Apple spent twenty years building a religion of secrecy on Chinese soil. The bill for leaving comes due in a folder named com.apple.factorydata.

In the spring of 1983, at a beachside retreat, Steve Jobs warned his Macintosh team to guard their documents. An Apple salesman in Chicago had just been offered a complete Lisa sales plan — handed over, allegedly, by someone at IBM. Jobs vowed to come down "real hard" on any vendor who leaked. Then he walked to the board, sketched an inverted pyramid, and told the room their work would send a "giant ripple through the universe."

That was the founding moment of a corporate religion. Secrecy at Apple was never a policy; it was infrastructure — body searches at factory gates, color-coded badges, prototypes chained to desks, suppliers who could be ruined for a single photograph. Over the next four decades Apple built the most disciplined secrecy apparatus in industry, and it built almost all of it in one place: China.

This morning, a folder named com.apple.factorydata turned up on the open internet.

The News

A group calling itself World Leaks posted more than 630 gigabytes of data — over 204,300 files — that it claims to have stolen from Tata Electronics, the Indian conglomerate that has become Apple's anchor manufacturer outside China. Inside the dump, according to 9to5Mac: a 52-page document bearing Apple's markings detailing iPhone circuit-board quality-inspection standards, dozens of files referencing Tata's Hosur assembly plant, and those com.apple.factorydata folders.

Tata confirmed the incident in careful language: "A few weeks ago, Tata Electronics identified a cybersecurity incident on some of our systems… the incident has had no impact on our operations." A person familiar with the matter said a ransom had been demanded. Apple did not comment.

The Tax Nobody Budgeted

The reflex reading is "another ransomware story." It isn't. It's an invoice — the first visible bill for a decision Apple made under enormous pressure, and the line item it forgot to price.

For five years Apple has been hauling its supply chain out of China. The progress is real: roughly a quarter of all iPhones are now made in India — about 55 million units last year, up more than half from the year before. Tim Cook now tells investors that the majority of iPhones sold in the U.S. have a country of origin of India. Tata, which bought Wistron's Karnataka plant for $125 million in 2023 and later took control of Pegatron's India operation, is the company Apple chose to carry that bet.

But you can fly machines to Hosur. You cannot fly the religion. The secrecy apparatus that kept twenty years of iPhone designs off the open internet was hard-won, soil-specific, and built supplier by supplier over decades of ruthless conditioning. As Patrick McGee documents in Apple in China, India's plants are still mostly doing final assembly, test, and pack — components shipped in from China, real engineering maturity a five-to-ten-year climb away. The tooling moved first. The security culture is the part that doesn't fit on the plane.

What History Says

Apple already knows its supply chain leaks. It just used to leak in ways it could survive.

In November 2018, when Apple slashed iPhone XR production by 11 million units, the cut was on Nikkei's wire within days — Foxconn and Pegatron had halted their ramp, and Apple's own operations executive confirmed the report was accurate. Cook wanted financial penalties. The lesson McGee draws is the uncomfortable one: the larger and more distributed the supply chain, the more surfaces it has to leak from, and no badge-and-bodysearch regime fully closes them.

That was one country. Diversification doubles the perimeter. Every additional geography is another secrecy apparatus to build from zero, another partner whose security maturity Apple inherits whether it has audited it or not. The move that reduces Apple's exposure to Beijing's leverage — the equipment-export freezes and visa games China has used to slow the India shift — increases its exposure to everything else.

The Deeper Pattern

There's a darker echo underneath. McGee's central argument is that Apple, by teaching China to build at Apple's standard, manufactured its own most formidable rivals. The blueprints went into the world and came back as competitors. A 52-page quality-inspection spec sitting in a ransom dump is the same story compressed: Apple's hardest-earned manufacturing knowledge, loose again — this time not transferred by training but simply taken.

Jobs understood, in 1983, that the ripple goes outward. What he didn't say at the beach is that it also comes back.

What Happens Next

India's climb will continue; the economics demand it and so does Washington. But the Tata breach reprices the bet. Security maturity — the audits, the network segmentation, the supplier conditioning that took twenty years in Shenzhen — is now the unbudgeted line on the diversification spreadsheet, and it cannot be rushed the way a factory can.

Apple built a kingdom on the principle that nothing leaves the room. It is now operating in several rooms at once, in a country still learning the rules. The ripple, it turns out, was always going to run in both directions.

Also This Week

Design loses its seat. Bloomberg's Mark Gurman reports that Apple's industrial-design team "no longer has a true seat at the executive table" — reduced, in one description, to "a service bureau." For the company Jony Ive once ran as a design sovereignty, it's a quiet inversion: the org that defined the object now waits on the org that defines the intelligence.

Apple hands Samsung the crown jewel. Samsung Display has been approved to begin module production of the foldable iPhone's OLED panel as the sole qualified supplier — re-creating, on Apple's highest-margin new product, exactly the single-vendor dependency that the company has spent a decade trying to escape.

The Mac's playbook, copied. Microsoft's new Surface Pro and Surface Laptop went on sale June 16 with Qualcomm's Snapdragon X2 chips and an 80-TOPS NPU, starting at $1,499 — the closest the Windows world has come to the Arm-efficiency, all-day-battery proposition that made Apple Silicon feel uncatchable in 2020.

From the Library

This newsletter draws on a shelf of books about Apple's history. Today's issue featured stories from:

Apple in China by Patrick McGee — the 2018 supply-chain leak, India's "ass-backwards" assembly climb, and the argument that Apple's manufacturing knowledge always finds its way back into the world.

Return to the Little Kingdom by Michael Moritz — the 1983 beach retreat where Jobs sketched the inverted pyramid and made secrecy a founding article of Apple's faith.

In 2006, a glass executive named Wendell Weeks took a phone call from Steve Jobs, who wanted the hardest glass in the world and wanted it in six months. Weeks told him Corning had invented something like that back in the 1960s — a nearly unbreakable glass meant for fighter-jet cockpits — and then shelved it. Nobody had wanted it. There was no market. As Walter Isaacson tells the scene in Steve Jobs, Weeks had to cut Jobs off mid-lecture: "Can you shut up and let me teach you some science?" Then Apple committed to buy the stuff by the millions. Corning converted a Kentucky plant, ran the line around the clock, and the iPhone shipped with a glass face. Within a few years, every premium phone on earth wore a slab of the same glass.

Apple hadn't invented the glass. It had been willing to buy a thing nobody else could justify — and that willingness became the industry standard. It's a move Apple makes again and again: take a component too expensive to be ordinary, put it in everything, eat the cost on the way up, and let scale make it normal. Rivals follow years later, once the price has fallen. The willingness to pay early is the weapon.

Last week, at its developer conference, Apple did it again — with the most expensive component on the planet right now.

The spec nobody mentioned

The keynote was about Siri and Apple Intelligence. The load-bearing detail was buried in a developer session. Apple's new on-device system — the one it now calls Core AI, the engine behind the smarter Siri and the better dictation — needs twelve gigabytes of memory to run on the phone. The iPhone 18 is reported to ship with exactly that, up from eight. Apple's entire AI pitch is that the intelligence runs on your device, privately, instead of in someone else's cloud. That pitch has a bill of materials. On-device thinking runs on memory, and Apple just made a lot of it the floor.

It could not have chosen a more expensive moment. The price of DRAM has roughly doubled this year, with some grades headed toward three or four times their old price by year-end. The cause is the AI boom: data-center builders are buying memory by the warehouse, and the chipmakers are selling to whoever pays most. Memory is the substance an entire industry suddenly needs more of than exists.

Why this is a moat, not a mistake

Here's what the wallet-focused coverage missed. The crunch doesn't hurt Apple most. It hurts Apple least.

When the memory makers came for their price increase this winter, Apple agreed to roughly double — and the open market climbed further than that. Apple had stockpiled inventory months ahead and carries the fattest margins in the industry to absorb the rest. Counterpoint Research put it flatly: Apple is "best-positioned to weather the next few quarters," while it "will be tough for others." Apple's smaller rivals — Honor, Oppo, the Android middle — are passing 15 to 25 percent straight to customers, because they have no cushion to hide it in.

Now lay the AI bet on top. To match Apple's on-device intelligence, a competitor has to put twelve gigabytes of the same scarce memory into its phones — and either swallow a cost it can't afford or raise prices in a market already squeezing it. Most won't. They'll push the AI back to the cloud, or ship less of it, or wait. Apple will put the memory in, absorb the premium, and call it the standard. The shortage isn't a threat to Apple's AI plan. It's the thing that prices everyone else out of copying it.

And there's a second edge. Because Apple designs the silicon, the operating system, and the model as one system, the Neural Engine and the rest of the phone feed from a single shared pool of memory — so the same twelve gigabytes does more thinking on an iPhone than on a handset assembled from three companies' parts. Apple wrings more from each gigabyte and can afford more of them than anyone. The moat, doubled.

What Cook actually said

Which is why Cook's words this week are worth reading carefully. In an interview with The Wall Street Journal, he called price increases "unavoidable" and the situation "unsustainable," and said he'd never seen anything like it in over 40 years. That sounds like distress. But asked why Apple doesn't simply build its own memory factories, he gave the answer that has guided him for twenty-five years: "We can't do everything. We know what we're good at."

He isn't panicking. He's narrating the sorting of a market. Apple will raise some prices — it can, because its customers pay — and absorb what it must, and come out the far side with its AI story intact and its rivals thinner. The "unavoidable" increase is the cover charge for a room only Apple can still afford to sit in.

The pattern

Gorilla Glass was this exact shape. So was the Retina display, and the machined aluminum unibody — every time, Apple made an exotic part ordinary by being the buyer who showed up first with the most money, then let volume drag the price down for everyone who came after. The difference now is that the part isn't glass or metal. It's memory, and there isn't enough of it. Apple is about to make twelve gigabytes the price of a modern phone at the precise moment that is hardest to do — which is exactly why, as Patrick McGee documents Apple's supply playbook in Apple in China, it's the kind of move only Apple can make.

What happens next

Watch whether the iPhone 18 holds its price or quietly climbs, and by how much — that number is the width of Apple's cushion. Watch whether the Android middle ships real on-device AI this year or punts it to the cloud and calls it a feature. Memory will get more expensive before it gets cheaper; the relief isn't expected until 2027.

For fifty years, memory was the thing Apple bought better than anyone. Now it's the thing Apple is betting its rivals can't afford to buy at all.

Also This Week

Find My learns to keep a secret: iOS 27 quietly added a way to hide your location from a specific person for twelve hours — with no notification that you did it. The company that invented anti-stalking alerts now helps you vanish from the people you know.

Apple ships an off switch for its own design: After a year of complaints that Liquid Glass was hard to read, iOS 27 buried a slider that lets you dial back the transparency. A flagship design that arrives with a way to turn it down.

A second iPhone Air: Mark Gurman reports a dual-camera follow-up to the thin iPhone Air for spring 2027, fixing the first model's biggest complaint.

From the Library

This newsletter draws on a shelf of books about Apple's history. Today's issue featured stories from:

Steve Jobs by Walter Isaacson — the 2006 cold call to Corning and the six-month sprint to put glass on the iPhone.

Apple in China by Patrick McGee — Corning reviving its shelved 1960s glass, and the supply discipline that lets Apple absorb what rivals can't.

Apple by David Pogue — Jobs killing plastic for glass in weeks, and the iPhone 4 redesign that turned a premium part into the standard.

In the flat country east of Cambridge sits a vacant turkey barn. In the spring of 1991, thirteen men carry bargain desks in between the old farm beams and start a company that will end up inside almost every phone on earth. They've come from Acorn Computers, a struggling British PC maker, and they work for a joint venture formed a few months earlier that nobody expects much from: Acorn contributes the engineers, VLSI Technology the design tools, and Apple £1.5 million in cash — about $2.5 million — for a 43 percent stake, matched share for share by Acorn.

Apple is in the barn because of a Hobbit. AT&T's Hobbit processor, the chip inside early Newton prototypes, was buggy — and it belonged to a giant with plans of its own. When VLSI tried to grab a larger stake as the deal closed, Larry Tesler, negotiating for Apple, told its man to back down or Apple would walk — then slammed down the phone, and called back to confirm the Newton would run on Arm. He wanted something nobody else in the industry was selling: a chip company with no products, no factory, and no agenda beyond the instruction set itself. A landlord, not a rival. Arm's first job was rescuing the Newton; its ARM610 shipped inside the 1993 MessagePad. AT&T took the loss hard — an Arm engineer later spotted an Apple poster on an AT&T wall, repurposed as a dartboard.

The Newton died. The arrangement lived. For thirty-five years Arm designed the architecture and leased it to everyone — Apple, Samsung, Amazon, Qualcomm, Alphabet, Huawei, Meta, Tesla, bitter rivals all buying from the same quiet supplier — while building nothing itself. The Switzerland of silicon, the industry called it.

In March, Switzerland started shipping weapons.

The Landlord Builds a House

On March 24, Arm announced the AGI CPU — the first chip of its own in company history: 136 cores on TSMC's 3-nanometer process, aimed squarely at AI data centers. On June 2, Supermicro unveiled five server lines built around it, claiming over twice the performance per rack (Arm's estimate, not an independent one). At Computex, CEO Rene Haas said Arm's $15 billion chip-revenue target may land ahead of schedule, with reported commitments from Meta, Oracle, and ByteDance. Wall Street did the repricing: Bank of America to $335 on Thursday, Mizuho to $500, the stock closing Friday at $368 — a share Apple helped price at $51 as a cornerstone investor in the 2023 IPO.

The number isn't the news. The business is. A company that spent thirty-five years as everyone's neutral supplier now competes with its own licensees — against Qualcomm and Ampere for data-center sockets, reportedly poaching their engineers, and bidding for the same TSMC wafers Apple once locked up years in advance.

The Stake That Paid the Rent

In 1996, CFO Fred Anderson walked into an Apple he compared to an emergency-room patient. As David Pogue recounts in Apple, he flew to Tokyo to beg for time on $400 million in loans, sold factories, squeezed suppliers — and found, nearly forgotten on the books, Apple's stake in ARM. When ARM went public on April 17, 1998 — dual-listed in London and on Nasdaq, up 46 percent on day one, a billion-dollar valuation on $5.4 million of profit — Apple started selling. The 10-Ks record the drip: $245 million in fiscal 1999, $372 million in 2000, $176 million in 2001, down to a final 278,000 shares sold in 2003 for $295,000. Call it $840 million all in; James Ashton's history of Arm lands on $838 million following the same trail. Legend rounds it up to $1.1 billion; neither the filings nor the definitive account support that. Jobs later said Apple had been ninety days from bankruptcy; Sculley said the Arm money kept the doors open.

Apple sold every share. It kept the lease. And the most important part of that lease was bought in a moment of dread. In April 2008, at a hotel offsite near St Albans, Arm's executives ate a grim dinner with football on the television — one of them called the mood "like a blooming funeral" — convinced Apple's silence meant Intel had won the iPhone and Arm was about to be dropped. Then the call came. Apple didn't want out. It wanted an architecture license: the broadest kind, the right to design its own Arm cores. That license, later extended "beyond 2040" in Arm's own words, sits under every A-series and M-series chip Apple has shipped. Apple sold the barn and kept the right to live in it forever.

The Playbook Comes Home

Apple wrote this script. Learn the stack, then own it: PA Semi in 2008 became the A4, the A4 became Apple Silicon, and three chip suppliers — Motorola, IBM, Intel — were fired the moment their roadmaps served someone else's empire. Now Apple's last remaining chip dependency is running the same play in reverse, integrating from architecture down into product.

And Arm has shown what it does to a licensee standing in the way — in a fight that ran straight through Apple's own alumni. Nuvia, the chip startup Qualcomm bought in 2021, was founded by the architects who had built Apple's A-series silicon. Arm sued in 2022, demanding the designs be destroyed, and in October 2024 moved to cancel Qualcomm's license outright. A Delaware jury sided with Qualcomm; the judge finished the job last September. Arm lost the case. What survived wasn't legal precedent — it was behavioral: the steward of the architecture will go to war over the lease, even against the people who once built the most celebrated chips in its own ecosystem.

Apple's contract is safe past 2040. What's not contractual: first call on TSMC capacity, in an order book Arm itself now bids into alongside Nvidia and OpenAI. The engineers. And the oldest assumption of all — that the referee has no team.

What Happens Next

Watch three things. Whether Arm's silicon stays in the data center or creeps toward client chips, where Qualcomm lives today and the M-series' territory begins. Whether TSMC's allocation starts favoring the landlord's own designs. And whether Apple's tone toward Arm changes — cornerstone investors don't usually stay warm toward companies that start competing with them.

In 1990, Apple paid $3 million for a chipmaker designed to want nothing. The bet paid twice: once when the architecture took over the world, once when selling it kept Apple alive. The third act opened in March. The company in the turkey barn finally wants an empire of its own — and this time, Apple is the tenant.

Also This Week

The Intel Mac era ends: macOS 27 "Golden Gate" supports zero Intel Macs, and the ecosystem followed within the week — CrossOver is going Apple Silicon-only. Twenty-one years, almost to the week, after Jobs announced the Intel switch at WWDC 2005. AppleInsider

Siri AI launches without the EU and China: Apple publicly blamed the DMA for the European delay in an unusually pointed statement while Chinese approval stalls; the Commission is visibly irritated.

From the Library

This newsletter draws on 38 books about Apple's history. Today's issue featured stories from:

The Everything Blueprint: The Microchip Design That Changed the World by James Ashton — the definitive history of Arm: Harvey's Barn, Tesler's ultimatum, the 2008 "funeral dinner" that became Apple's architecture license, the $838 million stake sale, and the Nuvia fight over Apple-trained talent

Apple by David Pogue — Fred Anderson's emergency-room rescue and the forgotten ARM stake that paid for Apple's survival

Apple: The Inside Story of Intrigue, Egomania, and Business Blunders by Jim Carlton — the Newton's no-compromise origins

January 9, 2007. Moscone West. Steve Jobs has just pulled the iPhone from his pocket, and now he does something no Apple keynote would do today: he invites another company’s CEO on stage. Eric Schmidt of Google bounds up the steps. He has sat on Apple’s board for four months, and he opens with a joke about how close the two companies have grown. “If we just sort of merge the companies, we could call them AppleGoo.” Then the line underneath the joke: “you can actually merge without merging.” The iPhone ships with Google search in its browser, Google Maps on its home screen, YouTube one tap away.

The friendship lasted thirty-one months. In August 2009, with Android in the market, Jobs pushed Schmidt off the board. When an HTC phone shipped with multi-touch the following spring, Jobs vowed to spend “my last dying breath if I need to, and every penny of Apple’s $40 billion in the bank” to destroy Android — “I’m willing to go thermonuclear war on this.” Schmidt asked him to coffee at a Palo Alto café to talk settlement. Jobs cut him off before he could open the menu: “If you offer me $5 billion, I won’t want it. I want you to stop using our ideas” (Walter Isaacson, Steve Jobs).

Yesterday, at WWDC, Craig Federighi stood where Jobs once stood and announced that Apple has “partnered deeply with Google.”

The Merger Ships

The product is Siri AI — the conversational assistant Apple promised in 2024 and finally delivered, built on Apple Foundation Models “co-created” with Google. Apple is emphatic that the models contain “none of Gemini”: they were trained on Apple’s data and refined using outputs from Gemini frontier models — Google as teacher, not tenant. But the hardest questions route to a model that runs on Nvidia GPUs in Google’s cloud. Apple tried running it on its own Private Cloud Compute hardware first, The Information reported. It was too slow. The brain Apple rents lives in a building Apple doesn’t own, on chips Apple doesn’t control, encrypted by Nvidia rather than guarded by Apple silicon. Bloomberg puts the price at roughly $1 billion a year.

One detail says everything about how Apple feels about this: its press releases never contain the word Google. The partnership exists on stage and in press briefings — nowhere in writing Apple controls.

And the money is flowing the wrong way. For twenty years, every dollar in this relationship moved from Mountain View to Cupertino: Google paid Apple roughly $20 billion in 2022 alone to stay the default in Safari. The Gemini deal is the first time Apple has ever paid Google. The junior partner and the senior partner have quietly switched seats.

The Maps Lesson

Apple knows what leaving Google in one leap costs, because it tried. In September 2012 it ripped Google Maps out of iOS and shipped its own — and the world got a melted Brooklyn Bridge, a Jacksonville hospital rendered as an abandoned supermarket, roads that dove into canyons. Tim Cook published an apology recommending Bing, MapQuest and Waze by name. Scott Forstall refused to sign that letter and was gone within a month (David Pogue, Apple). This April, Cook called the Maps launch his “first really big mistake” as CEO.

The Gemini deal is the Maps fiasco run in reverse: rent openly while you build quietly, and don’t jump until the replacement is ready. Nobody gets fired for shipping the working version of someone else’s technology. They get fired for shipping the broken version of your own.

The Search Lesson

Search teaches the opposite lesson: some rentals never end. The default deal survived Schmidt’s exit, thermonuclear war, a patent crusade that burned through $60 million in fees, the Maps divorce — even a federal monopolization verdict. Judge Mehta let the payments continue; they just have to be non-exclusive now, renewed one year at a time. The most durable thing Apple and Google ever built together is the check.

But notice where the alliance now lives. The search payments renew annually under court supervision. The Gemini contract is already discovery material in Elon Musk’s lawsuit against Apple and OpenAI. A relationship that began with a joke on a keynote stage is now administered by federal judges.

The Doctrine Bends

January 21, 2009. One week after Jobs went on medical leave, an analyst asked Tim Cook, then the acting CEO, what would happen if Jobs never came back. Cook didn’t dodge. He recited — “as if reciting a creed he had learned as a child in Sunday school,” Adam Lashinsky wrote in Inside Apple — the sentence that became the Cook Doctrine: “We believe that we need to own and control the primary technologies behind the products we make.”

Seventeen years later, the doctrine survives by definition rather than amendment. Apple owns the model weights; it rented the teacher and rents the data center. And the doctrine always had this flexibility — its author is the man who closed Apple’s factories in 1998 and outsourced assembly forever. “Own and control” never meant owning everything. It meant Apple decides what counts as primary. For nineteen years, silicon was primary and search was not, and the line held. Intelligence is the first technology that refuses to stay on one side of it.

How Rentals End

Apple’s ledger of dependencies reads like a rule: Motorola’s 68K, ten years, replaced. PowerPC, twelve years, replaced. Intel, fourteen years, replaced. Google Maps, five years, replaced. When the rented thing defines the product, Apple eventually builds its own. When the rented thing merely pays — search — the lease rolls over forever.

Gemini is the first rental that breaks the rule’s clean edges: it defines the product, and Apple writes the check. By the doctrine’s own logic, that shouldn’t be a stable state. Watch three things: whether the cloud model migrates off Google’s Nvidia fleet onto Apple silicon servers; whether Apple’s distilled models stop needing a teacher; and what the Musk discovery unseals about how long this contract actually runs.

Schmidt’s joke had two parts, and everyone remembers the wrong one. AppleGoo was the laugh line. “You can actually merge without merging” was the plan — and nineteen years later, it shipped. But companies that merge without merging can also divorce without divorcing. Apple and Google have done it once already.

Also This Week

Apple’s first AI paywall. iCloud+ subscribers now get higher daily limits on Apple Intelligence features that use server models — the Services playbook, applied to compute.

Apple and Brussels blame each other. Siri AI is delayed indefinitely in the EU; Apple blames DMA interoperability demands, and the Commission’s spokesperson answers that the decision is “Apple’s and Apple’s only.”

Liquid Glass gets a dimmer switch. iOS 27 and macOS 27 add a system-wide transparency slider — and default to less transparent. The second consecutive walk-back of last year’s redesign.

From the Library

This newsletter draws on 29 books about Apple’s history. Today’s issue featured stories from:

Steve Jobs by Walter Isaacson — thermonuclear war, the $5 billion coffee, and Schmidt’s exit from the board
Inside Apple by Adam Lashinsky — the Cook Doctrine’s birth scene, a creed recited from memory
Apple by David Pogue — the Maps implosion that taught Apple how not to leave Google

On April 12, 2007, Apple did something it almost never does. It admitted it was late. Mac OS X Leopard, promised for that spring, would slip to October. The reason was buried in a terse statement: the company had pulled its key software and QA engineers off the Mac and onto a phone it hadn't shipped yet. The Mac would wait. The iPhone could not.

Leopard arrived that fall, ambitious and a little raw. So two years later Apple did something even rarer. It shipped an apology. Mac OS X Snow Leopard, August 2009, sold on a single boast: zero new features. No new icons, no new apps. Just the same operating system rebuilt underneath — faster, smaller, 64-bit. Apple spent a year making the Mac better at being the Mac. It was the most confident thing the company could do, because in 2009 the operating system was the product. The magic lived on the machine in front of you.

Apple is about to do it again. On June 8, at what may be one of Tim Cook's last keynotes as chief executive, the company is expected to unveil iOS and macOS 27 — and the early reporting frames them as Snow Leopard reborn: a stability release, legacy code stripped for speed, the excesses of last year's Liquid Glass redesign sanded down. Tahoe earned that cleanup. John Gruber, who has covered Apple longer than most of its executives have worked there, refused to even install it, calling Liquid Glass “goofy,” Fisher-Price, “half-baked who-thought-this-was-OK-to-ship” work. Apple is reportedly retooling the worst of it for 27.

But the headline of the keynote won't be the cleanup. It'll be Siri — rebuilt, conversational, finally the assistant Apple promised in 2024 and didn't ship. And here is the thing worth noticing. The new Siri isn't really part of the operating system at all. It's a cloud service the OS calls out to. The intelligence runs in a data center, on a model Apple licenses from Google for a reported billion dollars a year. The phone asks the question. Someone else's computer answers it.

That is the quiet inversion inside an unassuming release. In 2009, Apple could afford to spend a year polishing the OS because the OS was where it competed. In 2026, it can afford to make the OS boring for the opposite reason: the part that sells the phone has left the phone. It moved to the cloud — and the cloud is rented.

For Apple, this is close to heresy. This is a company whose scars are all cloud scars. When it first tried to live in the cloud — MobileMe, 2008 — it launched so broken that Steve Jobs hauled the team into the Cupertino auditorium and asked them to explain what the product was supposed to do. They told him. “So why the f*** doesn't it do that?” (Walter Isaacson, Steve Jobs). The first Siri, in 2011, shipped as a “beta” — a word Apple never used — because the thing it had bought from a DARPA-funded lab was clumsy with accents and helpless without a signal (Adam Lashinsky, Inside Apple). And when Apple tried to wrench its maps away from Google in one leap, it gave the world melted bridges and a hospital stranded in a park, and it cost Scott Forstall his job. The lesson Apple drew from twenty years of this was plain: the things that matter belong on the device, under your control, where the network can't fail you.

So is renting Google's brain the new permanent shape of Apple — the device a beautiful dumb terminal, the intelligence forever someone else's? Maybe. Or maybe it's the oldest move in the playbook. Apple rented Google's maps before it built its own. It rented Intel's chips for fifteen years before Apple Silicon made them obsolete. It rented the modem before the C1. Each dependency looked like surrender until the day Apple brought it home. Cloud Gemini may be a placeholder — a credible brain bolted on to stop the bleeding while Apple's own models grow up — and the day they are good enough, the intelligence comes back down onto the Neural Engine, where the religion says it belongs.

We won't know on June 8. But there is a way to read the tea leaves. Watch how Apple spends its keynote minutes — how many on the operating system, the old crown jewel, and how many on the Siri it had to rent. The ratio will tell you where Apple now thinks its product lives.

Snow Leopard worked because Apple spent that year fixing what was its own. This time the cleanup is real and the discipline is admirable. But the part everyone will remember, the part that answers when you ask — that is snow Apple didn't make. We will find out soon enough whether it learns to make its own again, or whether the most valuable thing your iPhone does is now, and always will be, someone else's snow.

Also This Week

Apple guts the Vision Pro roadmap. Ming-Chi Kuo and Mark Gurman both report the mixed-reality roadmap scrapped for smart glasses, Vision Pro 2 “on ice” — a pivot personally signed off by John Ternus, the incoming CEO.

India's $38 billion question. Apple agreed to hand its India revenue data to antitrust regulators rather than risk a fine that could reach $38 billion — the China-concession playbook, replayed in Delhi.

From the Library

Today's issue leaned on three books about Apple's history.

Steve Jobs by Walter Isaacson — the MobileMe auditorium reckoning and the cloud pivot that followed.

Inside Apple by Adam Lashinsky — Siri's quiet debut as an admitted “beta.”

Apple by David Pogue — the Maps face-plant that took down Scott Forstall.

At five in the morning on October 3, 1992, a prototype Macintosh booted in a Cupertino lab. Nothing about it looked remarkable. The Finder loaded. A cursor blinked. But the chip underneath was a PowerPC, a processor no shipping Mac had ever used. Apple had just swapped the engine of its entire computer line while the car was still moving — and made it look like nothing happened.

It would do this twice more. In 2006 it tore out PowerPC and bolted in Intel; Bill Gates, who'd once sneered at Apple's "crazy-colored cases," called the seamless switch something that "sounds impossible." In 2020 it threw out Intel for its own silicon, and the new M1 Macs ran old Intel software — emulated — faster than the Intel Macs had run it natively. Twenty-hour battery life. Reviewers sounded stunned.

Three times Apple changed the most fundamental part of a computer. Three times the industry said it couldn't be done without breaking everything. Three times the secret was the same: Apple owned the chip, the operating system, and the apps, and it built a translator good enough that the old software just — worked.

Last week, Nvidia copied the destination. It has no idea how hard the road is.

The blueprint, unbundled

On May 31, on a stage in Taipei, Jensen Huang unveiled RTX Spark — a laptop chip fusing an Arm CPU and an Nvidia GPU with up to 128 gigabytes of unified memory and a full petaflop of AI horsepower in a 14-millimeter chassis. Thirty-plus laptops this fall — Dell, HP, Lenovo, Asus, MSI, and Microsoft's own Surface Laptop Ultra. Satya Nadella promised "unmetered intelligence to every home and every desk."

Read the spec sheet again. Arm CPU. Integrated GPU. Unified memory. Thin body. All-day battery. That is the MacBook. That is the exact architecture Apple shipped in 2020 and has sold as a generational lead ever since. For six years, "Arm plus unified memory beats Wintel" was Apple's private thesis, proven by Apple alone. Now it's a merchant part. Anyone can buy it.

And the company selling it is the one Apple fired. In 2008, Nvidia's graphics chips began dying inside MacBook Pros — solder cracking under heat, an episode the industry called Bumpgate. The relationship never recovered. Apple shipped its last Nvidia-powered Mac in 2015; modern macOS still refuses to run an Nvidia GPU. The spurned supplier went on to become the most valuable company on earth. Now it has handed every Apple competitor Apple's own blueprint.

The hard part

Here is what the spec sheet doesn't tell you. A laptop chip is only as good as the software that runs on it — and Windows on Arm is a fourteen-year graveyard. Windows RT in 2012. Surface Pro X in 2019. Qualcomm's Snapdragon in 2024. Each promised the world and shipped a machine that couldn't run half your programs. It's better now; Microsoft says users spend ninety percent of their time in native apps. But Nvidia's chip leans on an emulator, Prism, that Microsoft tuned for Qualcomm's silicon — not Nvidia's. Huang promised RTX Spark will run "every Windows app ever made." That is the single hardest promise in computing. Apple is the only company that has ever kept it — and it kept it by controlling the emulator, the chip, and the operating system all at once. Microsoft controls one of the three.

Consider what "all at once" bought Apple. When it designed the M1, it buried a hidden switch in the silicon — a mode that makes the chip order memory the way an Intel chip does, the one place Arm and x86 fundamentally disagree. Apple flips it on only while the translator is running. It cost transistors. It bought nothing but faster emulation of software written for the chip Apple was throwing away. That is why M1 Macs ran old Intel apps faster than Intel Macs did. Only a company that builds the processor and the operating system and the translator would ever spend silicon that way. Nvidia builds the chip. Microsoft builds everything else. They cannot do this for each other.

Nvidia's other moat is CUDA — the software language the entire AI world is written in, and the thing Apple has no answer to. That moat looks bottomless. But this spring, Anthropic's Claude Code took a complete CUDA program and rewrote it for a rival chip in thirty minutes, no human touching the code. If AI agents can translate between chips for free, the language that locks developers to Nvidia stops locking. And the company best placed to walk through that opening is Apple, whose own framework already runs local models two to three times faster than the standard tools — on Apple's own silicon.

What history tells us

Apple has made this bet before. Own the whole stack, control the architecture, and wait for the rest of the industry to commoditize the parts you don't. It made the bet against Motorola, against Intel, against the entire PC business. It has also lost it: PowerPC was supposed to bury Intel, and instead Apple spent a decade shackled to chips that "sucked," in Steve Jobs's words, before he tore up the alliance.

So the question — a week before Apple takes the WWDC stage and is expected to admit that Siri's brain will be rented from Google — is not whether Apple lost the laptop. It didn't. Nvidia copied the body. The seamless soul of the thing, the part that took Apple thirty years and three transitions to learn, doesn't come on a spec sheet.

Gates had the word for it in 2006. Sounds impossible.

Everyone's about to find out how impossible.

Also This Week

Apple patches its newest silicon, twice in one day. Apple shipped iOS 26.5.1 and macOS Tahoe 26.5.1 simultaneously to fix unexpected M5 Mac shutdowns and a wired-charging failure on the iPhone Air and iPhone 17 — quality wobbles on the year's flagship chips, one week before WWDC. (MacRumors)

Apple Glasses slip to late 2027. Mark Gurman reports Apple's first smart glasses — a $200–500 Meta Ray-Ban rival with cameras but no in-lens display — have slipped from early to late 2027, with a Vision Air to follow by 2029. It's framed as Tim Cook's top priority before he hands the CEO chair to John Ternus in September.

A win for Apple in Musk's AI suit. A Singapore court denied xAI's requests for documents in Elon Musk's antitrust case against the Apple–OpenAI partnership — a procedural setback for the only live challenge to Apple's AI alliances.

From the Library

This newsletter draws on a shelf of books about Apple's history. Today's issue leaned on three:

Insanely Great by Steven Levy — the PowerPC leap: the 90/10 insight that made emulation thinkable, and the prototype Mac that booted at 5 a.m.

Steve Jobs by Walter Isaacson — Jobs hanging up on Motorola ("your chips sucked"), and Gates calling the seamless Intel switch "impossible."

Apple by David Pogue — the M1 transition that turned a dreaded migration into a rout.

Simson Garfinkel pulls on fireproof pants and a respirator and points a steel-cutting torch at the matte-black box. Nothing happens. He tries a gas torch. Nothing. A magazine called NeXTWORLD has rented a brick-and-steel burn cell at Lawrence Livermore to give the NeXT Cube a funeral pyre, and the funeral will not start. The Cube is built from a magnesium alloy Steve Jobs picked for its strength and its sheen. "This is so NeXT," Garfinkel tells the reporter beside him. Eventually the pyrotechnician runs a three-inch gas pipe into the cell, lights it, and tosses in magnesium shavings. The black box pools, then flashes white. Witnesses leave dusted with powder, like fallout.

The year is 1993. NeXT has just killed its own hardware. The market killed the Cube long before the torch ever could.

This is a Geoffrey Cain story. More precisely, a story Cain and his two co-authors—Pixar's Ed Catmull and NeXT's Dan'l Lewin—pulled out of an oral history project and stitched into Steve Jobs in Exile, a chronicle of the seven years between Jobs's two Apple tenures. It does not appear in any of the other 37 books in our library. We checked.

We've extracted 7,088 stories from those 37 books. Steve Jobs in Exile contributed 255. Of those, 69 are told by no other book in the library—not by Isaacson, not by Schlender, not by Lashinsky.

An honest caveat: most of these stories are not really about Apple. They're about NeXT, mostly, with Pixar at the edges and a few late-1996 scenes leaking into the Apple return. The Apple Library was built to chronicle Apple. So when Cain surfaces a NeXT dealer in Tulsa or a NeXTSTEP pivot meeting in Chicago and no one else has it, that uniqueness is partly an artifact of scope. Cain isn't pulling stories the others missed—he's pointing a different camera entirely.

That doesn't make the stories less worth telling.

The Funniest

The exile years were not light, but Cain's sources—reuniting decades later at memorials, dinners, and the Computer History Museum—remember a lot of dark comedy.

Burning the indestructible Cube. The funeral pyre at Livermore is its own joke about NeXT. Jobs had specified an exotic magnesium alloy so durable the press could not destroy it on camera. A weapons-lab pyrotechnician spent four hours trying. Garfinkel watched and decided it was the perfect metaphor: ambition, perfection, consumption by flames. Even failure at NeXT was overengineered.

NeXT's Christmas dump at the spy base. December 1992. NeXT needs the quarter's revenue. The National Reconnaissance Office—the secret satellite agency that hides behind a Los Angeles Air Force Base address—has agreed to buy thousands of Cubes, but does not want delivery until January. NeXT delivers anyway. A dozen tractor trailers roll to a closed base during the holiday shutdown and unload thousands of black cubes. One trailer crushes a picnic table. A salesman named Jim gets in "a little trouble." NeXT books the revenue and lives another quarter.

The rave paid for by the NSA. A NeXT software developer named Andrew Stone receives a $300,000 software order from what looks like a small town in Texas. He uses part of the proceeds to throw a psychedelic rave at the Palace of Fine Arts: smart drugs, fog, colored lights, vegetarian buffet, Genesis P-Orridge on the speakers, ecstasy and LSD on the floor. Sometime after the dancing ends, the phone rings. The buyer was the NSA. The Texas town was cover. Stone flies to Fort Meade conflicted, then discovers the agents he meets are also Deadheads.

Elsewhere Cain documents NeXT salesmen building a cardboard "Steve on a stick" to substitute for the man himself, and Ross Perot toasting a Stanford dinner with a warning: he was the grain of sand inside the oyster, and it would be hell on the oyster.

The Most Surprising

NeXT's commercial story is one long slide. Cain's surprise is that the slide kept producing accidental immortality.

The web is born on a NeXT Cube. Tim Berners-Lee unboxes a matte-black NeXT Cube in October 1990 at CERN. Objective-C, Interface Builder, drag-and-drop menus—all of it lets him compress a year of work into months. By December, info.cern.ch is online. The first web server in the world is one machine, in one office, with a red-ink label stuck to its case: "This machine is a server. DO NOT POWER IT DOWN!!" For years that label was the entire World Wide Web's redundancy strategy.

Dell's one-week WebObjects miracle. Michael Dell wants a website where buyers can choose processors, memory, features, and check out with a credit card. He calls IBM. IBM quotes two years. He calls NeXT. Sina Tamaddon takes notes, vanishes mid-meeting, and returns in the same meeting with a working WebObjects demo. "We can get that ready for you in one week," he says. Within a year, Dell.com is selling $3 million a day—the strategy that lets Dell embarrass HP and Compaq through the late 1990s.

The $54 million monthly pilgrimage. For nine years before Pixar's IPO, Ed Catmull walked into NeXT's offices, briefed Jobs, and asked for another check. Jobs wrote one. By the end he had put in $54 million—roughly half his net worth—while NeXT itself was failing. Two companies on life support, both quietly carrying the seeds of two industries. The investment that defined the 1990s for Jobs is the one almost nobody saw him make, $54 million spread across nine years of pilgrimage.

Cain also reports the Q3 1991 channel-stuffing collapse, where Jobs flew to London and purged seven European country heads one by one in a single afternoon, and the Tulsa NeXT dealer who watched a $275,000 police contract evaporate and then received a $156,000 bill from NeXT for the inventory that had wrecked his business.

The Most Insightful

These are the stories that change how you read the rest of the Jobs canon.

Andy Grove's question stumps NeXT. Santa Cruz, summer 1991. Intel's CEO walks into a NeXT offsite and asks the leadership team, one by one: "What business are you in?" Six years in, the answers splinter. Mainframe on a desk. Apple rival. Scholar's workstation. The room cannot agree. Ron Weissman remembers Grove's questions as profoundly simple and lethal. The silence is the diagnosis.

The customer tells NeXT the product is software. O'Connor & Associates—a Chicago trading firm—is the customer keeping NeXT's payroll alive, buying souped-up, overpriced workstations purely as a vessel for NeXTSTEP. An O'Connor principal closes the door of Phil Wilson's office and says it plainly: the value is the software, not the hardware. Wilson emails Jobs the case. Jobs snaps back: "You're an HR guy," and threatens his job. Within two years, Jobs kills NeXT hardware anyway. The customer was right; the founder needed two more years to hear it.

The ex-NeXTers return to Redwood City. October 2011. Nearly 200 ex-NeXTers gather at a Sofitel hotel near where NeXT began. Todd Rulon-Miller stands up and tells the IBM 125-page contract story, the Cube color arguments, Ross Perot's Texas drawl. Then his voice softens. Jobs, he says, was tough as nails. He was caring. He had a big heart, carefully hidden. He had created their futures. The survivors laugh, wince, mourn. The most honest accounting of Jobs is held by people who carry the damage and the gift in the same sentence.

Cain also covers Andy Grove's other warning—NeXTSTEP would die without Intel—and the Pixar acquisition note where Jobs explicitly rejected the disposable-crew model that defined movie production. The asset, he insisted, was the team.

What Cain Adds

The exile years are not unwritten. Schlender's Becoming Steve Jobs moves through them. Isaacson covers them. Catmull's own Creativity, Inc. sits in our library. Randall Stross's Steve Jobs and the NeXT Big Thing, written from inside NeXT itself, hasn't been ingested yet but is on the shelf.

Cain's posture is different. He and Catmull and Lewin function less as biographers than as conveners—leaning on a Computer History Museum oral history project to put NeXT and Pixar veterans back in the room with each other. What comes out is a book full of details only the people in the room remember: the picnic tables crushed by trailers, the magnesium pooling on a burn-cell floor, the $300,000 software order that turned out to be from Fort Meade. Stories small enough that no biographer reaching for the larger arc would stop to record them.

I attended the Computer History Museum's live event tied to this book—Cain moderating three famous NeXT alumni—and it's the same dynamic on stage. The video is on YouTube and highly recommended. Watch the alumni interrupt each other to correct dates and argue about which color the Cube fight was actually about.

The Cube at Livermore would not burn. The exile years were similar: built to last in ways the market did not need, almost destroyed, then turned into the platform a billion people now use. Cain's gift is showing that most of that turn was accidental, attended by people who only realized later what they'd done.

Adding to the Library

Steve Jobs in Exile by Geoffrey Cain with Ed Catmull and Dan'l Lewin — 255 stories, 69 found nowhere else in the Apple Library. The oral history of the seven years Apple's library tends to skim.

Tim Cook's biggest hardware bet is being quietly buried. That's usually how Apple finds the next thing.

In late 1996, while still negotiating his return to Apple, Steve Jobs called Gil Amelio two or three times a day. Some calls were about lawyers and deal terms. Some were about something else. Kill Newton, Jobs would say. People would cheer.

The Newton was John Sculley's project. A handheld digital assistant with a stylus and handwriting recognition and a name that nodded at gravity. Launched in 1993, it was supposed to be Apple's iPhone before there was an iPhone. Amelio refused to kill it. Within months he was gone, and Jobs did the thing he had been urging.

At a San Francisco Macworld in early 1998, the Newton was killed. So were the eMate, the printer division, and the licensed Mac clones. More than three thousand people lost their jobs. "God gave us ten styluses," Jobs said. Let's not invent another.

The engineers who had been building handhelds did not go home. They were reassigned. Three years later, Apple shipped the iPod. Nine years later, the iPhone.

The Quiet Burial

This week, Bloomberg's Mark Gurman reported that Apple will not ship a new Vision Pro — or any new mixed-reality headset — for "around two more years at least." The cheaper Vision Air, meant to be the headset's path into normal people's homes, was quietly canceled in October. A small M5 refresh of the original Vision Pro shipped the same month, and then, mostly, silence.

The Vision Products Group still exists on the org chart. Last month, marketing chief Greg Joswiak put out a statement insisting it has "not been entirely dissolved" — a phrase of the kind only PR departments produce. According to Gurman, the bulk of Apple's mixed-reality hardware talent has been pulled onto other things: lightweight smart glasses, AirPods with cameras, an AI pendant, Siri.

Apple has conceded the headset.

The Wrong Answer to the Right Question

This is not a small concession. Vision Pro shipped in February 2024 after eight years of work and nearly five thousand patents. Mike Rockwell, recruited from Dolby in 2015 to hunt for the platform that would come after the iPhone, built it from a skunkworks in the old Mariani 1 building. His team once hung a fifty-pound prototype from a crane. It was the largest hardware ambition of the Cook era — Apple worked on it longer than it took to build the original Macintosh.

And like Sculley's Newton, it turned out to be the wrong answer to the right question.

The question — what comes after the phone? — is real. Meta has been selling Ray-Bans with cameras at a pace nobody at Apple expected. Google is shipping Gemini-tied glasses. The form factor is settling, and it does not look like a 1.4-pound goggle. It looks like eyewear. Apple knows this. The engineers Rockwell built up are now working on it.

The Closer Template Is Titan

The most recent template for what Apple is doing here is not the Newton. It is Project Titan.

On February 27, 2024, Jeff Williams and Kevin Lynch told the engineers of Apple's car program that it was over. Ten billion dollars. A decade of work. About two thousand people reassigned, many to AI and machine learning. Six months later, Apple Intelligence was announced at WWDC.

The redirect was the announcement. The car got killed so the engineers could be sent to the next thing — which, it turned out, was the thing Apple actually needed.

Vision Pro is now eighteen months behind on the same script. The biggest hardware bet of the Cook era is dying so the engineers can be moved to a hardware bet that may define his successor. On September 1, John Ternus takes over. He inherits a Vision Products Group that exists on paper and a smart-glasses program that exists in practice.

Did the Moonshot Fail, or Did It Pay for the Next Thing?

The Newton failed. The iPhone is, by some accounting, what Apple's engineers built once they had nine more years of components, batteries, networks, and silicon to work with. Project Titan failed. Apple's AI hardware story is, partly, what its engineers are doing now that the company has admitted it is not going to ship a car. Vision Pro has already failed at the thing it was supposed to do — put a spatial computer in a hundred million faces — but the engineers who built it now know exactly how to make the next thing smaller, lighter, cheaper, and tied to whatever AI Apple ends up using.

Sculley's tablet died so Apple's future could happen. We will see what happens to Cook's headset.

From the Library

This newsletter draws on a library of books about Apple's history. Today's issue featured stories from:

Steve Jobs by Walter Isaacson — the Newton kill at Macworld 1998, and Jobs's "ten styluses" line.

Becoming Steve Jobs by Brent Schlender — the broader 1998 purge: clones bought out, printers shut, more than three thousand laid off.

On the Firing Line by Gil Amelio — Jobs's pre-close calls urging Amelio to kill Newton, two or three times a day.

Apple by David Pogue — Mike Rockwell's 2015 skunkworks, the 50-pound crane prototype, and the February 2024 Titan cancellation.

Eddy Cue gets a trophy from Cannes in June. The Formula One rights he wants are locked up till 2034.

In April 2003, Sony Music's new chief Andy Lack flew to Tokyo carrying an iPod in his jacket pocket. Sony CEO Nobuyuki Idei watched from the front of a room of two hundred Sony managers. Howard Stringer sat near him. Lack stood up, reached into his pocket, and held the iPod aloft.

"Here it is," he said. "Here's the Walkman killer."

He told the room there was no mystery left. Apple had built the thing Sony's hardware engineers and Sony's music label had been trying not to build for a decade. The future had shipped. On the Apple side, Steve Jobs had led the label charm offensive; the operational lieutenant who ran the licensing slog underneath him was an Apple lifer named Eddy Cue.

That deal was for the United States only. Getting iTunes onto European charts took another fourteen months of country-by-country bargaining. The Beatles took seven years. Cue learned the shape of the rights business in those years: it doesn't sell whole; it sells by territory, in pieces, slowly.

What happened this week

On May 19, Sky paid a billion pounds to extend its Formula One broadcasting rights in the United Kingdom through 2034 and in Italy through 2032. Apple's existing F1 deal — five years, around $150 million a year, U.S. only — runs until 2030. The U.K. window Apple wanted just closed for eight years.

Cue, asked about it, said the obvious thing. "The sport doesn't get licensed on a global basis... Do I hope that we are able to grow into other areas and markets? Yeah, I do."

That was the quote of a man who has done this before.

Why 2026 isn't 2003

In 2003 Cue was bargaining with a music industry that was bleeding from piracy. Napster had been shut down. Kazaa was eating CD sales. As Steven Levy recounts in The Perfect Thing, Jobs went on a label-by-label charm campaign that fall, pitching iTunes as a "five percent virus" — a small Mac-only experiment that couldn't possibly hurt the labels' business. The labels were terrified enough to let him try.

Sky isn't terrified. It is operating a healthy oligopoly. The United Kingdom has Premier League football locked down through Sky and Comcast's NBC. Italy has Sky Italia. France has Canal+. Germany has Sky Deutschland. These broadcasters are not staring at a piracy gun. They are signing nine-year deals. Apple's pitch — bundle F1 into a subscription that also has F1 the Brad Pitt movie and Ted Lasso — does not produce the same panic.

This isn't a one-off mismatch. It is the structure of the global rights business. The U.S. Major League Soccer agreement that became Apple TV's MLS Season Pass worked because the league owns its global rights centrally — one signature, one product, every country. F1 doesn't. The NBA doesn't. Most leagues don't. Apple's checkbook only works when the seller has done the consolidation for it.

What Cue actually knows

Cue has spent twenty-eight years negotiating with the people who control how Apple's products reach customers. He started as a logistics manager. Brent Schlender writes in Becoming Steve Jobs that Cue walked into his first meeting with Jobs in 1998 carrying a paper mockup for an online Apple store. Jobs's verdict: "These suck." Cue lost his temper. "We lost two billion dollars last year! Who gives a fuck about the channel?" Apple was bleeding money the year Jobs returned, and a big part of the bleeding was the retail channel — CompUSA, Sears, the resellers Apple depended on to reach customers. Cue's pitch was to bypass them and sell direct. Jobs lit up. Cue got the store. It sold a million dollars in six hours.

That fight put him in charge of distribution at Apple. Carriers next, then labels, then publishers, then Hollywood studios. Now sports leagues. Cue's job, in every case, is to negotiate with someone who owns something Apple wants to sell. He almost never gets the whole thing on the first try. The Apple TV+ launch itself, David Pogue writes in Apple, happened only after Apple gave up on cable: if Apple couldn't own live television, it would own the shows.

Which is why the second piece of Cue news this week matters. On June 22, he will pick up the Cannes Lions Entertainment Person of the Year award and deliver a keynote with Jerry Bruckheimer — producer of Apple's F1 movie. The trophy is for what Cue made when he couldn't buy what he wanted. The pattern is the lesson.

What happens next

Watch the broadcast contracts that expire first. Foxtel in Australia, DAZN in Japan, the long tail of European F1 deals — they fall on different calendars, and Apple's expansion will be paced by their expirations, not by anything decided in Cupertino. Some Cue will win. Some he won't. By the early 2030s the F1 map will look something like the iTunes Music Store map looked in 2010: most of the world covered, a few holdouts, a few markets where the incumbent still owns the customer.

The man being honored at Cannes in June already knows this. He has done it once. Twenty-three years ago the music business unbundled itself one country at a time because Napster forced the panic. This time the sport never bundled in the first place, and there is no panic at all. Just nine-year contracts, in pounds, in countries Apple wanted.

Also This Week

Mike Rockwell is now spending most of his time on Siri. Bloomberg's Mark Gurman reports the Vision Products Group has been folded back into broader engineering, and the man who built Apple's headset from 2015 now runs Siri and visionOS together. No new Vision Pro for at least two years. The pattern echoes the breakup of the original Macintosh team in 1985 and the absorption of the Newton group in 1998 — Apple's standing answer for what to do with moonshot talent once the moonshot misses its market.

Fortnite is back on the App Store worldwide. Epic Games's Tim Sweeney is doing victory laps before the final ruling. The five-year procedural defense of the thirty-percent commission rate just ran out of countries.

Johny Srouji reshuffles product design. Kate Bergeron loses oversight; Shelly Goldberg and Dave Pakula split her territory. Silicon and industrial design now sit under one umbrella for the first time since 2019, when Jony Ive left.

Apple's first retail stores turn twenty-five. Tysons Corner and Glendale Galleria opened on May 19, 2001. At the time, analysts called them a Gateway-store copy and predicted they would close inside two years.

From the Library

This newsletter is built from a working library of books about Apple's history. Today's issue drew on four of them.

Steve Jobs by Walter Isaacson — Andy Lack's Tokyo iPod demo; the 2010 Beatles boardroom summit.

Becoming Steve Jobs by Brent Schlender — Cue's 1998 outburst that won him Apple's online store; Lack's eventual capitulation to iTunes.

The Perfect Thing by Steven Levy — Jobs's 2002 "five percent virus" label-by-label campaign.

Apple by David Pogue — the Apple TV+ pivot, after cable broke Apple's first run at television.

September 7, 2005. Steve Jobs stood in Moscone West holding a Motorola ROKR — Apple's first phone, of a sort. Plastic. Cluttered. Capped at one hundred songs by firmware. Jobs could barely fake enthusiasm. He demoed it briefly and moved on to the iPod Nano, which everyone in the room — including Jobs — actually wanted to discuss.

The ROKR was a hedge. Eddy Cue had been hunting since 2003 for a phone that could run iTunes. Motorola, fresh from RAZR dominance, looked like a partner. The result was, in Wired's eventual headline, a disgrace. But Apple paid a small price and learned an enormous lesson — that the carriers and the committee designs would never let Apple ship the thing it wanted. Sixteen months later, Macworld 2007. Different room. Different product.

The report

This week The Information's Aaron Tilley reported that Apple is "exploring" how to let third-party AI agents run inside the App Store. The verb matters. Apple is not announcing. Apple is internally discussing. The concrete worry in the report: agents that "go haywire and delete all of a user's emails," and a class of apps that can "spin up smaller apps on the spot" — software that was never submitted to App Store review. Current guidelines prohibit apps that produce other apps on iPhone. The thing AI agents are best at is the thing Apple explicitly disallows.

WWDC opens June 8. The Information's caveat: it is unclear if Apple is ready to unveil anything.

Plumbing without a product

Apple has been laying pipes for this for two years. App Intents shipped in iOS 16 — a framework for apps to expose actions to Siri. The on-device Apple Foundation Model arrived with iOS 18. The Gemini deal in April brought cloud-scale reasoning under a private contract. The infrastructure exists. The product does not. Apple knows it must let agents do things on its platform; it does not know what those things should look like, charge for, fail at, or be reviewed for.

iPad shipped under the same dynamic in 2010. Jobs unveiled a tablet that pundits called "a big iPod Touch," and even Jobs could not fully explain what tablet software was for. The App Store filled it in over five years: Procreate, Paper, GarageBand, eventually the iPad Pro use cases that justified the price tag. Apple shipped the pipes; developers told Apple what flowed through them.

The constraint Apple chose

There is a sharper reason than curiosity that Apple needs third parties here. Two years ago at WWDC, Craig Federighi unveiled Private Cloud Compute and made a promise: Apple's AI servers would not see your data. That promise is architectural and binding. It is also a tax. PCC limits how big the model can be, how often it can be called, and what data it can reach. Google Pixel and Samsung Galaxy face no such limit.

Third-party agents on the App Store are the architectural escape hatch. A user who wants Claude or Perplexity-grade reasoning installs an app, signs the privacy bargain with that vendor, and Apple's promise at the OS layer stays clean. The user pays the privacy tax per app. Apple keeps the platform pristine. The same logic that built the 2008 App Store — Apple could not write two hundred thousand apps alone — now applies to AI: Apple cannot match Pixel capability under PCC alone. Developers carry the load Apple won't.

On the same day Tilley's report appeared, Apple was in Brussels defending Google's right to keep Android closed against EU rules that would force Google to open it to AI rivals. The draft measures, Apple said, would create "profound risks for user privacy, security and safety." Apple opens its store to capability it cannot ship itself, and defends Google's gates against rules that would force the same opening on Android.

What review can't catch

App Store review assumes apps are deterministic. A reviewer installs the app, performs an action, observes the result, decides. AI agents are probabilistic. Two reviewers running the same agent against the same prompt may see different behaviors, only one of which is destructive. An agent that "spins up smaller apps on the spot" has never been submitted to review at all; the agent invented it after the user asked.

Apple's defense against malware on iPhone has always been the slow, deterministic chokepoint of human review. Agents are the first class of software that don't fit that chokepoint. The Information's reporting reads less like a strategy and more like a list of fears.

What to watch

The ROKR sold poorly, got discontinued within a year, and taught Apple nothing about phones except what not to build. The iPhone came from somewhere else inside Apple — Scott Forstall and Tony Fadell working on the real thing while the ROKR moved through Cingular stores.

The App Store agent experiment, if it ships at WWDC, will be the ROKR. Limited surface. Artificial constraints. A handful of vibe-coding apps and chat assistants sandboxed within an inch of their lives. The constraints will reveal what Apple is actually scared of.

The tell will be where Apple's best engineers are. If they're building agent review tooling for the App Store, this is the strategy. If they're building a Siri rebuild that doesn't need third-party agents — or the next ambient device that makes "agent on phone" feel quaint — then this week's news was the placeholder. The cheap education Apple bought while it figured out the real answer.

Also This Week

iOS 26.5 ships with RCS end-to-end encryption. Apple drove an E2EE upgrade for green-bubble messaging ahead of Google's own rollout. The embrace-and-extend reflex, applied to a standard Apple resisted for years.

India sets May 21 final hearing in $38B antitrust case. Apple is refusing to disclose global financials to the Competition Commission of India. The largest potential antitrust fine in history, on the soil of Apple's fastest-growing manufacturing base.

Apple TV International head Morgan Wandell departs. With Apple TV+ since before launch; leaves to start his own production company. The second senior streaming exit in months.

From the Library

This newsletter draws on the Apple Library. Today's issue featured stories from:

Becoming Steve Jobs by Brent Schlender — The ROKR fiasco, Forstall's secret SDK, the iFund call to Doerr.

Steve Jobs by Walter Isaacson — "The device that can eat our lunch": the iPod panic that birthed the ROKR.

Apple by David Pogue — The Forstall disobedience that saved the iPhone's future; the ROKR humiliation.

The Intel and Samsung talks aren't a reversal. They're a return to a doctrine that never went away.

In 2010, Jeff Williams flew to Hsinchu.

Apple's chip operations chief was there to make Morris Chang a deal. Apple would buy chips from TSMC. TSMC would borrow seven billion dollars and build a fab in eleven months — a pace Chang's engineers thought impossible. In exchange, TSMC would take half the order. Not all of it. Half. Williams pushed Chang to suspend TSMC's shareholder dividend and plow the cash into the build. Chang refused. The deal closed anyway.

The detail that matters isn't the seven billion dollars or the eleven months or the dividend fight. It's half. Apple flew across the Pacific to give a foundry the gift of its business — and structured the gift so no foundry would ever own it alone.

Five days ago, Bloomberg reported that Apple is in early talks with Intel and Samsung to fabricate Apple Silicon in the United States. The leading candidate process is Intel 18A-P, scaling next year. A Cupertino team has visited a Samsung fab in Texas. Commerce Secretary Howard Lutnick has been on the phone with Tim Cook for the better part of a year. The Trump administration owns ten percent of Intel.

None of this is confirmed. Mark Gurman, who has the best chip-plan sources inside Apple, posted his hedge the day after the scoop: still early, not signed, Apple is still concerned about Intel's technology. Apple itself has said nothing.

The mainstream framing came in fast and uniform. Apple is bailing out Intel. Apple is reversing its Silicon strategy. Apple is capitulating to political pressure.

It's none of those things.

Apple still designs the silicon. The M5 is Apple's chip. The A20 is Apple's chip. A future M-series chip printed at an Intel fab would still be an Apple chip. Intel and Samsung — if any of this happens — would do what TSMC alone has done since 2016: print Apple's designs in their fabs.

The 2020 transition was about the design layer. Cook never said Apple would never want a third party near its chips. He said the design and the silicon would be Apple's. They still are. What might change is the fabrication layer. And on that layer, multi-sourcing has been Apple's default for the entirety of Cook's career.

Cook arrived at Apple from Compaq in 1998. Before Compaq he spent twelve years at IBM, where he watched the company rack up fifteen billion dollars in cumulative losses by the early nineties — losses driven, in no small part, by supply concentration that left IBM unable to negotiate or pivot. He arrived at Apple ninety days from bankruptcy. His first move was to cut Apple's suppliers from 100 to 24, warehouses from 19 to 9, inventory from 30 days to 2.

That looks like concentration. It wasn't. It was consolidation of redundant vendors per part. The ones that didn't pass the bar got cut. The ones that remained still competed against another supplier on the next bid.

Look at the iPhone today. DRAM: Samsung, SK Hynix, and Micron. OLED panels: Samsung, LG, BOE. Cameras: Sony and Samsung. Assembly: Foxconn, Pegatron, Wistron, Luxshare, BYD. Even when Samsung threatens Apple's sovereignty over its own product — copying the iPhone, training engineers on Apple's specs — Apple still buys panels and chips from Samsung. A second source is not a backup. It is the plan.

The single exception, since 2016, has been the brain.

It wasn't always. The A4 through A7 ran on Samsung. The A8 ran on TSMC. The A9 ran on both — same iPhone, two fabs, two processes (Samsung 14nm, TSMC 16nm). Reviewers eventually noticed. The TSMC variants ran cooler and lasted longer. Apple's PR called the difference not representative of real-world usage. Chipgate, in 2015, was the last time Apple multi-sourced its main processor.

Every A and every M since has been printed in Taiwan, by one company, on one island, fifty miles from the world's largest standing army. Patrick McGee's Apple in China describes the 2020 commitment as doubling down on a single-island supply chain. The corpus doesn't call this a strategy. It calls it a fragility.

The fragility didn't persist because Cook stopped believing in second sources. It persisted because there was no second source. Samsung fell behind. Intel fell apart. GlobalFoundries gave up on the leading node. TSMC was, for a decade, the only foundry on earth that could fabricate Apple's most ambitious chips at the speed Apple ships them.

There's also the matter of slots. Apple is no longer TSMC's biggest customer — Nvidia is, 19% of TSMC's 2025 revenue to Apple's 17%, fed by AI capex from every hyperscaler at once. TSMC's leading-edge nodes are booked solid through 2026. On Apple's Q2 earnings call ten days ago, Cook named the supply constraint on Mac sales as "the availability of the advanced nodes our SoCs are produced on." Every TSMC customer is getting less than they want, Apple included. The doctrine and the squeeze point in the same direction.

Whether Intel 18A-P qualifies as that second source is the open question. Ben Bajarin, who tracks foundry economics for a living, calls 18A "a little bit rough" and 18A-P the version that "cleans a lot of stuff up" — the first credible second source. Ming-Chi Kuo expects the lowest-end M-series on 18A-P as soon as next year. Apple's own engineers, per Gurman, are not yet convinced. None of the Cupertino-Hillsboro-Pyeongtaek triangle is signed.

Lutnick is pushing on an open door. The political pressure to put logic in American fabs is real, and it ought to be named. But it isn't what's pulling Apple toward Intel and Samsung. The pull was already there. It has been there since 2010, when Jeff Williams flew to Hsinchu and asked for half.

Cook learned something at IBM in 1991 that he has spent thirty-five years applying. The supplier you don't have is the one that kills you.

Also This Week

AirPods with cameras hit advanced testing. Bloomberg reports the hardware is near-final, but launch is gated on the Gemini-powered Siri shipping in iOS 27. Apple's next form-factor calendar is now downstream of software it doesn't own.

Canada's Bill C-22 — second Five Eyes encryption fight in twelve months. Apple signaled it would pull features in Canada if the bill passes, mirroring the UK ADP withdrawal. Same week iOS 26.5 RC2 shipped E2E-encrypted RCS. Sometimes the juxtaposition writes itself.

India CCI sets May 21 final hearing. The fastest-growing manufacturing base for Apple is also the venue for what could be the largest antitrust fine ever levied on the company. Both numbers are heading up at the same time.

Apple settles $250M Siri "personalized" class action. A rare admission-by-settlement that the WWDC 2024 demo was oversold. Cash receipt for a feature Apple couldn't ship in-house. The Gemini deal looks better in this light, not worse.

From the Library

This newsletter draws on a library of books about Apple's history. Today's issue featured stories from:

Apple in China by Patrick McGee — Williams's 2010 trip to Hsinchu, TSMC's eleven-month fab build, the dividend negotiation, the framing of the 2020 Apple Silicon move as a single-island bet, and Cook's IBM-and-Compaq formation in supplier discipline.

Steve Jobs by Walter Isaacson — Cook's 1998 arrival at Apple, the supplier consolidation from 100 to 24, and the inventory-as-the-enemy doctrine that has governed Apple operations for twenty-eight years.

Apple is paying Google a billion dollars a year for a tutor. The terms include the right to learn.

The story most outlets told about Apple's Gemini deal was a story of surrender. Bloomberg said Apple was paying roughly a billion a year because it couldn't catch up internally. The Financial Times called the deal "a worrisome sign, suggesting that Apple is still struggling to build its own LLM." Fortune ran the headline "Google wins in AI deal that highlights Apple's own AI struggles."

On Dithering, Ben Thompson made the smartest version of the same argument. He compared Apple's situation to a chip foundry: once you skip a process node, you don't get to come back later. "Like global foundries bailed at, was it, fourteen nanometers or ten nanometers," he said. "They don't get like now decide, oh, I'm going to come back in at two nanometers. It's like you need to like go down that path." The corollary: Apple has just committed itself to depending on third parties for AI for the long run.

It is a respectable argument. It is also, I think, wrong — though I am hedging that think, and I'll tell you what would change my mind.

The buried lede

In late March, The Information reported something that didn't make the headlines. Apple's deal doesn't just license Gemini's outputs. It grants Apple full access to a custom 1.2-trillion-parameter Gemini running in Apple-controlled datacenters — and the right to distill that model into smaller variants. Distilled models, as the report put it, "learn to imitate the internal computations" of their teacher, not just its outputs.

If you have been paying attention to AI in 2025 and 2026, you know this is no longer a curiosity.

In federal court in April, Elon Musk was asked under oath whether xAI used distillation on competitors' models to train Grok. He answered "Partly," and described it as "a general practice among AI companies." DeepSeek released a 32-billion-parameter distillation of its R1 model that outperformed OpenAI's o1-mini on five of six benchmarks. Anthropic in February caught Chinese frontier labs — DeepSeek, Moonshot, MiniMax — distilling Claude across more than sixteen million exchanges, using over twenty-four thousand fraudulent accounts to do it. Google's own open-weights Gemma family is itself a Gemini-distillation.

Distillation is the skip-ahead lever. The fab-node analogy — you have to go down that path — does not apply, because the teacher is doing the path-walking.

What "fullstack" means in Cupertino

There is a shape to how Apple uses partners that recurs across forty years.

In 1997, Apple took a hundred and fifty million dollars from Microsoft. The framing at the time was rescue capital — humiliating, structurally subordinate. What Apple actually did with that breathing room was ship iTunes on Windows in 2003 and ride Microsoft's installed base into the iPod's billion-unit decade. Apple borrowed the rival's platform for long enough to make itself unborrow-from-able.

In 2006, Apple announced that Mac would run on Intel. The decision was framed as a confession: PowerPC could not keep up. What Apple did during the next fourteen years is well-known now and was barely visible at the time. It bought P.A. Semi in 2008. It shipped the A4 in 2010. By 2020 it had walked off Intel's path entirely, into a vertically integrated silicon stack that Intel itself cannot replicate. The transition that began as licensing ended as ownership.

The cautionary inverse is Apple Maps in September 2012. Apple left Google before its own map data was ready. The product shipped misshapen; Cook apologized; the rebuild took years. The lesson Maps taught Apple is the same lesson the Gemini deal seems to have absorbed: leave the teacher when the substrate is ready, not before.

This is what "fullstack" means in Apple's vocabulary, and it is not the same thing it means in Mountain View. Apple's stack is silicon plus on-device inference plus private user data. AFM-on-device today runs at roughly three billion parameters in about a gigabyte, performing favorably against Qwen 2.5-3B — and the M5 added Neural Accelerators inside the GPU specifically for LLM inference. Frontier scale is not the game Apple is trying to win. It is the game whose output Apple is trying to inherit.

What I'd watch

Apple has run this kind of apprenticeship before, but never with a model. This is a thesis under uncertainty. Several conditions could prove me wrong, and I'd rather name them now than back into them later.

Maybe copying isn't enough. Apple is shrinking Gemini into a smaller model. The smartest parts might not survive the shrink. If they don't, Apple stays a step behind. Watch: WWDC June 8 — does Siri actually get smarter?

Apple's chips might not be strong enough. Private on-device AI only works if iPhones and Macs can run real models, not toy ones. Watch: iPhone 18 and M5 Pro — how big and how good is the AI that runs on the device?

The deal won't last. Once Google sees Apple using Gemini to build its own AI, Google can change the terms or end the contract. Watch: what the next deal looks like, and whether Apple has a backup ready (Anthropic, or its own model).

Maybe Apple is just guessing. Apple already uses Claude inside the company. OpenAI said no to Siri before Google said yes. That looks less like a careful strategy and more like Apple grabbing whatever it can. Watch: does Siri actually get better, or does Apple keep switching partners?

Maybe people don't care about AI privacy. Privacy helped sell iPhones. It might not sell chatbots — most people are fine handing their questions to ChatGPT or Gemini. Watch: how many people actually turn on Apple Intelligence.

A billion a year is what apprenticeship costs at this scale. The interesting question isn't whether Apple is the apprentice. It's what the apprentice does once the master has been copied.

From the Library

This newsletter draws on books about Apple's history. Today's issue rhymed with stories from:

— Becoming Steve Jobs by Brent Schlender — for the 1997 Microsoft rescue and the years Jobs spent letting iTunes ride Windows.
— Steve Jobs by Walter Isaacson — for the 2006 Intel transition framed as confession, and for what licensing-then-ownership looked like before the model was a model.

April Fools' Day, 1996 — Apple's twentieth birthday. Fred Anderson, the company's brand-new CFO, walked into Cupertino and found a house on fire. Losses for the year would land near $750 million. Cash on hand: roughly $500 million, dwindling. $150 million in loans were due to Japanese banks within weeks. Bankruptcy counsel — Harvey Miller of Weil Gotshal — was already on retainer, in case.

The most embarrassing fact wasn't that Apple was broke. It was who was bankrolling it. Suppliers were paying upfront for tooling so Apple could keep making products. LG fronted the iMac molds. The personal-computer industry's most cash-rich era, and Apple had become the supplier's charity case.

Anderson sold the Fountain factory. Raised $661 million in an oversubscribed bond. Bought time. Steve Jobs returned, fired half the company, and started building a war chest.

Apple turned fifty four weeks ago. The war chest now stands at $147 billion.

The Cash Weapon

In 2005, Apple wrote a $1.25 billion check to lock up flash memory for the iPod nano. In 2008, as Adam Lashinsky recounts in Inside Apple, Tim Cook's team made "billion-dollar forward purchases" of components that "starved rivals' access to the same parts." That same year, as Patrick McGee documents in Apple in China, Apple bought roughly 10,000 CNC machines — essentially everything FANUC could build for three years — and turned them into the unibody MacBook Pro. Rivals saw the product. They couldn't see the supply chain. By the time they could buy the same machines, Apple had moved on.

This is the trick nobody calls a strategy because it doesn't sound like one. Pay early. Pay more. Take the supply off the market.

The Quiet Announcement

On April 30, in the middle of an otherwise routine Q2 call, CFO Kevan Parekh said something that should have been the headline:

"Net cash neutral has been a valuable framework for our capital structure… we are no longer providing net cash neutral as a formal target."

Apple has held that target for over a decade. Returning roughly all free cash to shareholders. It was a discipline: don't accumulate, don't get lazy. Then, with $147 billion in cash and a $100 billion buyback authorization in his pocket, Parekh quietly retired it.

Same call: R&D hit $11.4 billion for the quarter, up 34% year-over-year — a record. Cook said the constraint hurting Mac sales was "the availability of the advanced nodes our SoCs are produced on." TSMC's 3nm and 5nm are 100% booked through 2026. DRAM is forecasted to climb 25% over the next two quarters. Advanced packaging — the part of chip-making that bonds AI dies to memory — is the new bottleneck behind the bottleneck. Intel is begging the U.S. government for help to compete with TSMC.

The buyback is the cover. The R&D number is the story. The retired target is the optionality.

The Mac Neo Is the First Salvo

Cook said it twice on the call, almost like he didn't trust himself the first time:

"The customer response to Mac Neo has just been off the charts, with higher than expected demand."

Mac mini and Mac Studio "are amazing platforms for AI and agentic tools, and the customer recognition of that is happening faster than what we had predicted."

The Mac Neo isn't a creative-pro product. It's a local-AI host. Agentic workloads — coding assistants, research tools, software that actually does things — want to run on the same machine as the developer. The cloud isn't fast enough. The privacy isn't good enough. Every minute the user spends waiting on a remote model is a minute they're reminded that someone else owns the compute.

For three years, the consensus has been that Apple is losing the AI race. The Magnificent Seven's other six are pouring more than $100 billion a year into AI infrastructure. Apple's capex through the first half of FY26: about $4.3 billion. To Wall Street, the gap looks like cope.

The Right Kind of Shortage

Read it the other way. The semiconductor industry hasn't been this constrained in decades, and it isn't constrained in just one place. Leading-edge fabs, sold out. Memory, climbing. Advanced packaging, gated. Alternative foundries — Intel, Samsung — wobbling. Multiple inflection points, all at once, none with a clear winner.

In a moment like that, the right move isn't to bet on a single supplier. It's to keep the dry powder for whichever bottleneck bites hardest. Apple just did. The retired target gives the balance sheet optionality — to pre-buy DRAM if the squeeze deepens, to lock 2nm if TSMC ramps faster than expected, to fund a second-source foundry if Intel ever delivers, to acquire if the right target appears. The deepest pocket doesn't have to be right about which constraint will matter most. It just has to be the one with cash when the constraint arrives.

Apple won't say this out loud. They never do. The 2008 CNC story didn't surface until McGee reported it fifteen years later. The flash prepayments showed up in earnings notes, not press releases. The trick is to take the supply off the market, then stay quiet.

What Happens Next

Cook flagged the June quarter as the one "where things will start to get worrisome." TD Cowen projects a 25% memory cost increase through Q1 2026. Apple has pre-bought as much as it can. So has the rest of the industry. The price competitors will pay to catch up is the price Apple already paid in 2005 and 2008 — except this time the asset isn't a single component. It's room to maneuver across a supply chain with too many bottlenecks at once. The stopwatch is running.

The 1996 question — who's bankrolling whom? — has a 2026 answer. Apple is bankrolling its own R&D. Apple is bankrolling its silicon roadmap. Apple is bankrolling whichever supplier wins the next round, whichever bottleneck breaks first. And, quietly, Apple is bankrolling its bet that the next decade of AI will run on the device sitting on your desk, not in a server farm somewhere in Virginia.

Net cash neutral is gone. Something else is taking its place.

Also This Week

Gurman: Vision Products Group dismantled. The headset team has been split up; "a large chunk of the visionOS software team was reassigned to fix Siri last year." Apple was retreating before it admitted retreat. [Bloomberg via MacRumors]

EU lands first DMA fine: €500M. Apple is choosing to pay rather than comply with the Commission's anti-steering ruling. Same playbook as Epic. [Lexology]

Touchscreen MacBook arriving 2027. Steve Jobs's "ergonomically terrible" doctrine officially walked back under Ternus. [9to5Mac]

From the Library

This newsletter draws on more than two dozen books about Apple's history. Today's issue featured stories from:

Apple in China by Patrick McGee — the 10,000 CNC machines and the FANUC three-year buyout that built the unibody MacBook Pro.

Inside Apple by Adam Lashinsky — the billion-dollar forward purchases that gave Cook's procurement team a household name.

On the Firing Line by Gil Amelio — the 1996 cash crunch from inside the burning house.

On April 20, Apple announced two successions. Only one made the front page.

In 1986, Apple bought a Cray. Not a workstation — an actual Cray, ten feet wide, blue and glowing in an antiseptic room near Cupertino with its own power substation. The Cray was there to design something Sam Holland's Advanced Technology Group called Aquarius: a single piece of silicon with four processors on it, what Holland's team called "a Cray on a desktop." Sculley signed the check. He had been told Apple could vault Motorola entirely if it built its own chips, and he liked the sound of it. Three years and roughly $20 million later, Aquarius was dead. Apple, the engineers admitted, had no fab experience and no business pretending it did. The team quit in disgust. The Cray went somewhere quieter.

That was Apple's first attempt to own its silicon. It would not be its last.

The second press release

On Monday, April 20, two press releases went out from Cupertino. The first announced that Tim Cook would step aside as CEO at the end of summer, and that John Ternus, senior vice president of Hardware Engineering, would take over. That release was carried on every front page on Earth.

The second was eight paragraphs long. It said that Johny Srouji had been promoted to Chief Hardware Officer, a title Apple had never used before. Buried in Ternus's memo, in a single line, was the structural news: Tom Marieb would take over Hardware Engineering — Ternus's old kingdom — and report to Srouji.

Ternus is the new CEO. Srouji runs the metal.

What it means

In the Jony Ive years, design ran Apple. In the Cook years, operations did. Now the building that matters is the chip lab.

Apple gave away the AI brain to Google last week, the deal Sundar Pichai pulled out of Cupertino as a marquee partnership at Cloud Next. That was Apple admitting it didn't need to own the substrate of intelligence. The substrate of everything else — the silicon — it would not let go of.

Apple-Gemini and Srouji-elevation are the same announcement, made in two parts.

The forty-year arc

The Srouji empire was forty years in the making. It started with Aquarius's blue Cray and a power substation; it almost died there. In 1997, Steve Jobs called Motorola's CEO Chris Galvin and told him Apple was done with PowerPC because "your chips sucked," then hung up. By 2006, Macs ran on Intel — a sweetheart deal Jobs cut with Paul Otellini that delivered the transition "without a noticeable glitch," as Bill Gates would later admit was something he'd thought impossible. In 2008, Jobs walked into Bob Mansfield's office and asked him to do quietly what Sculley's team had failed at twenty-two years earlier: design Apple's own chips. Mansfield recruited a young IBM engineer named Johny Srouji, told him there was no backup plan, and pointed him at the iPhone. The first Apple chip, the A4, shipped two years later in the iPad. Apple Silicon shipped in Macs in 2020. The Intel divorce — once thought impossible — completed in just under three years.

In 2026, the chips are not just made in-house. The chip guy runs the house.

What to watch

Two questions follow. The first is what Srouji's empire chooses to include. AirPods, Watch, Vision Pro, the rumored pendant — every one of these depends on what the chip team can deliver in a given thermal envelope. When Mark Gurman lists Ternus's six new product categories — AirPods with cameras, smart glasses, the pendant, HomePad, a tabletop robot, a security camera — every one is a silicon problem before it is a design problem.

The second is what the empire chooses to abandon. Vision Pro production is reportedly halted at Luxshare, and Mike Rockwell, the architect of the program, is said to be weighing exit. Apple's loudest statement of intent in 2024 looks, eighteen months later, like a project that gets quietly downgraded when the chip lab has more pressing demands. Watch which products Srouji's team continues to feed and which they let go hungry.

The press release Apple did not write would have said: the company has chosen what its crown jewels are. The world will hear that announcement in installments — through the products that ship, and the ones that don't.

Also This Week

Apple's Mac playbook arrives at the iPhone. Leaks suggest the base iPhone 18 will get a quietly cheapened chip, display, and manufacturing process — last-generation tech in a price-protected body — while the foldable iPhone Ultra reportedly goes the other way at $2,000–$2,500. The Mac line has worked this way since the M-series Air landed: pro tier moves upmarket, base tier inherits the year-old silicon. Same logic, new product line.

India's $38 billion antitrust hearing arrives May 21. Apple is openly withholding 2022–2024 financial data from the CCI, which is calculating the potential fine on global turnover rather than India revenue. Match Group is co-complaining. The pairing with Apple's 20% Q1 surge in China — back to the country's #1 smartphone after years of Huawei advance — is the geopolitical mirror of the year so far: winning where they expected to lose, losing where they expected to win.

WWDC's quiet tell. Mark Gurman noted that the WWDC 2026 keynote graphic features a glowing "Search or Ask" prompt, suggesting iOS 27 will ship a redesigned Siri with a standalone app. The conference begins June 8.

From the Library

This newsletter draws on books about Apple's history. Today's issue featured stories from:

Apple by David Pogue — Project Aquarius's blue Cray and the 2008 conversation in which Steve Jobs asked Bob Mansfield to design Apple's own chips with no backup plan

Steve Jobs by Walter Isaacson — Jobs's 1997 phone call with Motorola's CEO, the one that ended PowerPC at Apple

Becoming Steve Jobs by Brent Schlender — the Otellini handshake that made the Intel transition seamless

Apple in China by Patrick McGee — the 2020 Apple Silicon migration to TSMC, and the strategic cost of putting every chip on a single Taiwanese island

Google made the announcement today. Apple still hasn't said what it means for PCC.

April 22, 2026. Las Vegas. Thomas Kurian stands on the Cloud Next keynote stage, the Apple logo on the screen behind him.

"We're collaborating with Apple," he says, "as their preferred cloud provider, to develop the next generation of Apple Foundation Models based on Gemini technology."

The phrase to notice is not Gemini. It's preferred cloud provider.

For two decades, Apple has policed how partners describe the relationship. Search deals: announced by Apple. Map data sources: footnoted by Apple. Suppliers: no comment, please. Today, the CEO of the partner names Apple from his own stage. Apple's newsroom carries no statement. Federighi has not clarified. Cook has not posted. There is, as of this writing, nothing new on the Apple Security blog.

This is the biggest AI deal in Apple's history. We still don't know what it means.

The Questions Apple Hasn't Answered

Where does Gemini inference run? On the iPhone? On Apple's M5-based Private Cloud Compute servers? On Google's TPUs in a Google data center?

Which Siri queries cross the PCC trust boundary? All of them? Some of them? Only the ones a user explicitly approves?

Do PCC's five guarantees — stateless computation, enforceable guarantees, non-targetability, verifiable transparency, no privileged runtime access — still apply to a query that ends up at Google? Or do they apply only to the part of the request Apple still owns?

Will Google retain prompts? Train on them? Even temporarily?

Apple has answered none of these.

Why the Questions Matter

Two years ago, almost to the week, Craig Federighi unveiled Private Cloud Compute. The pitch was that Apple was extending the privacy of the iPhone — sealed, attested, cryptographically enforced — into the cloud where bigger AI models had to run.

The Apple Security blog laid out the five guarantees. The framing line was the point: "Security and privacy guarantees are strongest when they are entirely technically enforceable."

Not promised. Not audited. Technically enforceable.

Apple opened the silicon to security researchers. Posted bounty terms. Said PCC was the first time the industry-leading privacy of Apple devices had ever been "extended into the cloud."

Today, the cloud is somebody else's. And the company that wrote those words has not yet explained which of the five guarantees survive the handoff.

What We Do Know — And It Doesn't Add Up

In January, Cook told analysts Gemini-Siri would "continue to run on the device and run in Private Cloud Compute."

On Google's most recent earnings call, Sundar Pichai called the company "Apple's preferred cloud provider."

In February, Bloomberg reported the two companies were discussing hosting the model "directly on Google servers running powerful chips known as TPUs."

These are three architecturally different statements. They point at three architecturally different deals.

If Cook is precisely right, Apple hosts a Gemini-derived model on its own M5 silicon inside the existing PCC trust boundary, and the five guarantees apply. If Bloomberg is precisely right, the model runs on Google's hardware in a Google data center, where PCC's attestation does not reach. If Pichai is precisely right, the relationship is something in between — and "preferred cloud provider" is a label that does not yet have a public technical meaning.

Apple has shipped a partner-handoff architecture before. The ChatGPT integration in iOS 18.1 is a three-tier system: on-device, then PCC, then a partner outside the trust boundary, asked-before-sent, IP obscured, contractual no-retention promise. That model exists. Whether Gemini-Siri uses it, replaces it, or layers on top of it is not something anyone outside Apple and Google currently knows.

The Maps Inversion

There is a precedent for this trade, and it points the other way.

In September 2012, Apple shipped its own Maps app to break free of Google. The bridges melted. Hospitals moved to the wrong decade. Cook apologized in writing. Forstall lost his job. The lesson Apple took: escaping a powerful partner is right; doing it before you're ready is fatal.

Fourteen years later, the polarity reverses. Where Maps was an attempt to build out from under Google, Gemini-Siri is an admission that the brain — not the maps, the brain — should be rented from Mountain View. Apple bet on the chip. They were right. They bet on the model. They were not.

You can hear the strain in Cook's January phrasing. Every word in that sentence is true. None of them quite address the question.

The Carve-Out PCC Always Implied

Privacy is a perimeter, and perimeters have doors.

The five guarantees apply inside the perimeter. The ChatGPT carve-out exists outside it. So does the Google search default that pays Apple roughly twenty billion dollars a year and ships every Safari query off-device. So does Apple Maps, which still leans on third-party data feeds. So, soon, will Gemini-Siri.

What's enforced cryptographically: the silicon Apple controls. What's enforced contractually: everything else. The wall is real. It is also shorter than the marketing implied.

What to Watch For

The disclosure language will arrive between now and WWDC. Apple is good at the language. We will hear about IP obscuring, no-retention contracts, opt-ins per-feature or per-session or per-something. Whatever the words turn out to be, they will be enforced by signature, not by silicon.

The point is not that the deal is bad. Gemini is a better model than anything Apple has shipped, and the bargain may well be fair. The point is that technically enforceable meant something specific in 2024, and nobody at Apple has yet said what it means in 2026.

Federighi, two years ago: industry-leading.

Kurian, today: preferred cloud provider.

Both phrases are accurate. Only one of them came from Apple. Only one of them came with an attestation.

From the Library

This newsletter draws on a growing library of books, oral histories, and primary sources about Apple's history. Today's issue drew on:

Apple in China by Patrick McGee — for the long arc of Apple's "we'll do it ourselves" reflex and how it bends under partner leverage.

Becoming Steve Jobs by Brent Schlender and Rick Tetzeli — for the 1997 Microsoft pact, the canonical case of Apple swallowing a rival's leverage in public.

Inside Apple by Adam Lashinsky — for the Maps debacle and Forstall's exit.

Four A.M. Cupertino. The phone is glowing.

He has been doing this for fifteen years. Between four and five, before the gym, before the day starts — he reads customer emails. Hundreds of them. Sometimes seven hundred. Sometimes eight. He answers some.

Today, the email he is writing is not to a customer. It is to the company. "It has been the greatest privilege of my life to be the CEO of Apple." On September 1, John Ternus takes over.

Five weeks ago, on Good Morning America, he said he could not imagine life without Apple after twenty-eight years. He was already writing the letter.

The pattern

April 1998. He had been employed for one week. Friends in Houston had told him not to take the job — Compaq was thriving; Apple was drowning. He came anyway.

His first weekly ops review was scheduled for ninety minutes. It ran thirteen hours.

A manager reported a number: two hundred thousand. Cook cut in.

"And fifty?"

He had it memorized. He had everything memorized. By that September he had cut Apple's inventory from thirty days to six. By the next September, two. Suppliers from a hundred to twenty-four. Warehouses from nineteen to nine. Inventory is fundamentally evil, he told them. Like a dairy. Products should be sold fresh.

This is the man who, on hearing about a supplier crisis, said someone should be in China driving this — and then, thirty minutes later, looked at the operations executive still sitting across from him.

"Why are you still here?"

The executive stood up, drove to SFO, and flew to China. He became one of Cook's top deputies.

The pattern that wasn't on the spreadsheet

Early 2009. Steve Jobs was bedridden, gaunt, swollen with ascites. The cancer was eating him.

Cook went quietly to a doctor. He researched living-donor liver transplants. He found out his blood type matched. He got himself tested.

Then he went to Jobs's house in Palo Alto and offered part of his own liver.

Jobs cut him off mid-sentence. "No. I'll never let you do that."

It is the scene that ruins the cool-operator caricature. The man who would later be called a spreadsheet had quietly, secretly, gotten himself onto a transplant list for his boss.

There is a related story from 1998. Jobs, having decided Cook was indispensable, called Cook's mother in Robertsdale, Alabama, to ask her to convince her son to start a family. Mentor as meddler. The friendship was already that close.

The things he kept

He almost didn't take the Apple job because California is twice as far from Auburn football as Texas. His Palo Alto home is reportedly a shrine to the Tigers — a J&M outpost in the South Bay. He played trombone at Robertsdale High. Teachers remembered him as long-limbed and meticulous.

He works out at an off-campus gym, specifically so he won't run into Apple employees. He collects no cars, no art, no watches. Most evenings he is alone.

In October 2014, three years into his tenure, he wrote a 1,200-word essay in Bloomberg Businessweek. "I'm proud to be gay, and I consider being gay among the greatest gifts God has given me." He was the first Fortune 500 CEO to come out. He framed it as service: "a brick in the path toward justice."

It was not a product launch. It was the moment he made Apple — the most secretive consumer company on earth — brave on his behalf. The company did not flinch. He had earned the right to ask.

The handoff

August 11, 2011. A Sunday. Jobs called Cook to his Palo Alto home. "Now." He had decided. Cook would be CEO.

Cook asked the question every successor asks. What would you do? How would you decide?

Jobs's answer is the most important sentence in modern Apple history.

"Don't ask what I would do. Just do what's right."

Today's letter to the company — written between four and five A.M., the same hour as fifteen years of customer mail — does not quote that line. It doesn't have to. The whole letter is a version of it, addressed to John Ternus.

Cook will stay on as Executive Chairman, focused on what the announcement carefully calls "engagement with global policymakers." Translation: he keeps the China file, the Trump file, the EU file. The political-fixer machine is the one piece of him he is not handing over.

The rest is already gone. Two-day inventory. The 4 A.M. inbox. The off-campus gym. The audit-class precision. The 2,300 supplier factories. The four trillion dollars.

It will all run tomorrow without him. That was always the point.

From the Library

This newsletter draws on books about Apple's history. Today's issue featured stories from:

Inside Apple by Adam Lashinsky — the 13-hour ops review and "And fifty?", the SFO sprint.

Apple in China by Patrick McGee — Cook's 1998 supply chain rebuild, "inventory is fundamentally evil."

Becoming Steve Jobs by Brent Schlender — the liver offer, the mother phone call, the "Now" handoff.

Steve Jobs by Walter Isaacson — Jobs's "Don't ask what I would do" advice to Cook.

In September 1997, Apple ran the first version of Think Different. Black-and-white portraits — Einstein, Dylan, King, Picasso, Lennon and Ono. Here's to the crazy ones. The misfits. The rebels. The troublemakers.

In Asia, Apple cut the Dalai Lama. For fear of offending China.

The company that had just told the world to honor rebels had, on its very first day back, already decided which ones to edit out.

Patrick McGee, in Apple in China, calls it Apple's "first known kowtow." It turned out to be the template.

The ruling

On Friday, April 17, a federal judge in Chicago granted a preliminary injunction against the Trump administration. Judge Jorge Alonso ruled that the Department of Justice and Department of Homeland Security likely violated the First Amendment when they pressured Apple, Google, and Meta into removing apps and Facebook groups used to track ICE agents in public.

Alonso quoted the Supreme Court's 2024 ruling in NRA v. Vullo: government officials cannot "coerce private parties in order to punish or suppress views that the government disfavors." He called the administration's public statements "thinly veiled threats."

FIRE, the speech group that backed the suit, put it plainer. The court, they said, had blocked the government from "strong-arming Apple and Facebook."

The court ruled for the speakers. Apple was not a party.

The app

On October 3, 2025, Apple removed ICEBlock — a tool with roughly a million users that let people report observable locations of ICE agents in public. The removal came the same week Attorney General Pam Bondi told reporters: "We reached out to Apple today demanding they remove the ICEBlock app from their App Store." Apple complied within a day.

Apple's explanation: the app "violated the app store guidelines." Apple added that it had "received information from law enforcement suggesting the app's purpose is to provide location information about law enforcement officers that can be used to harm such officers."

That is almost exactly what the Chinese state press said about HKmap.live, the Hong Kong mapping tool Apple pulled in October 2019 — that it was a platform to "harm Hong Kong police."

Same frame. Different uniform.

The pattern

Apple's compliance playbook is twenty-nine years old and remarkably consistent.

1997. Dalai Lama cut from Think Different in Asia.

2008. iTunes quietly blocked a Songs for Tibet promotion during the Beijing Olympics after Chinese users complained.

2017. 674 VPN apps deleted from the China App Store.

2019. HKmap.live pulled days after Chinese state-media criticism.

2022. AirDrop range restricted in China after protesters used it during the White Paper demonstrations.

2025. ICEBlock removed within a day of the U.S. attorney general's demand.

Five of these happened in China. One happened in the United States. Apple treated them the same way.

The pattern was never about geography. It was about posture. When a sovereign pushes, the platform removes, explains, and moves on.

The letter that isn't here

There is one moment in Apple's history that makes all of this look strange. On February 16, 2016, Tim Cook posted "A Message to Our Customers" on apple.com. The FBI was asking Apple to build software that would unlock a dead terrorist's iPhone. Cook said no — in public, under his name, at length. Forty days later, the Justice Department dropped the case.

That letter is why people still associate Apple with principle. It made the cover of magazines. It is cited in law-school casebooks. It is, for a generation of users, the thing Apple is for.

It is worth noticing, though, what Cook was defending. He was defending encryption — the unlocked phone as a demand the state had no right to make. That defense turned into a brand. Since 2016, Apple has spent a decade making privacy the centerpiece of its marketing: the "Privacy. That's iPhone." billboards, App Tracking Transparency, Private Cloud Compute, a dedicated page on apple.com with its own animated film. Privacy is now a product Apple sells. It is printed on the box.

The First Amendment is not. There has never been an "Apple defends speech" campaign, because Apple does not. There is no Cook letter on behalf of an app developer. There is no billboard that says The App Store protects rebels. When the FBI came for the phone, Apple drew a line. When the DOJ came for the app, Apple did not.

The difference is not accidental. Encryption is a wall the company already owns — defending it costs nothing after the architecture is built. Speech is messier. It requires sometimes hosting things sovereigns don't like. That was never part of the brand.

In October 2025, there was no Cook letter. The DOJ demanded. Apple removed. Bondi thanked Apple on camera. Apple issued a sentence about guidelines.

The 2016 letter was not the beginning of a tradition. It was the beginning of half a tradition — the half Apple could sell.

What the court actually did

Here is what the injunction does, and does not, do.

It does not order Apple to restore ICEBlock. Apple is not a defendant. The court has no jurisdiction over Apple's App Store decisions.

What it does is order the government to stop making the threats that produced those decisions. It protects the speakers — the app developers, the Facebook group administrator, the users — from the coercion that Apple complied with.

The protection did not come from Apple. It came from the people Apple's compliance had silenced. They sued. They won. And the ruling now shields them from the next such demand — whether Apple fights it or not.

This is the structural fact worth sitting with. In 2016, Apple was the defender. In 2025, Apple was the infrastructure the court had to reach around.

Coda

Twenty-nine years ago, Apple edited out the Dalai Lama and built a company on "Here's to the crazy ones." Most of the years since, those two sentences have been held in the same mouth without visible discomfort. The company celebrates rebels in English and removes them in every other language the local censor requires.

Last week, a federal judge in Chicago did for Apple what Apple used to do for itself: write a letter on principle.

The government didn't have to go around Apple.

The court did.

Also This Week

Apple is sending ~200 Siri engineers to a multi-week AI coding bootcamp. Seven weeks before WWDC, the team that builds Siri is being retrained on how to build with AI. Marketing has already teased the redesign. Engineering is in remedial training. WWDC is June 8.

India's CCI advanced to a final hearing on what could be Apple's largest antitrust fine ever — up to $38B. Apple is refusing to hand over global financial data. In 2013, Cook's response to a Chinese regulator was a Mandarin apology letter. In 2026, the response to an Indian one is silence. Different playbook.

India manufactured $2.5 billion in components that shipped back to China. India isn't replacing China. It's feeding it. The supply chain isn't pivoting — it's fracturing into bilateral specialty flows.

From the Library

This newsletter draws on the Apple Books archive.

Apple in China by Patrick McGee — the twenty-nine-year arc from the 1997 Dalai Lama edit to HKmap.live, the VPN purge, and the AirDrop restriction. McGee names the pattern.

Steve Jobs by Walter Isaacson — the shape of Apple's principled posture before principle became a brand pillar.

In the fall of 2006, Eddie Cue and Steve Jobs sat at a conference table inside the Four Seasons in Palo Alto, across from two AT&T executives. They had four hours.

Weeks earlier, they had pitched the iPhone to Verizon. Apple wanted control of the interface, the price, a cut of data fees. Verizon refused. Wireless was, an executive told them, our playground.

AT&T was mocked for its coverage and hungry for a halo product. In four hours, they accepted the iPhone almost sight unseen. Apple controlled the UI. Apple set the price. Apple got up to ten percent of monthly data revenue.

Pick the partner who needs you most. It worked for twenty years.

The news

Last week, Amazon signed an agreement to acquire Globalstar for $11.57 billion. Separately, Apple signed a long-term deal routing iPhone satellite services — Emergency SOS, Find My, Messages — through Amazon's Project Kuiper.

The first reading is that Amazon bought Apple's satellite partner. The second reading, which matters more, is that Amazon rescued it.

The pattern

Look at twenty years of Apple partner decisions.

AT&T over Verizon, 2007. Works. Carriers cannot consolidate in a weekend.

TomTom over Google Maps, 2012. Rocky. Works eventually. Mapping is software.

TSMC over Intel, 2020. Works. Taiwan is geopolitically insulated.

Globalstar over Starlink, 2022. Works, briefly.

The playbook assumed one thing and never stated it: the small partner had to stay small and stay viable. AT&T and TSMC are massive companies for whom Apple is one of many customers. They were small only in the sense that Apple had leverage. They could survive without Apple.

Globalstar could not.

The captive partner

Globalstar was not a hungry partner. It was a captive one. Twenty-four aging satellites. Years of losses. A new constellation that existed only because Apple fronted $1.5 billion to build it. Eighty-five percent of network capacity dedicated to a single customer.

When Apple picks a partner like that, there are three ways the relationship ends. Apple buys the partner — as with PA Semi, Authentec, Shazam. Someone else buys it. Or it fails.

Apple didn't buy Globalstar. Satellite infrastructure is capital-intensive and sits outside Apple's competence. So option one was off the table. That left two and three. Amazon chose two.

What the deal actually is

This is closer to an arranged marriage than an acquisition. Amazon needs a functioning satellite business to make Kuiper pencil out. Apple needs a working constellation to keep Emergency SOS alive. The $11.57 billion solved Globalstar's capital problem. The Apple contract gives Amazon an anchor customer for a rollout that is years behind Starlink.

Both parties had other options in theory. Neither had any in practice.

For Apple, this is the first time in twenty years the playbook has produced a partner it neither controls nor can credibly leave. AWS runs Apple services. Amazon is one of the largest resellers of Apple hardware. Alexa, Audible, Kindle, Prime Video all compete with Apple businesses. And now, Amazon owns the pipe that carries the most intimate signals Apple has — your location when you're lost, your distress when you can't call 911.

It's not 1997. Apple isn't insolvent and Jeff Bezos isn't Bill Gates on a giant screen. But the shape is familiar. A rival has become indispensable. The stage doesn't say it. The filings do.

What comes next

Near-term: nothing visible. The FCC still has to approve the acquisition. iPhone satellite service keeps working. Nine to twelve months before anything shifts.

Medium-term: Amazon paid a premium for Globalstar because the Apple contract, the 85% capacity guarantee, and the specialized satellites now belong inside Amazon's long-term satellite strategy. Enterprise connectivity. AWS ground stations. Services Amazon hasn't announced yet. Apple's leverage at contract renewal will look different than it did in 2022.

Coda

In 2007, Apple walked away from Verizon's playground. In 2022, Apple picked a partner small enough to control. In 2026, Apple discovered there is a category of partner the playbook never accounted for: the one too small to play alone.

Apple picked the hungry partner. It turned out hungry wasn't the problem.

Starving was.

Also This Week

Perplexity and OpenAI launched rival Mac agents on the same day. Two always-on AI assistants with system access, shipped within hours. The agentic layer is being contested on macOS — and not by Apple. WWDC is June 8.

The $599 MacBook Neo sold out through April. Apple's affordable laptop is validating the "iPhone SE for Mac" thesis. Price-ladder discipline still works.

From the Library

This newsletter draws on the Apple Books archive.

Becoming Steve Jobs by Brent Schlender — the Four Seasons negotiation and the small-partner playbook.

Steve Jobs by Walter Isaacson — the 1997 Microsoft lifeline and the shape of rival-as-indispensable-partner.

Apple in China by Patrick McGee — the Intel-to-TSMC pivot and Apple's geopolitically insulated silicon bet.

In 1999, Wayne Miller rode a golf cart across a stretch of dirt north of Shenzhen while Terry Gou pointed at the air.

Gou was describing buildings — dormitories, canteens, whole city blocks — that did not yet exist on a piece of land called Longhua. Miller, an Apple operations engineer, saw only empty fields. But when workers passed "Uncle Terry," they straightened up the way farmhands do when a governor walks by. The campus was small. The power structure was not.

Twenty-seven years later, the fields have become a city, the city has become a country, and the country has become a piece of legislation.

The Law

On April 7, 2026, China's State Council published eighteen articles that consolidate Beijing's export controls, anti-sanctions powers, and industrial policy into a single national-security framework. The Regulations on Industrial and Supply Chain Security took effect on publication. There was no transition period.

Article 15 authorizes investigations when foreign entities "interrupt normal transactions," impose "discriminatory measures," or cause "substantial harm" to Chinese supply-chain security. Article 16 requires China-based personnel to "strictly execute" any resulting countermeasures, creating a direct conflict with foreign compliance obligations.

The Morgan Lewis readout was careful enough to quote the triggering language verbatim, because the language is unusually candid for Chinese regulation. The law specifically flags "terminating supply to Chinese customers" and "exiting China-related supply chains" as prohibited conduct — particularly when those decisions stem from foreign regulatory pressure. The document names its own policy objective in plain view: "increasing global supply chain decoupling pressures."

In other words, diversification is the offense. Intent is not required.

The Garden and the Walls

To understand why this law is about Apple without naming Apple, you have to understand what Apple built.

Patrick McGee's Apple in China documents the scale plainly, because the plain numbers are themselves the argument. Apple trained roughly twenty-eight million workers in China beginning in 2008. Its internal documents put the annual investment figure at $55 billion by 2015, before even counting components. Apple-owned machinery inside Chinese factories reached $7.3 billion by 2012, wearing no Apple logo. The campuses had no signs. The engineers were deliberately junior. Foxconn handled every awkward conversation with every local official. It was, as McGee calls it, an invisible empire.

The empire served a specific purpose: it made China the only place iPhones could be built, which made iPhones profitable, which funded more training, more tooling, more investment. For a decade and a half, the flywheel spun. China got the industry. Apple got the margins. Everyone agreed not to talk about whose balance sheet was whose.

Then Beijing wrote it into law that the flywheel is now spinning in one direction only.

The Diploma

Here is the part almost nobody is covering. Apple's own capex data says China no longer needs the tutoring.

Apple's "long-lived assets" in China peaked at $13.3 billion in 2018. By the early 2020s they had fallen nearly two-thirds, to roughly $4.8 billion — the lowest level since 2011. The money didn't disappear; it changed hands. In July 2020, Luxshare bought two Wistron iPhone assembly plants for $472 million. A year later, BYD Electronic paid $2.2 billion for Jabil's Chengdu operations. Grace Wang, a former Foxconn line worker from 1988, had turned Luxshare from an $81 million connector shop into a $38 billion company by imitating Foxconn's playbook and improving it. Xi's Made in China 2025 was explicit on the question of method: the fastest path to "mastering core technologies" was working "intimately" with Apple.

The Chinese supply chain was not bought off. It was graduated.

Which reframes the April 7 regulations entirely. The law does not exist because China is trying to stop Apple from leaving. The law exists because China has finally absorbed enough of what Apple was teaching that it can afford to make Apple sit still for the final exam.

There is a harder reading of this — the one you only see if you take China's point of view seriously for a moment. For twenty-five years, Beijing tolerated the indignities of hosting Apple because Apple was the single most effective industrial-policy subsidy any foreign company has ever delivered. But there is no obvious next Apple. There will not be a second firm that arrives with $55 billion a year and the patience to train twenty-eight million workers. The Western firms Beijing might once have courted are — at Washington's insistence — moving the other way.

So either China has decided it no longer needs another Apple, or China has decided that if no second teacher is coming, the first teacher should not be allowed to leave the room.

The law is consistent with both readings. That is what makes it unnerving.

The Move, From the Other Side

There is a scene in the database that haunts this one.

In early 1999, LG was Apple's largest iMac assembler. It had the build documentation, the three-continent footprint, the long relationship, and the swagger that came with all of it. When an LG executive started talking about leverage, Steve Jobs and Tim Cook made a phone call. Terry Gou answered it with a single sentence: I can fix this. By March, Apple had redesigned the iMac end-to-end — new enclosure, new internals, no fan, roughly half the bill of materials — and Foxconn had delivered tooling in twenty-five days that would have taken LG twelve weeks. Gou called it China speed. It was, more precisely, the speed at which the student you trained becomes the rival you can't undercut.

Apple learned that move as an apprentice. It used the move against LG as a master. It spent the next two decades teaching the move to China. And now the move is being used against Apple, by a student who has the build documentation, the three-continent footprint, and the long relationship — and who has just written a law that says the documentation stays.

I can fix this is a phrase that only works in one direction.

The Trap

For twenty-five years, Apple's China playbook had an escape valve. When Beijing got angry, Tim Cook wrote a Mandarin apology letter, upgraded a warranty policy, posed for a photo at a Luxshare plant, and the pressure released. The 2013 CCTV apology is the canonical version: eighteen days from attack to contrition, and a replacement policy sweeter than the one Americans got. You could apologize your way out because the dispute was about image, not infrastructure.

The April 7 regulations remove the valve.

There is no Mandarin letter that settles a capex dispute. The concession Article 15 asks for is not a policy tweak — it is the reversal of the India build-out, which Washington has made a precondition for Apple's continued tariff exemptions. Freeze the India shift and the American politics collapse. Continue it and Chinese investigators can begin work without warning, without evidence of intent, and without a transition period. Apple's supply chain now has to clear two legal regimes that are designed to contradict each other, and every marginal decision — where to qualify a new vendor, which engineer flies where, which assembly line goes live in Tamil Nadu — now creates liability under one or the other.

Article 16 is the twist of the knife. It requires China-based personnel to "strictly execute" any countermeasure Beijing imposes. Apple has roughly one thousand engineers in China. In a serious enforcement scenario, the legal obligation of Apple's own employees would run against the operational directives of their American employer. The org chart now has a fault line through it, and the fault line was drawn by a government.

This is the dilemma Cook does not have a playbook for. The 2013 version was a question of what Apple would say. The 2026 version is a question of what Apple is allowed to do, and neither answer is free. The only move that works is the move that does not exist: a way to be in both countries at full scale at the same time — which is the exact arrangement Beijing just banned, and the exact arrangement Washington just mandated.

Back in the golf cart at Longhua, Terry Gou was right about the buildings. What he did not say — and did not need to — was whose city it would be.

From the Library

This newsletter draws on the Apple Books corpus. Today's issue featured stories from:

Apple in China by Patrick McGee — the invisible empire, the 28-million-worker figure, Longhua's construction, Luxshare and BYD's rise, the capex retreat, Terry Gou's 1999 "I can fix this" call, and the 2013 CCTV apology.

The private dining room at Maddalena's in Palo Alto was supposed to be neutral ground. HR had booked it to broker peace between Apple's CEO and the man who ran its soul — the debonair Frenchman with the diamond earring whom John Sculley had chosen, five years earlier, because he was "the closest thing to a Jobs."

Dessert ended the diplomacy. Jean-Louis Gassée looked across the table and told Sculley that Apple's engineers didn't respect him. That he himself didn't respect him. That Apple had the wrong man at the top.

Two days later, Gassée was fired.

About 150 engineers marched outside De Anza 7, hoisting hand-lettered signs reading "JLG 4 CEO." A funeral wreath of black flowers arrived at his office door. Gassée remembers not fury but relief. He walked away with $1.7 million in severance and the seed money for Be Inc.

Apple kept everything he'd built.

The Departure

This week, John Giannandrea officially left Apple. The timing was precise: April 15, the day his next stock award vests. He'd been on the payroll since December in an advisory role — "resting and vesting," in Silicon Valley parlance — after Tim Cook announced his retirement with a press release that thanked him for "building and advancing our AI work."

The language was warm. The reality was not. Seven months earlier, Cook had stripped Siri from Giannandrea's control after losing confidence in his ability to ship product. Mike Rockwell took Siri. Sabih Khan took AI infrastructure. Craig Federighi got the rest. By the time the retirement announcement came, Giannandrea was a senior vice president of nothing.

His replacement, Amar Subramanya, had been at Apple for all of four weeks — poached from Microsoft, where he'd spent five months, after sixteen years at Google leading the engineering team behind Gemini. An outsider replacing an outsider.

Within weeks, at least five AI researchers followed him out the door. Stuart Bowers, a senior Siri executive, left for Google DeepMind. Others went to Meta's superintelligence lab. The immune system had activated — but this time, the antibodies were attacking the host.

The Pattern

Apple has been running this playbook for four decades.

In 1983, Steve Jobs flew to New York and walked John Sculley through Central Park, the Met, and up to a San Remo terrace where the Pepsi president clung to the wall, afraid of heights. As Walter Isaacson recounts in Steve Jobs, Jobs lowered the blade: "Do you want to spend the rest of your life selling sugared water, or do you want to come with me and change the world?" Sculley came. Apple got corporate credibility, a grown-up face for Wall Street, ten years of stability. Then the cupboard was bare, and Apple pushed him aside.

In 1996, Gil Amelio negotiated what one witness called "outrageous terms" — chairman plus CEO, $2.5 million minimum, a $5 million loan, 200,000 shares a year — to save a company bleeding cash. As he recounts in On the Firing Line, he stabilized the finances and acquired NeXT. Eighteen months later, on July Fourth, Ed Woolard called him at Lake Tahoe from Wimbledon: the board had been meeting for 36 hours without telling him. Jobs had already been consulted.

In 2001, Jon Rubinstein phoned Tony Fadell on a Vail ski slope and offered an eight-week contract to see if a pocket music player was feasible. Fadell shipped the iPod. Then the iPhone. Then he slammed his badge on the table shouting "Wrong, wrong, wrong!" when Jobs leaned toward Intel — and eventually walked out. As Adam Lashinsky documented in Inside Apple, when Rubinstein himself later took a job at Palm, Jobs called back "about four seconds later," furious. The capability stayed. The person left.

In 2012, Scott Forstall — who'd quietly built the App Store against orders, who inspected icons with a jeweler's loupe — was gone within weeks of Apple Maps melting bridges into rivers. As Tripp Mickle wrote in After Steve, Cook apologized publicly. Forstall refused to sign the letter. He was out. Apple kept the ecosystem.

The pattern has six phases: seduce with mission. Empower inside a walled kingdom. Extract peak value. Watch the cultural antibodies activate. Execute a swift surgical removal. Absorb the value, discard the person.

The Exception

Every previous outsider left behind a capability Apple could own. Sculley left corporate structure. Gassée left engineering culture. Amelio left NeXT's operating system. Fadell left the iPod and ARM chip strategy. Forstall left iOS.

Giannandrea left... what, exactly?

Siri still can't reliably set a timer and send a text in the same breath. The overhauled version, promised at WWDC 2024, slipped to 2026 after hallucinating too often. Apple pulled a Bella Ramsey ad campaign for a product that didn't exist yet. And in January, Apple signed a deal to power its assistant with a custom Google Gemini model — effectively conceding that the foundation model isn't Apple's to build.

Mark Gurman put it plainly in Bloomberg's Power On: "The top of Apple is run like a small family business with few decision-makers. And if you're not in the inner circle — which is nearly impossible to crack — you're simply not empowered enough to drive real change."

The immune system works beautifully when the outsider's job is to bootstrap a capability insiders can then own — chips, supply chains, retail, operating systems. You build it once, then iterate. It fails when the capability keeps evolving faster than the institution can absorb it. AI isn't a chip architecture. It isn't a supply chain. It's a field where the frontier moves weekly, where the best researchers have standing offers from DeepMind, and where Apple's secrecy culture — the very thing that powers the immune system — repels the open-research talent that builds foundation models.

The Second Outsider

Subramanya's appointment is the tell. Apple's immune system rejected Giannandrea, but it didn't promote an insider. It hired another outsider — one who'd spent four months at Microsoft and sixteen years at Google. He reports to Federighi, the lifer who once viewed AI as "unpredictable and difficult to control" before ChatGPT changed his mind.

The question isn't whether Subramanya can ship. It's whether Apple has learned anything from 40 years of this cycle. Every outsider who failed at Apple failed for the same reason: the institution was designed to absorb their expertise and move on, not to let them reshape the institution itself.

Gassée understood this. At Maddalena's, he didn't just criticize Sculley. He named the deeper problem. After his meetings, Sculley's own lieutenants would huddle to decide "what part of John's plan we'll implement." The outsider proposes. The institution disposes.

Thirty-six years later, the private dining room is different. The funeral wreath is digital. But the mechanism is the same.

No one marched for Giannandrea. No signs appeared outside De Anza 7. His exit was so quiet the only marker was a stock grant, vesting on tax day.

Also This Week

Apple Testing Four Smart Glasses Designs: Premium acetate frames, no display, camera-only — Apple's post-Vision Pro pivot is fashion glasses, not spatial computing. Custom N401 chip, 2027 launch target. A direct shot at Meta Ray-Ban.

RIP iWork (2005–2026): Apple quietly stopped providing downloads for the original Pages, Keynote, and Numbers. Only "Creator Studio" versions remain. Classic Apple: kill your own product, force everyone forward.

From the Library

This newsletter draws on stories from 4 books about Apple's history:

Steve Jobs by Walter Isaacson — The original "sugar water" seduction on a San Remo terrace

On the Firing Line by Gil Amelio — The Lake Tahoe phone call, told by the man who received it

Inside Apple by Adam Lashinsky — Rubinstein's "loyalty test" and the four-second callback

After Steve by Tripp Mickle — Forstall's jeweler's loupe and the Maps catastrophe