John Ternus inherits a $4 trillion AI problem

TL;DR

• Tim Cook's exit was last month. The story now is what the new CEO has to fix, and it is narrower than the coverage suggests.
• Apple's AI gap is real: Siri still awful, Apple Intelligence adoption tepid, the flagship Personal Context feature never shipped, eight-plus senior AI researchers gone to Anthropic and DeepMind in twelve months.
• The durable advantages are hardware: M-series silicon, Neural Engine, on-device inference, distribution, and user-data trust.
• Five prioritised in-tray items: decide whether Apple is a model company; fix Siri on a 2026 clock; retain silicon and AI talent; relaunch or retire the Apple Intelligence brand; renegotiate the OpenAI relationship.
• WWDC in June is the forward indicator everyone should watch. If R&D as % of revenue does not move from ~8% toward ~11%, the market will repricing.

The job, narrowed

John Ternus is not inheriting Apple's supply chain — Sabih Khan runs that, and runs it well. He is not inheriting the services business — Eddy Cue has it. He is not inheriting hardware direction — that's his own background, his own instinct, his own muscle memory.

What he is inheriting is a company whose single biggest strategic liability in 2026 is that it has publicly fallen behind on the one technology shift that now defines the industry. Everything else about Apple works. The AI story does not.

Where Apple actually stands

Let's be specific, because the coverage often isn't.

Siri. Still the punchline it was in 2017. The promised overhaul has slipped twice. Internal timelines now point to 2027 for the rebuilt assistant.

Apple Intelligence. Launched 2024 with Cook personally on stage. Two years on, adoption is tepid, writing tools are rarely better than free-tier ChatGPT, and the flagship Personal Context feature never shipped. The OpenAI partnership papered over the gap but handed OpenAI the customer relationship — users know they're talking to ChatGPT, not to Apple.

Foundation models. Apple has models. They are behind the frontier. The lab has bled senior talent to Anthropic and Google DeepMind at a pace that rattles everyone who tracks it — at least eight named principals in twelve months, per Bloomberg.

Silicon. The one genuine asset. M-series chips and the Neural Engine give Apple an on-device inference advantage no competitor can match at the platform level. This is real, durable, and the foundation Ternus actually has to build from.

Distribution. Billions of devices capable of running meaningful on-device AI, paired with a user base that trusts Apple with the most sensitive data of any tech company. A moat the AI-native labs do not have and cannot buy.

Why the moat is not enough

The moat buys time. It does not buy direction.

Apple's product position has always been we are late, but we are better when we arrive. It worked for MP3 players, smartphones, watches, AirPods, silicon. It requires one condition: the eventual Apple product must be materially better than what's on the market when it ships.

That is the bet Ternus has to deliver, and the market is no longer giving Apple unlimited time to place it. The frontier moves every six months. Each delay widens the capability gap that has to be closed at ship.

The in-tray, prioritised

Priority one — decide whether Apple is a model company. The current strategy is a compromise: small on-device models plus OpenAI for the hard stuff. This is unstable. Either Apple invests $40–60 billion over three years to credibly train frontier models in-house, or it commits fully to a platform strategy and makes AI model access the user's choice (the way it handled default browser and search). Both defensible. The current middle path is not. Ternus has to pick.

Priority two — fix Siri for real, on a short clock. Not the 2027 date. A credibly improved Siri at WWDC 2026, shipping in iOS 20 in September. The rebuilt assistant is a commercial requirement, not an R&D objective. If Amazon Alexa+ or Google's Gemini assistant lands first on a meaningful device, Apple loses an interaction category it has owned since 2011.

Priority three — retain silicon and AI talent. Golden handcuffs, restructured comp, visible project ownership. Departing researchers told reporters the same thing: not enough product impact, not enough autonomy, not enough compute. Fixable with executive will. Cook did not fix them. Ternus has to.

Priority four — Apple Intelligence branding. The current brand is wounded. Most users who have tried it don't reach for it. Ternus will need to either relaunch with a genuinely different product, or retire the name and rebuild under a new one. Holding the current position through another underwhelming release is the worst of the three options.

Priority five — the OpenAI relationship. Renegotiate or replace. Cook's deal gave OpenAI enormous distribution for modest revenue share. In a post-Mythos world, dependency on any single lab is a board-level risk. The pragmatic move is multi-vendor — Gemini, Claude, possibly a Chinese model for the Chinese market — with user choice as the headline consumer feature.

The hype to deconstruct

The prevailing narrative — "Apple is a hardware person, software is the gap, Ternus is a mistake" — has it backwards. The AI question for Apple in 2026 is not a software question. It is a systems question: silicon, thermals, battery, memory bandwidth, on-device inference, privacy architecture. These are hardware problems in software clothing. Ternus has shipped every generation of Apple silicon. He knows what he's holding.

He also has the institutional permission Cook did not. Cook's exit was widely read as Apple admitting the AI gap needed a different kind of leader. Ternus can spend political capital Cook had depleted.

Cross-layer implications

The AI reset has second-order effects beyond Cupertino: on semiconductor roadmaps (TSMC's N2 line; Apple's AI compute now competes with its own iPhone compute for wafer allocation), on the Asia-Pacific developer ecosystem (Apple-platform AI apps can't credibly compete with cross-platform agent products unless the framework changes), and on the $120bn+ services growth story that assumes continued ecosystem stickiness in a world where users start talking to agents, not apps.

What this means for you

• Apple shareholder: next two earnings calls plus WWDC in June are the forward indicators. Watch R&D as a percentage of revenue — it needs to move, and the market will punish Ternus if it doesn't. Current R&D runs ~8%. A credible AI push takes it to 11–12%. Consensus estimates have not priced that in.
• Business on Apple hardware (most Australian professional services): expect a significantly better on-device AI story at WWDC in June. Changes what you can deploy locally for privacy-constrained workloads. Brief your CIO now; the procurement cycle that matters is calendar-year 2026 H2.
• Developer on the Apple platform: the AI APIs you build on in the next twelve months will be materially different from today's. Current Apple Intelligence framework unlikely to survive as-is. Architect loosely. Plan for a second migration.
• AI engineer or researcher currently at Apple: the next six months is the window in which your leverage is highest. Ternus needs the team. Use the moment.
• Consumer: no action today. Decide with your wallet at the iPhone 18 launch in September. If Siri still feels like Siri, buy the phone for the camera. If Siri feels genuinely new, Apple has turned the corner. You'll know.

Uncertainty ledger

• Ternus's actual org design — rumoured elevation of the AI group to a direct report; possible new Chief AI Officer role outside the engineering org.
• Fate of the internal Ajax / foundation model programme. Three scenarios: accelerate, partner-out, discontinue. Ternus has not signalled.
• OpenAI contract renewal date and whether Apple has notified OpenAI of any change in intent.

• Whether Apple attempts an acquisition. Perplexity, Character.ai, Anthropic all rumoured; all implausible at current valuations. Mistral at a different price is not implausible.

Bottom Line

Ternus has one job, one eighteen-month window, and two genuine assets — silicon and trust — that cannot be bought by the labs ahead of him. He does not need to catch Anthropic on frontier capability. He needs to ship the first believable AI product Apple has had since the launch of Siri, on-device, privacy-native, on hardware no one else can match. If WWDC in June shows he understands that, the $4 trillion holds. If it shows another polished roadmap deck, the market reprices by year-end.

Written in the tradition of — E.

Sources

• Tier 1: Reuters — Ternus appointment and strategic priorities coverage (April 2026); Bloomberg — Apple AI talent departures reporting (2025–2026); Apple — Q2 FY2026 earnings call commentary
• Tier 2: Forbes The Prompt — Apple AI reset analysis (April 2026); Business Insider — Apple Intelligence adoption data