The Last Engineer
The appointment everyone reads as continuity is the one the next paradigm reads as exit.
On September 1, 2026, John Ternus will succeed Tim Cook as Apple’s next CEO. He is a mechanical engineer. He joined Apple’s product design team in 2001, became a vice president of Hardware Engineering in 2013, and joined the executive team as senior vice president of Hardware Engineering in 2021. He helped translate Apple silicon into the consumer hardware line, above all through the Mac’s post-Intel resurgence. The board’s public statement emphasized his engineering depth, his fixation on product quality, his discipline with the supply chain.
All of this is true. Together these truths compose a judgment more important than the appointment itself: Apple’s board has decided that the next decade will be run by a hardware fundamentalist.
The decision is not strange in itself. What is strange is when it was made.
I. The Historical Function of a Personnel File
The twentieth century produced a class of appointments that are studied only in retrospect.
Thomas Watson Jr. inheriting IBM in 1956. The last executive reshuffle under Stuart Saunders at Penn Central before the 1971 bankruptcy filing. Carly Fiorina taking HP in 1999. Scott McNealy refusing to hand off Sun Microsystems at its peak in the late 1990s.
At the time, each was treated as ordinary corporate governance. In retrospect each is understood as something else: a paradigm, shortly before it lost its dominant position, appointing the person best suited to manage its dusk.
There is a stable pattern to these appointments. The appointee possesses every capability the past decade required, and is for that exact reason deficient in the capability the next decade demands. The board believes it is choosing continuity. It is in fact choosing ossification. The appointee’s strength converts, in the new paradigm, into weakness. The process usually takes five to eight years to be publicly acknowledged by the capital markets.
Ternus fits the full pattern.
II. The Physics of the Appointee
Ternus’s engineering philosophy has a clear, verifiable record. The hardware teams he led restored the scissor-switch keyboard, killed the Touch Bar, vetoed a smart-speaker prototype with a screen, and drove the migration to in-house silicon. The internal logic of every one of these decisions was the same: physical certainty over aesthetic risk, engineering redundancy over feature expansion, on-device cost control over category experimentation.
Between 2014 and 2024 this was a nearly unassailable philosophy. It rescued Apple from the late-Ive formalism, consolidated the efficiency lead of the M-series, and held the iPhone’s hardware premium far above the industry average.
The same philosophy, applied to the central strategic problem of 2026, produces a systematic misjudgment.
The current AI paradigm takes as its basic competitive form: ship the broken software first to collect data, then iterate the model backward from the data at scale. OpenAI, Anthropic, and Google DeepMind all release in a manner that a zero-defect mindset could never tolerate — absorbing hallucination, error, ugly edge cases — trading deployment scale for data and data for model-iteration speed.
The whole of Ternus’s professional formation — a quarter century inside Apple’s hardware organization — has been devoted to resisting exactly this way of working.
III. The Sculley Event, Recurring
This appointment is structurally identical to the 1985 event in which Apple’s board chose Sculley and exiled Jobs.
What “Sculley-ization” means is routinely misread. The board’s error was not “expelling a genius.” It was answering a paradigm shift in progress with the solution to the previous paradigm’s problem. In 1985 Apple did not need better operational management; it needed the continuation of product vision. The board’s instinct was operational discipline. Twelve years later Apple was near bankruptcy, having paid the cost of the delayed penalty.
The 2026 board is structurally identical to the 1985 one. As of April 2026, its core members include a former CEO of Johnson & Johnson, a former CEO of Avon, a former CEO of Northrop Grumman, a co-founder of BlackRock, the former president of an aerospace corporation, the former president of an education foundation, and the founder of Calico. This is a board of compliance, dividends, political lobbying, and brand defense. Not one of them has, in the past decade, been deeply engaged in the core engineering of large language models, reinforcement learning, or neural-network architecture.
This is not a criticism. It is the observation of a fact: when Apple needs, at the highest level of governance, to judge whether to commit tens of billions of dollars to compute infrastructure that may produce no revenue before 2030, the people making the judgment are professionally incapable of making it. Their fiduciary duty, their professional training, and their personal incentives all point the other way — protect current profit, sustain the dividend, defend against antitrust.
The board’s instinct is engineering discipline, exactly as the 1985 instinct was operational discipline.
But on one level this event is colder than Sculley-ization.
Sculley-ization was reversible. What Apple excised in 1985 was the product-vision organ. Jobs was expelled, but he remained the same species of person — he could return twelve years later and repair the excision. When the board called him back in 1996, he had accumulated, at NeXT and Pixar simultaneously, both product vision and corporate management, and returned as an upgraded version of himself.
Ternus-ization is irreversible. What Apple failed to develop, by 2026, is the paradigm-adaptation organ for compute. That organ never grew inside Apple at all — through the Cook era Apple never built real muscle in AI fundamental research, never built a cloud-native compute foundation, never built an engineering culture oriented toward iterating on broken software.
This is not “the board excised an organ that existed.” It is “the board confirmed that an organ which never developed will now never develop.”
If, in 2034, the board discovers Apple has missed the AI era, whom does it call back? There is no Jobs-equivalent waiting outside, having accumulated both AI and hardware capability at some NeXT and some Pixar. The defining figures of the AI paradigm — Sam Altman, Dario Amodei, Demis Hassabis — already command entities harder to dislodge than Apple’s market capitalization at the time. They will not return to Apple. Their existence is itself the cause of Apple’s decline.
More to the point: when Jobs returned, the problem Apple needed to solve — redefining the human-computer interface of personal computing — was still inside Apple’s reach. The problem Apple needs to solve in 2034 — redefining compute distribution in the age of AI agents — is no longer inside Apple’s reach.
Sculley-ization was a delayed resurrection. Ternus-ization is a confirmed marginalization.
IV. The Physics of the Paradigm
The four trillion dollars of market capitalization the iPhone created were not the personal achievement of Cook or Jobs. They were the physical byproduct of a specific historical window: from 2007 to 2024, the entire value growth of the internet economy flowed through one chokepoint — the personal, always-carried gateway to compute. As mobile compute penetration moved from zero toward above ninety percent, whoever held the gateway was automatically granted the era’s largest right of taxation.
That window is now closing — but not in the way most people imagine.
One fashionable account holds that compute will naturally detach from the screen, evolving into glasses, earbuds, wearables, ambient sensors, and that the iPhone is therefore being slowly diluted by time. This account is sound in physics and wrong in timetable. The Humane AI Pin is dead. The Rabbit R1 never became a category. Meta’s Ray-Ban glasses are the exception that proves the timetable — a genuine hit, the dominant share of a real smart-glasses market, and still an accessory layer rather than a phone-replacing compute gateway. Even the most successful AI wearable to date has not retired the chokepoint.
The threat does not come from a startup’s small hardware. The threat comes from the paradigm-definer stepping directly onto the field.
In April 2026, Ming-Chi Kuo reported that OpenAI is developing its own phone jointly with MediaTek and Qualcomm, with Luxshare holding the exclusive system-level manufacturing contract, targeting mass production in 2028. This is not “another AI device.” It is the AI paradigm-definer entering the phone category in a frontal hardware war. That OpenAI chose this path — rather than waiting for the “compute detaches from the screen” trend to mature — is itself a judgment: before 2030, whoever holds the hardware chokepoint holds the largest right of taxation in the AI economy.
That judgment is identical to the instinctive judgment of Apple’s current board. The difference is this: OpenAI has decided to seize the chokepoint, and Apple has decided to hold it. The judgments are the same. The resources are not. The resolve is not. The culture is not.
Apple will not collapse in any single year. Apple will, over a decade, decline from definer of the era to one of the era’s physical shells. This is not the Nokia trajectory of 2007 to 2013. It is closer to the railroad trajectory of 1920 to 1970 — long, concealable, repeatedly deceiving the capital markets with earnings rebounds.
V. The Remaining Moat, and Its Failure Condition
Two countervailing forces deserve honest treatment.
Over the past fifteen years Apple quietly built a second layer of infrastructure: Secure Enclave, biometrics, Wallet, Passkeys. This is a hardware-rooted personal-sovereignty authentication system. When AI agents become autonomous economic actors — placing orders, signing contracts, moving money on a user’s behalf — they will need a hardware-grade identity anchor that law and banks recognize, in order to initiate authorization. Ambient compute can be formless; the physical device that authorizes the money must be tangible, certifiable, traceable by regulators.
This is Apple’s real moat in the AI era — deeper than “make a better iPhone.”
But it will be diluted by two parallel processes.
The first is protocol-layer dilution. The FIDO Alliance, the W3C, and national digital-identity frameworks — the EU’s eIDAS 2.0, India’s Aadhaar, China’s eID — will progressively standardize hardware-root authentication into a multi-vendor protocol layer. Apple will be permitted to participate, but will lose exclusivity. Its Secure Enclave will be demoted from the de facto global identity infrastructure to one of many compatible devices.
The second is sovereign expropriation. When AI agents become significant economic actors, governments will require that agent authorization admit state-level certification. This is not conspiracy; it is the necessary extension of KYC and AML regulation — you cannot permit an AI to move a billion dollars a day on a privately issued hardware key with no state oversight. Once state-level certification is mandated, Apple’s Wallet and Passkeys are bypassed the way Visa is bypassed by a central-bank digital currency.
There is no historical precedent for a dedicated identity-verification device surviving as an independent giant. The physical key is attached to the lock system. The signature and seal are attached to the paper-contract system. The bank card is attached to the clearing network. The iPhone is attached to the personal compute gateway. Identity verification has never existed as a standalone product — it is always the parasitic layer of some dominant functional paradigm. When the dominant paradigm changes, the verification layer does not survive independently and pass to new hardware. It is absorbed whole into the new dominant paradigm.
The ancient Chinese tiger tally was once the sole physical anchor for the movement of military authority. When the logic of mobilizing troops evolved from a bronze object split in two to a document of the Bureau of Military Affairs, the tiger tally was left with only the fate of the museum. It still existed. It was still exquisite. It was still a masterpiece of craft. But it no longer defined what military authority was.
Apple’s Secure Enclave will share that fate. In 2040 it will still exist, still function, still be used in certain scenarios. But it will no longer define what identity is.
The second countervailing force is luxurification. When compute becomes formless and cheap, the visible hardware device may instead reacquire a status premium. Rolex’s market capitalization has nothing to do with telling time. The iPhone already performs this function across several emerging markets.
Apple could travel this path and run in steady state at around two trillion dollars of market capitalization for twenty years. Its margins might even rise. But that path must be chosen actively by the board — actively surrendering the identity of a technology company, accepting an Hermès-like existence. The current board does not have the cognitive capacity to make that judgment, and Ternus’s appointment is the personnel evidence of that absence.
Apple will pretend it is still a technology company while the capital markets push its head down and demote it. The demotion is more painful than the voluntary acceptance of demotion would have been.
VI. The Lag Function of the Capital Markets
The capital markets acknowledge the end of a paradigm with a structural delay. The cause is no longer analyst cognitive failure; it is the mechanical holdings of passive index funds. Vanguard, BlackRock, and State Street hold Apple by the logic of its S&P 500 weight, not by fundamental judgment. As Apple begins to lose its fundamentals, passive holdings lag the deterioration — this is structural buying that does not sell on its own.
The threshold of abandonment will not be a single failed keynote. It will be the moment Apple falls out of the top ten weighted constituents of the S&P 500. Passive funds are forced to trim; reflexivity begins. Until then, Apple floats on a high market-cap plateau for years — every quarter’s earnings reporting services revenue at a new high, every year a marginal five to eight percent decline in market cap, every keynote mocked a little more.
The reasonable estimate for that threshold is between 2030 and 2032. Until then, Apple will continue, in the numbers, to look unassailable.
The railroads were still the peak of American market capitalization in 1920. It was not until the 1971 Penn Central bankruptcy that the capital markets publicly acknowledged the end of that paradigm. A full fifty years of delay. No single year was the year of collapse — only the year-over-year marginal erosion, until one day a single large institution’s finances could no longer carry it.
VII. Kodak’s Immune System
Kodak invented the digital camera. In 1975. More than a decade before Canon, Nikon, Sony. The Kodak engineer Steven Sasson brought the prototype to management and was told: this is interesting, but don’t tell anyone.
Kodak was never truly at the digital-camera table. It invented the table and then left the room.
Why? Because Kodak’s profit model was film as a consumable. Every photograph was a small payment. The digital camera eliminated that payment stream. Kodak’s management was structurally incapable of accepting the destruction of its own profit model by its own invention.
This was not a misjudgment. It was the organization’s instinctive protection of its own profit stream overriding its judgment about the future.
Apple’s isomorphism with Kodak is not at the product layer. It is at the structural layer.
Apple’s profit model is the hardware premium. The logic of the AI paradigm is compute consumption. For Apple to enter the AI table at full force, it must accept the gradual replacement of hardware-sales revenue by compute-subscription revenue — which means destroying the entire commercial logic on which its four trillion dollars of market capitalization is built.
Apple’s board, Apple’s financial structure, Apple’s shareholder expectations — all of these forces work to prevent that self-destruction. This is not the failure of any individual decision. It is the immune system of the organizational entity at work.
Kodak’s immune system drove it to bankruptcy in 2012, with a market capitalization of one hundred forty-five million dollars, fallen from thirty-one billion in 1996. Kodak was not defeated by any single product. It was slowly poisoned by its own immune system.
Apple’s immune system will not drive it to bankruptcy — it is too rich, its moat too deep. But the same immune system will leave it structurally marginalized in the AI era.
VIII. Ternus’s Real Task
Back to the appointment itself.
Ternus will not be told he was appointed to manage the dusk of an era. He will not be told that the board, professionally incapable of judging whether the AI investment is sound, therefore chose a CEO who would not demand it. He will be told he will lead Apple into the next decade.
He will work hard. He will do what he can. M-series efficiency will keep improving. Apple Intelligence will be deployed incrementally. The next Vision Pro will be lighter, cheaper, better. These will all be genuine engineering achievements.
They will not change Apple’s position in the new paradigm.
On June 8, 2026, Apple unveiled a rebuilt Siri AI, the foundation of which is the Apple Foundation Models. The most revealing fact of the keynote was not in the keynote. It surfaced in the technical session afterward: Apple tried first to run the new Siri model on its own Private Cloud Compute hardware, the model ran too slowly in testing, and the heavier workloads were therefore pushed onto Google’s servers — specifically, onto Nvidia Blackwell B200 chips inside Google’s cloud, with Nvidia’s confidential-computing feature handling the encryption. Apple’s own AI executive conceded that the top cloud model, AFM Cloud Pro, is comparable to Google’s frontier Gemini models. Apple had evaluated OpenAI and Anthropic before choosing Google.
Read that sequence in order. Apple attempted self-sufficiency. Self-sufficiency failed in testing. Apple fell back to running a Gemini-distilled Apple model in Google’s cloud, on Nvidia’s chips. This is not merely a procurement decision. For this workload, at this scale, it is a capability defeat, disclosed as a footnote.
After the keynote, Apple executives convened a press session specifically to deny that Siri is a Gemini reskin, insisting that users never touch Google’s code, that the foundation models are distilled from Gemini but are Apple’s own.
Companies do not convene post-keynote clarification sessions around accusations they consider harmless. The denial itself is a confession — that Apple knows precisely where its most fatal vulnerability lies, and knows the whole world has seen it. What it denies is not the facts: the model is distilled from Gemini, the heaviest workload runs in Google’s cloud on Nvidia’s chips, and Apple could not run that class of workload at acceptable speed on its own Private Cloud Compute hardware. All of that it concedes. What it denies is only the word reskin. It is arguing: the shell, at least, is paint I applied myself. This is exactly the last defense a company makes as it declines from paradigm-definer to physical shell.
The clearest read on the keynote came from the market’s own accountants. The single most bullish analyst price target published that week justified its optimism on Apple by citing the read-through to Alphabet — that is, the logic for buying Apple was that Apple now runs on Google. When the bull case for a company is that it has become a distribution channel for its competitor, the verdict is already in; only its acknowledgment is delayed.
Ternus’s real historical task is to preside, without knowing it, over Apple’s transition from definer of the era to one of the era’s physical shells. He will step down around 2034, leaving behind an Apple still vast in market capitalization, still healthy in margin, but already, on every meaningful dimension, no longer defining the future. By then, another analysis will review his tenure in language much like this.
If, before stepping down, he actively drove the luxurification transition, he would be remembered as Apple’s second savior — saving the company itself by acknowledging the end of the paradigm. This requires a cognitive act he has not demonstrated across his career: surrendering engineering perfectionism, accepting the marginalization of the category.
Acts of that kind are exceptionally rare in history. This is not a judgment of him personally. It is the statistical observation of the cognitive pattern of managers at that level.
In the keynote that defined the direction of Apple’s next decade, the man who will lead Apple into that decade did not appear. John Ternus takes over as CEO in September; the stage of June 8, 2026 belonged to a tearful, departing Cook. An offensive succession puts the new CEO on the most important stage to set the tone. A managed decline lets the old CEO take his bow while the new one waits backstage until September. Apple chose the stagecraft of the latter — and it spoke the nature of the succession more honestly than any analysis could.
In the room where Ternus was appointed, no one was present to tell the board what it was doing. Jobs was certainly not. Even if he were alive, he could not solve Apple’s problem in 2026 — because the problem is no longer the loss of an organizing principle but the deprecation of the category that organizing principle defined. Jobs’s genius was taste. Taste was the scarce input in 1997. In 2026 the bottleneck is compute and electricity. Taste has no power to rescue.
Four trillion dollars of market capitalization was no individual’s achievement. It was the physical byproduct of mobile compute penetration traveling from zero to saturation. When the window closes, that number closes too — slowly, at length, concealed by repeated rebounds, until one day it is confirmed by an ordinary index-fund rebalancing notice.
Ternus will have stepped down by then. He will not be permitted to see the verdict.