"Few people understand what it takes to build products that change the world the way John does."
That is the line Tim Cook chose for his last memo as Apple's CEO, and it lands like a thesis statement. The man Cook anointed, John Ternus, runs hardware engineering. The moment he was anointed is the one everyone says belongs to models, not machines.
Compress a single month and the strategy stops looking like a set of contradictions. Apple posts its best-ever June quarter — $109.4 billion, up 16% year on year. It hands the chief executive chair to the person in charge of building the physical objects. And it presses a lawsuit against the AI lab it pays to help run Siri.
Rent the best model available this year, own the device it runs on, and litigate the line between them. Three moves. One bet.
Three moves. One bet.
The market has spent two years assuming intelligence was the product. Apple ran a different experiment: rent the best model available this year, own the device it runs on, and litigate the line between them.
Start with what Apple actually got paid for. iPhone revenue grew 22% in the June quarter. Mac grew 29%. Both set June-quarter records.
Services reached $30.7 billion, up 12% — also a record, and still short of analyst estimates.
Read the split slowly, because it is the entire argument. The hardware lines outran expectations. The intelligence-and-subscriptions line missed. Apple is being paid for the device, not for the smarts inside it.
Now the accident. Developers and enterprises have been physically clearing high-end Macs off Apple's shelves to run large models locally, cutting cloud costs and keeping data on-premises.
The reason is architectural. Apple silicon uses unified memory, where the chip's memory is shared across every component, which makes local AI computing very fast and skips the cost of copying data between CPU and GPU.
The demand got so hot that configurations vanished. By May, base Mac minis that once shipped in days faced four-to-six-week delays; higher-end Studios with 96GB or more of unified memory pushed past ten weeks, per Ars Technica's tracking. Scalpers listed base units at nearly double the price.
Cook admitted the surprise on an earnings call: "Both of these are amazing platforms for AI and agentic tools, and the customer recognition of that is happening faster than what we had predicted."
Note the word "accidental" appears everywhere in the coverage. That framing is doing too much work.
Yes, the specific spike — driven by open-source frameworks like OpenClaw turning cheap Mac desktops into hosts for persistent agents — caught Apple flat-footed on supply. But the machine that caught the demand was built on purpose.
Apple pulled its Mac mini and Mac Studio refreshes forward from the usual autumn window into late August, reportedly with features tailored for AI workloads. One new feature lets users link multiple Mac Studios into a single system big enough to run frontier models. The M5 Ultra reportedly scales to 512GB of unified memory, per webpronews.com. Apple even redrew its chip roadmap, skipping high-end M6 Pro and Max variants to accelerate an AI-focused M7 generation for 2027.
That is not the behavior of a company stumbling. That is a company that already believed the device was the durable asset, now sprinting to feed a hunger it under-forecast.
Which brings the coronation into focus. Cook came from operations and built the most efficient supply chain in consumer electronics. Ternus comes from hardware engineering — the iPhone Air, the MacBook Neo, AirPods with hearing health.
Apple chose a silicon-and-devices leader at precisely the moment the industry narrative crowned software models the prize. The promotion is not indifference to the AI race. It is a wager about where the money in that race actually pools: in the surface, not the model.
The rental half of the strategy confirms it. In January 2026, Apple rebuilt Siri on Google Gemini in a multi-year deal reported at roughly $1 billion a year. Gemini became the core engine; ChatGPT was demoted to an optional plugin for world-knowledge queries.
Do the arithmetic. Apple pays Google about $1 billion a year for the intelligence layer and books $109.4 billion a quarter for the things that intelligence runs on. The model is a component with a renewable contract. The device is the balance sheet.
Then there is the defense, and it is the most revealing piece of all.
On July 10, Apple sued OpenAI and former employees in the Northern District of California — Apple Inc. v. Liu — over trade-secret misappropriation tied to consumer hardware development. The named defendant, Chang Liu, spent eight years at Apple as a senior systems electrical engineer before leaving for OpenAI in January 2026.
Look closely at what Apple is fighting over. Not model weights. Not training data. Hardware designs, manufacturing methods, supplier information.
Apple is seeking a preliminary injunction to block OpenAI from building hardware based on its technology. The company alleges Liu kept access after leaving by exploiting "a rare, previously unknown authentication bug," and that when he learned of the investigation in June, he enlisted OpenAI colleague Yu-Ting Peng to help destroy evidence.
The escalation came when Liu's counsel handed over his old Apple work laptop this month. Apple called it "shocking evidence." Prior filings quoted Liu's own text messages, punctuated with crying-laughing emojis, acknowledging he still had access to Apple files.
The strategic tell is what Apple did not do. It kept the ChatGPT integration agreement explicitly out of the suit. It rebuilt Siri on Gemini rather than OpenAI. Apple did not sue the company running its assistant. It sued the company building a device — because OpenAI is making consumer hardware, and the object in your pocket is the contested ground.
Rent the model. Own the device. Litigate the boundary. The record quarter and the lawsuit are the same posture — one expressed as demand, the other as defense.
So the thesis holds. But the fragility around it is real, and it arrives from three directions at once.
The first is institutional memory. As Ternus takes over, the veterans are streaming out. Phil Schiller reportedly stepped back from the App Store after a long tenure at the company that began in 1987 and included a four-year break. COO Jeff Williams retired in December after 27 years; general counsel Kate Adams and policy VP Lisa Jackson had their retirements announced; Apple Pay's Jennifer Bailey retires in October. Vision Pro chief Paul Meade left in June — for OpenAI, a company where, per Apple's lawsuit filing, more than 400 former Apple employees now work.
A hardware-first bet depends on deep hardware knowledge. Apple is losing custodians of it at the exact moment it needs to repopulate the executive bench.
The second is supply. The same DRAM and NAND shortage feeding data-center AI builds is hitting Apple's own devices. Apple paid significantly higher RAM prices and reportedly capped some new Studio configurations at 96GB, down from the 512GB supported in prior generations, per webpronews.com — throttling the very high-memory configs driving the local-inference boom. It raised Mac prices across the board in June and leaned on a Klarna-backed leasing program to soften the blow.
The third is the demand Apple is leaving on the table. It has no engineering team dedicated to business customers and no developer-relations staff focused on AI. So when Mac stock ran dry, enterprises that couldn't wait bought Nvidia's DGX Spark systems instead.
That is the sharpest irony in the whole story. Apple's contrarian bet — that the device, not the model, is the durable asset — is being validated by buyers who show up wanting exactly that device.
And Apple, caught flat-footed, has no one at the door to greet them.