Answer-First Lead
Tim Cook takes the WWDC 2026 stage on June 8 for his final keynote as Apple CEO — John Ternus takes over September 1. The real story isn’t the goodbye. It’s the $1 billion-per-year Google Gemini deal powering a rebuilt Siri, the Extensions framework that lets any AI app plug into 1.4 billion iPhones, and the two years of broken promises Apple finally has to deliver on.
🔍 THE BOTTOM LINE
Apple’s AI moment of truth arrives at the exact moment its leadership changes hands — and the company is betting a billion dollars a year that Google’s model can make Siri matter.
The Last Keynote
Multiple reports confirm this will be Tim Cook’s final WWDC keynote as CEO. Cook steps down September 1 after more than 14 years at the helm — a tenure that saw Apple’s market cap grow from $350 billion to $4 trillion. John Ternus, currently SVP of Hardware Engineering, takes over.
Bloomberg’s Mark Gurman reports that Cook will open the keynote before handing the bulk of the presentation to Craig Federighi. Mike Rockwell — the Vision Pro chief who was quietly moved to lead the Siri overhaul — is expected to introduce the rebuilt assistant.
Here’s what makes this transition unusual: Ternus isn’t a software guy. He’s a hardware engineer who joined Apple in 2001 and spent two decades building iPhone, iPad, AirPods, and Mac. His appointment signals Apple believes the AI battle is won on silicon, not software — and he’s expected to push Apple more aggressively toward AI while expanding into smart glasses, robotics, and new product categories. It also follows the collapse of Apple’s OpenAI partnership — Apple’s AI strategy is now Google-first, not OpenAI-first.
Why it matters: The CEO transition at the exact moment Apple makes its biggest AI play isn’t coincidence. Cook’s legacy was supply chain excellence; Ternus’s mandate is AI relevance. If Siri flops again, the new CEO starts with a credibility crisis.
The $1 Billion Google Deal Powering Siri 2.0
The rebuilt Siri runs on a custom 1.2-trillion-parameter Google Gemini model licensed under a multi-year deal costing roughly $1 billion per year, per Bloomberg. Gene Munster of Deepwater Asset Management estimates the total deal value at up to $5 billion.
Key details:
- 8× larger than Apple’s own 150-billion-parameter cloud model
- Uses a mixture-of-experts architecture that activates only relevant parameters per query, keeping latency competitive despite the size
- Apple’s own ~3-billion-parameter model handles simple requests (timers, smart home) locally on-device
- Runs through Apple’s Private Cloud Compute (PCC) framework on Nvidia B200 Blackwell chips inside Apple’s PCC nodes — model weights stay on Apple hardware, not Google’s
- Google is contractually prohibited from using Siri interactions to train future Gemini models
- An independent ACM conference analysis in June 2026 confirmed Apple’s three core PCC privacy claims
The financial relationship is a fascinating reversal. Google already pays Apple an estimated $20 billion/year to remain Safari’s default search engine. Now Apple is paying Google $1 billion/year for AI. A DOJ antitrust challenge filed in February 2026 contests the Safari deal — and the AI licensing adds another layer to an already tangled relationship.
Why it matters: Apple isn’t building frontier AI — it’s buying it. The question is whether $1 billion a year buys relevance, or just a more expensive version of “sorry, I didn’t quite get that.”
The $250 Million Overhang
Apple arrives at WWDC 2026 with a credibility problem. At WWDC 2024, it showed a Siri capable of seeing what’s on screen, accessing personal data from Mail and Messages, and executing multi-step tasks across apps. Apple promoted those features in iPhone 16 ads featuring Bella Ramsey. Ran them for months. Then quietly pulled them.
None of the advertised features had shipped.
Apple acknowledged the delay in March 2025. A class-action lawsuit in the Northern District of California accused the company of false advertising. In May 2026, Apple settled for $250 million — $25 per eligible device, up to $95 if claim volume is low. A separate suit led by South Korea’s National Pension Service argues the delays caused stock-market losses.
Now Apple is presenting the same capabilities on the same slide, two years later.
Why it matters: When you’ve already paid $250 million for overpromising AI, the audience for your next AI promise is going to be sceptical. Apple needs to ship — not announce — today.
The Extensions Framework: App Store Economics, Round Two
iOS 27 introduces the Extensions framework — a system that lets any AI chatbot from the App Store plug directly into Siri, Writing Tools, and Image Playground. ChatGPT has early access. Gemini and Claude are confirmed next. Any AI app shipping through the App Store gains eligibility once Apple opens the framework.
The commercial catch: Apple hasn’t clarified whether it will charge a commission on Siri integrations. The Information reported that Apple told prominent developers it won’t charge “in the early stages” but hasn’t ruled out introducing one later. The structure gives Apple surface area to collect up to 30% commission on AI subscription revenue.
Some of the platform’s largest developers are waiting for pricing clarity before committing. In China, the dynamic is sharper — developers are hesitating over fee ambiguity layered on top of local AI data regulations.
Meanwhile, the technical barrier is absurdly low. Developer Blake Crosley added full Siri support to a hydration-tracking app using a single 80-line Swift file. If integration is trivial, Apple’s leverage comes entirely from distribution — exactly the same playbook it used with the original App Store.
Why it matters: This is the App Store story all over again, except the stakes are higher. Apple controls 1.4 billion iPhones. If it can make Siri the default gateway to every AI service, the 30% cut on AI subscriptions could dwarf App Store gaming revenue. But if developers smell a rent-extraction play and hold back, Siri’s ecosystem stays thin.
The On-Device AI Strategy
The Information reports that Apple plans to position on-device inference as its key differentiator — privacy-preserving and cost-saving compared to the data center buildouts from Anthropic and OpenAI. The company will use its custom Apple silicon to distill Google’s Gemini model into smaller versions that run locally.
Apple is also scouting acquisitions to advance model-shrinking work, with Liquid AI — a Massachusetts startup focused on running AI locally on devices — reportedly under consideration. The company also accidentally shipped Anthropic’s Claude configuration files in a support app earlier this year — a sign of how tangled Apple’s AI relationships have become.
There’s a tension: Apple’s original Apple Intelligence announcement said all cloud-bound queries would be handled exclusively by its own Private Cloud Compute infrastructure on Apple silicon. The Google deal changes that. Nvidia’s confidential compute technology within Google Cloud will handle some processing. Apple is expected to keep the Private Cloud Compute branding despite the shift.
Why it matters: Apple’s privacy brand is its most valuable asset. Any gap between “your data stays on Apple silicon” and “your data goes through Google Cloud, but encrypted” will get scrutinised hard.
What Siri 2.0 Actually Delivers
The visible changes in iOS 27:
- New activation: Glowing animation in the Dynamic Island, expanding into a translucent results card, swipable to full conversation view
- “Search or Ask”: System-wide panel triggered by swiping down from top center, replacing Spotlight as the primary search entry point
- Standalone Siri app: Persistent conversation history, file and image uploads, text-and-voice modes — layout described as “matching iMessage in structure”
- Personal context: Pulls from emails, messages, photos, notes, and calendar
- On-screen awareness: Siri can see what’s displayed and act on it
- Cross-app execution: Multi-step sequences through App Intents without opening individual apps
- Swappable AI backends: Users can route queries to ChatGPT, Claude, or Gemini through Extensions
- Camera integration: Visual Intelligence mode powered by Google Image Search
- Photos AI editing: Scene extension, object removal, natural-language edits
All three features Apple originally promised at WWDC 2024 — personal context, on-screen awareness, cross-app execution — are back on the menu. Two years late.
Why it matters: This is the feature list Apple should have shipped in 2024. If it works, Siri becomes a real assistant. If it’s another round of “coming later this year,” the $250 million lawsuit will have company.
The Investment Angle
The Technology: Apple is spending an estimated $1B/year on Google Gemini licensing, plus whatever it costs to run Nvidia B200 clusters in its PCC nodes. The company registered genai.apple.com a week before WWDC — infrastructure is ready.
The Opportunity: If Extensions works as promised, Apple becomes the toll collector for AI services on 1.4 billion iPhones. That’s a revenue stream that could exceed current App Store gaming income. Google wins too — it gets its model in front of the world’s most valuable user base while being paid for the privilege. Nvidia wins — B200 chips in Apple’s data centers. Developers who move early on Extensions get first-mover advantage in Siri integration.
The Risk: Apple has overpromised AI before and paid $250 million for it. The on-device strategy hits material limits — The Information reports Apple has struggled to run Google’s full Gemini model on its own PCC infrastructure. If the Ternus era opens with another Siri delay, Apple’s AI credibility is shot. And the Liquid AI acquisition is still “under consideration,” not done. As we’ve written before, big tech’s $700 billion spending spree may be building its own replacement — and Apple is now spending $1 billion a year on a model it doesn’t own.
For NZ investors: Apple’s NZ operations are modest, but the knock-on effects matter. Any company building AI apps for iOS needs to be planning for Extensions right now. The App Store commission question affects every NZ app developer selling AI subscriptions.
❓ Frequently Asked Questions
Q: What does this mean for NZ? NZ app developers need to prepare for the Extensions framework now — integration is technically simple (one developer did it in 80 lines of Swift) but the commercial terms are unclear. If Apple takes a 30% cut of AI subscription revenue, that changes the economics for every NZ-built AI app targeting iPhone users.
Q: What’s actually different this time? The Gemini model is 8× larger than Apple’s previous cloud model and uses mixture-of-experts architecture for lower latency. The Extensions framework is genuinely new — it turns Siri from a closed assistant into an AI platform. And the CEO transition means there’s political capital on the line if this fails.
Q: Should I upgrade to iOS 27 for the new Siri? Wait for reviews. Apple promised these exact features two years ago and didn’t deliver. The technology is more credible this time (backed by a $1B Google deal), but shipping on schedule matters more than announcing.
🔍 THE BOTTOM LINE
Tim Cook’s final keynote is Apple’s AI credibility test. The company is paying Google $1 billion a year to power a Siri that’s two years late, it settled a $250 million lawsuit for overpromising the last version, and it’s handing the CEO role to a hardware engineer at the exact moment software becomes the battleground. If Apple ships what it’s promising — on-device AI, Extensions framework, a Siri that actually works — it changes the AI competitive landscape by putting 1.4 billion iPhones in play as a platform. If it doesn’t, no amount of keynote polish fixes the credibility gap.
SOURCES
- Gadgets360 — WWDC 2026: Tim Cook’s Final Apple Keynote
- MacRumors — Apple to Reveal What Makes Its AI Different
- Oton Technology — Apple’s WWDC 2026 AI Overhaul Runs on a $1B Google Deal
- Bloomberg — Gurman Power On Newsletter
- AppleInsider — Tim Cook Expected to Head WWDC 2026 Keynote for the Last Time
- MacObserver — Tim Cook Prepares For His Final WWDC Keynote
