quality into a lasting AI moat when AI capability increasingly sits in models Apple doesn't own, and the answer will decide whether Ternus's tenure resembles Cook's or turns out to be something more complicated.
The final read
John Ternus is a serious executive running a serious company. Choosing not to compete directly for frontier model capability is not obviously wrong. It is a bet on a different theory of where AI value ultimately accrues. If that value ends up concentrated in the physical computing layer, meaning the sensors, the inference chips, the device integration, and the privacy architecture, then Apple's position is strong and Ternus's engineering instincts are exactly right for the moment.
If instead AI value concentrates in the model layer, with whoever trains the most capable general-purpose reasoning engine and controls the user relationship through it, then Apple's dependence on Gemini and ChatGPT is a structural weakness that no hardware upgrade cycle can fully fix.
The bet was placed before Ternus arrived, and his job is to make it pay. It is still early days in September 2026. The first earnings call, users' real reaction to the rebuilt Siri, and how long Google stays a willing partner on terms Apple can live with are all still ahead of him. One of the most interesting leadership assignments in tech in years has just begun. The clock is running.
Enjoyed this read? Be sure to follow me for more stories that make people feel things. You can also connect with me on LinkedIn or Telegram @kawsarlog for more updates on my work.
I do freelance scraping and data automation projects. Feel free to contact me on LinkedIn or email at hi@kawsarlog.com
Top comments (0)