Aya Research
DEEP RESEARCH
中EN

Apple: The Only Company in the Mag 7 That Refused to Buy a Receipt — It Bought Back Its Own Stock Instead

A panoramic research report — July 5, 2026

There are two checks crossing in the mail between Cupertino and Mountain View right now, and the difference between them is the strangest fact in large-cap technology.

Google pays Apple roughly $20 billion a year for the right to be the default search engine on the iPhone. Apple pays Google roughly $1 billion a year for the right to run a custom 1.2-trillion-parameter Gemini model as the brain of the rebuilt Siri. One company is renting Apple's front door. The other is renting Google's mind. The net flow — about $19 billion a year in Apple's favor — is a fair one-line summary of Apple's entire strategic position in the AI era: it owns the most valuable distribution real estate in consumer technology, it has effectively stopped trying to build frontier intelligence itself, and it charges everyone else for access while paying a comparatively trivial toll for the intelligence it needs.

Whether that position is a fortress or a slowly closing trap is the question this report exists to examine.

The occasion for examining it now is simple. Over the past several weeks we have audited the other six members of the Magnificent Seven — Nvidia, Meta, Alphabet, Microsoft, Amazon, and Tesla — plus the memory supplier underneath all of them, SK hynix. Across those reports a single organizing device kept proving useful: the receipt. In 2026, the market has stopped paying for AI stories and started demanding receipts — signed contracts, disclosed backlogs, take-or-pay commitments, revenue that has already landed. Microsoft shows a $627 billion RPO. Amazon shows a $364 billion backlog before counting its newest $100 billion customer. Alphabet shows $462 billion. SK hynix has sold out its HBM production years forward and trades at 5.5× forward earnings anyway. Tesla shows essentially no receipts at all and trades at roughly 196× forward earnings, which is why we called it the control case.

Apple is the seventh audit, and it breaks the framework in the most instructive possible way. Apple has no AI receipts because Apple is not selling AI. It has no capex receipts because it is not buying AI infrastructure — its roughly $13 billion of annual capital expenditure is a rounding error against the $660–690 billion its five hyperscaler peers will spend in 2026 alone, a ratio of about fifty to one. Apple is the only member of the group that declined to enter the arms race entirely. And the market's response to this abstention has been to award Apple the highest trailing multiple of any Mag 7 company except Tesla — 37.3× trailing earnings, 32.1× forward — attached to what was, until two quarters ago, the slowest revenue growth in the group.

The slowest grower wears the richest sober multiple, pays for nothing, promises nothing, and buys back its own stock at a pace no company in history has matched: more than $1 trillion returned since 2012. If the rest of this series has been about companies writing receipts for the future, Apple is the company that decided the future would come to it, as it always has, and priced admission accordingly.

This report follows the structure of the series: first the longitudinal story — how a company that was ninety days from bankruptcy became the most efficient cash machine ever built; then the cross-section — where every business line, legal siege, and competitive threat stands as of July 2026; then the synthesis — what 37 times earnings actually buys, what Warren Buffett's five-year exit says, and what to watch as a hardware engineer named John Ternus takes the keys in September.

Part I: The Longitudinal Story

1. Ninety Days from Zero

Every member of the Magnificent Seven has a near-death story somewhere in its past, but Apple's is the founding myth of the entire genre, and it is worth retelling precisely because the numbers have become hard to believe.

In 1997, Apple Computer was losing roughly $1 billion a year and had, by Steve Jobs's own later account, about ninety days of cash left before insolvency. The company that had gone public in December 1980 at $22 a share — the largest IPO since Ford in 1956 — had spent twelve years after Jobs's 1985 ouster shrinking into an irrelevant niche vendor with single-digit PC market share and a product line so confused that Jobs, upon returning via the NeXT acquisition, killed more than 70% of it in his first year.

The rescue came from the least likely direction imaginable. On August 6, 1997, at Macworld Boston, Bill Gates appeared on a giant video screen above the stage — a staging choice that the audience read, audibly, as Big Brother — to announce that Microsoft would invest $150 million in Apple preferred stock, commit to five more years of Office for Mac, and settle the companies' long-running patent disputes. The crowd booed. Jobs scolded them: "We have to let go of this notion that for Apple to win, Microsoft has to lose."

The scale of the reversal since is worth stating in the driest possible terms, because dry terms are the only ones it doesn't break. The company that could not fund ninety-one days of operations in August 1997 generated $28.7 billion of operating cash flow in the single quarter ending March 2026 — roughly one billion dollars every three days, or the entire 1997 rescue package every eleven hours. No turnaround in the history of public markets is comparable in magnitude, and the fact that investors now treat Apple as the definition of safety — the bond of the Mag 7, as we will argue later — is itself the most extreme possible measure of how completely the near-death has been forgotten. Forgetting, in markets, is not free; it gets priced. A meaningful fraction of the 37× multiple this report keeps circling is, functionally, the capitalized value of institutional amnesia about the fact that consumer-technology franchises can and do die.

Three details of that moment matter for the 2026 investor. First, the sum: $150 million. The company it saved is worth $4.5 trillion today — the rescue capital has been diluted into legend by a factor of thirty thousand. Microsoft, incidentally, sold its entire position by 2003, exiting one of the great compounding runs in market history for a modest gain. Second, the structural lesson Apple's leadership internalized: survival can depend on a rival's checkbook, and dependence is a thing to be engineered out of, permanently. The next twenty-five years of Apple strategy — own the silicon, own the operating system, own the store, own the customer relationship — are legible as one long reaction to the humiliation of August 1997. Third, the irony that organizes this report: in January 2026, Apple once again signed a deal in which a rival supplies something existential — except this time Apple is paying, the rival is Google, and the existential input is not cash but intelligence itself.

2. The Second Founding: 1997–2011

What happened between the Microsoft rescue and Jobs's death is the most compressed value-creation sequence in business history, and its rhythm — one category-defining product roughly every three years — established the expectation that has haunted every Apple executive since.

The iMac arrived in 1998 and stabilized the patient. The iPod arrived in October 2001 and gave Apple its first hit outside the computer. Then, on January 9, 2007, Jobs stood on the Macworld stage and introduced "three revolutionary products" — a widescreen iPod, a phone, and an internet communicator — and paused until the audience understood they were one device. The iPhone went on sale that June. Eighteen months later, on July 10, 2008, the App Store opened with about 500 applications, and the most profitable toll booth in the history of commerce began collecting.

The iPad followed in 2010, completing the portfolio. Then, on August 24, 2011, Jobs resigned as CEO and handed the company to his operations chief, Tim Cook. On October 4, 2011, Cook's first keynote introduced the iPhone 4S and a voice assistant called Siri — remember that name and that date, because the fifteen-year arc between Siri's debut and the Gemini rental agreement is the spine of Apple's AI story. Jobs died the following day.

Wall Street's consensus on Cook in 2011 deserves to be quoted against its subsequent record. He was the supply-chain guy, the operator, the caretaker — a man with no product vision inheriting a company whose entire value was product vision. Apple's market capitalization on the day of the handover was roughly $350 billion. It is $4.53 trillion today. The caretaker produced a thirteenfold return, and the way he did it — not by inventing the next iPhone, but by monetizing the existing one with an intensity nobody imagined possible — is the subject of the next section, and arguably the true precedent for what the market is now betting John Ternus can repeat.

3. The Operator's Trilogy: Services, Silicon, and the Buyback Machine

Cook's Apple made three strategic moves that, taken together, explain nearly everything about the company's 2026 income statement. None of them was a new product category. All of them were monetization architecture.

The services pivot. In fiscal 2016, Apple's services segment — the App Store, iCloud, licensing, Apple Music — generated about $24 billion, and Cook did something Apple had never done: he gave Wall Street a forward target, promising in early 2017 to double services revenue by 2020. Apple hit the target roughly six months early. By fiscal 2025 services revenue reached $109.2 billion, growing 14% a year, carrying gross margins around 75–76% against roughly 35–37% for hardware, and contributing over a quarter of total revenue. More than one billion paid subscriptions now sit on top of an installed base of 2.3 billion active devices. The strategic meaning is simple: Cook converted a hit-driven hardware company into a subscription annuity wrapped around a hardware refresh cycle. This is the single largest driver of the multiple expansion from the 10–15× of the early Cook years to the 37× of today.

The silicon migration. Apple had designed its own A-series phone chips since 2010, but on November 10, 2020, it announced the M1 and the abandonment of Intel across the Mac line — a transition completed in under two years, executed so cleanly that it is now taught as the benchmark for platform migrations. The pattern has since extended relentlessly: the C1 and C1X cellular modems (shipping in the iPhone Air), the C2 modem expected in the iPhone 18 Pro, the N1 Wi-Fi/Bluetooth chip that displaced Broadcom in the iPhone 17, an A20 Pro headed for TSMC's 2-nanometer node, and — most relevant to this series — a Broadcom-partnered AI inference server chip called Baltra, targeted at mass production in the second half of 2026. Each migration converts a supplier's margin into Apple's margin. It is capex-light vertical integration: Apple spends on engineers and TSMC wafer allocations, not on data-center shells.

The buyback machine. Since initiating its capital-return program in 2012, Apple has returned more than $1 trillion to shareholders, roughly $850 billion of it through repurchases — no other company in history is within shouting distance. The recent cadence: a $110 billion authorization in May 2024 (the largest single buyback authorization in U.S. history), $100 billion more in May 2025, and another $100 billion on April 30, 2026, alongside a dividend raised 4% to $0.27 per quarter. Over fiscal 2021–2025, Apple returned approximately $522 billion — about 97% of the free cash flow it generated in the same window. And in the most recent earnings call, management formally abandoned its long-standing "net cash neutral" target, with net cash sitting around $62 billion ($147 billion in cash and securities against $85 billion of debt). The signal buried in that abandonment: the buyback machine now has permission to run harder than cash generation itself.

The buyback's arithmetic deserves one more paragraph, because it is the quiet engine of the EPS growth the market capitalizes. Apple's share count has shrunk by more than a third since the program began — from over 26 billion split-adjusted shares to roughly 15 billion — which means a dollar of net income today is spread across roughly 40% fewer claims than in 2013. Decompose the consensus 14% EPS growth expected for fiscal 2026 and the machine is visible: perhaps two to three points of it come from share retirement alone, compounding mechanically and forever, independent of whether a single additional iPhone is sold. This is the least glamorous growth algorithm in the Magnificent Seven and by far the most reliable — no model risk, no execution risk, no dependence on an AI capex cycle resolving favorably. It is also, critics note fairly, the algorithm of a company that has run out of higher-returning uses for $100 billion a year: the same board that approved three consecutive $100+ billion repurchase authorizations approved roughly $13 billion of capital expenditure. The ratio of money spent buying the company's own past to money spent building its future is about eight to one, and no other line in the financial statements states Apple's strategic self-assessment as plainly.

Hold the trilogy against the series framework and the contrast sharpens. Amazon returns nothing to shareholders and pours everything into $200 billion of capex; the market grants it a permission slip. Meta bought back zero dollars of stock in the first quarter of 2026 — the tell we flagged in that report — because Meta Compute is eating everything. Apple is the mirror image: it builds almost nothing and returns almost everything. In receipt terms, the only receipts Apple writes are made out to its own shareholders.

4. The Trillion-Dollar Staircase

Apple's market-cap milestones, laid end to end, quietly narrate its change in status from leader to follower — a demotion in narrative rank that most investors have not consciously registered.

On August 2, 2018, Apple became the first U.S. company worth $1 trillion. On August 19, 2020, it was first to $2 trillion. On January 3, 2022, it was first to touch $3 trillion intraday. Three consecutive milestones, three times the undisputed frontrunner of American capitalism.

Then, on October 28, 2025, Apple crossed $4 trillion — third, behind Nvidia and Microsoft. The staircase's final step tells the story of the entire AI era in one line: the two companies that got there first were the arms dealer of the AI buildout and its most aggressive customer. The company that had defined the prior decade of market leadership arrived at $4 trillion as an afterthought, on a day when most financial coverage treated the milestone as confirmation of AI-era hierarchy rather than Apple's achievement.

The stock-price record inside that staircase is similarly two-toned. Apple returned +30.7% in 2024, a muted +9.1% in 2025 while AI names ran, and about -8.4% year-to-date in 2026 — which, remarkably, makes it the best-performing Mag 7 stock of 2026 so far, in a year when Microsoft is down about 24% and the entire complex has deflated. The company that missed the AI rally is winning the AI hangover. That inversion — laggard in euphoria, refuge in doubt — is the market telling you what it actually thinks Apple is: not a growth story at all, but the sector's bond.

5. The Fifteen-Year Nap: Siri's Long Road to Someone Else's Brain

Now the uncomfortable chapter, because no honest Apple report can route around it.

Apple shipped the first mainstream voice assistant on October 4, 2011 — ChatGPT's debut was eleven years away, and Siri was, briefly, the most futuristic thing a consumer could own. What followed was a fifteen-year institutional failure with remarkably precise documentation. Siri received no systematic architectural upgrade through the entire deep-learning revolution. The June 2024 response to ChatGPT — Apple Intelligence, announced with a ChatGPT integration at WWDC — shipped piecemeal starting that October to reviews ranging from indifferent to mocking. The centerpiece "personalized Siri" was demoed but never shipped; on March 7, 2025, Apple issued the rarest of documents, a public admission that the feature was delayed to "the coming year."

Then the organizational reckoning, in unusually visible stages. On March 20, 2025, after Cook reportedly lost confidence in AI chief John Giannandrea, Siri was taken from him and handed to Vision Pro creator Mike Rockwell, reporting to software chief Craig Federighi. In July 2025, Meta poached Ruoming Pang, head of Apple's roughly 100-person foundation-models team, with a package reported near $200 million; senior researchers followed him out the door, and the reported trigger for the exodus was itself damning — internal debate over whether to abandon Apple's own models for OpenAI's or Anthropic's. By December 2025 Giannandrea's retirement was announced, with former Google Gemini engineering leader Amar Subramanya taking over as VP of AI.

The strategic surrender was formalized on January 12, 2026: a multi-year agreement under which Google supplies a custom 1.2-trillion-parameter Gemini model — a mixture-of-experts design roughly eight times the scale of Apple's own 150-billion-parameter cloud model — to power the rebuilt Siri, for roughly $1 billion a year, running on Apple's own Private Cloud Compute infrastructure so that Google never touches user data. At WWDC on June 8, 2026, Apple unveiled the result: a reconstructed "Siri AI" shipping with iOS 27 this fall, finally delivering the personal-context, screen-awareness, and cross-app-action capabilities first promised in June 2024, with management claiming a 92% multi-step task success rate against the old Siri's 58%.

The architecture of the rental deserves attention, because it is where Apple salvaged strategy from surrender. The Gemini model runs exclusively on Apple's own Private Cloud Compute servers — increasingly built on Apple's own Baltra silicon — inside an attestation framework where Google receives neither queries nor user data, and where the user-facing brand, interface, and privacy promise remain entirely Apple's. Functionally, Apple has positioned Google the way it positions Foxconn: as a contract manufacturer, of cognition rather than casings, invisible to the customer and swappable by design. The bake-off that preceded the deal reinforces the point — Anthropic reportedly priced its model higher, OpenAI's Microsoft entanglements complicated terms, and Apple chose the vendor most willing to be commoditized. Whether frontier intelligence stays commoditizable is the trillion-dollar assumption underneath the whole arrangement; every hyperscaler audited in this series is spending nine figures a day betting it doesn't. But as a piece of deal-making, Apple converted a fifteen-year R&D failure into a supplier relationship costing 0.9% of one year's free cash flow, with no lock-in, no capex, and no brand transfer. It is the cheapest exit from a strategic hole any Mag 7 company has purchased this cycle.

Set the two dates side by side: Apple invented the category in October 2011 and rented the winning implementation in January 2026. In our Alphabet report we told the mirror-image story — Google invented the Transformer and watched all eight authors walk out the door to build its competitors. The pair of failures is symmetrical and instructive: Google had the research and fumbled the product; Apple had the product surface — the most valuable one on earth, 2.3 billion devices — and never built the research. Of the two, the market has decided Apple's failure is the survivable one, because intelligence, it turns out, can be rented for $1 billion a year, while distribution cannot be rented at any price. Google's own $20 billion annual payment for Apple's front door is the proof.

6. The Graveyard of Next iPhones

Before assessing what Apple is, tally what it tried to become and didn't — because the graveyard explains both the bear case (Apple can no longer invent) and, perversely, the bull case (Apple no longer bleeds money trying).

Project Titan, the car: launched 2014, peaked near 5,000 employees, pivoted repeatedly between full autonomy and conventional EV, consumed an estimated $10+ billion (Apple never disclosed the figure), and was canceled on February 27, 2024, with roughly 2,000 staff largely redirected to generative AI. A decade of effort produced zero shipped products — and, in the cancellation, one of the great silent write-offs of the promise-inventory era. Tesla's Dojo, we noted in the previous report, died the same kind of death in the same news cycle era: billions in narrated future value, extinguished without ceremony.

Vision Pro, the $3,499 face computer: launched February 2024 to approximately 390,000 first-year units against initial expectations of 700–800 thousand, then collapsed — full-year 2025 shipments are estimated at 80–90 thousand, the holiday quarter around 45,000, and assembler Luxshare has reportedly halted production. An M5-refresh in October 2025 changed nothing. The second-generation headset is shelved; resources have shifted to a cheaper Vision Air targeted for 2028–2029 and, more importantly, to camera-and-audio AI glasses (no display) now delayed to late 2027 at a rumored $299–499 price point — a segment where Meta, via EssilorLuxottica, sold over 7 million units in 2025 and is discussing capacity of 20–30 million. Apple invented the modern smartwatch and wireless earbud categories; in glasses, for the first time in the wearables era, it is the follower by two full generations.

The honest read of the graveyard: Apple's last organically invented mass-market category remains the Apple Watch (2015) — eleven years ago. Everything since has been either a refinement, a cancellation, or a flop. The bull's rejoinder is equally honest: the failures were cheap relative to Apple's scale, none was existential, and the discipline to kill Titan and starve Vision Pro — rather than pour hundreds of billions into pride — is exactly the capital discipline the market now pays 37× for. Apple's graveyard cost perhaps $15–20 billion across a decade. Its peers now spend that much on data centers per month.

Part II: The Cross-Section, July 2026

7. The Machine Still Prints — Loudly

Whatever one thinks of Apple's AI position, the operating results of the last two quarters ended the "stagnation" narrative with prejudice, and any current analysis has to start from that fact.

Fiscal Q1 2026 (the December 2025 quarter) was the best quarter in the company's history: revenue of $143.8 billion, up 16% — a growth rate Apple hadn't printed since the pandemic supercycle — with net income of $42.1 billion, EPS of $2.84 up 19%, and gross margin of 48.2%. iPhone revenue alone was $85.3 billion, up 23%. Greater China, the region that spent two years as Apple's biggest bear talking point, grew 38% to $25.5 billion.

Fiscal Q2 2026 (the March quarter, reported April 30) confirmed it wasn't a one-off: revenue $111.2 billion, up 17%, a March-quarter record; EPS $2.01, up 22%; gross margin 49.3%, up 220 basis points and an all-time high; iPhone up 22% to $57.0 billion; services at a record $31.0 billion, up 16%, with services gross margin of 76.7%; China up another 28% to $20.5 billion, crushing estimates. First-half fiscal 2026 revenue is running +16.4%; trailing-twelve-month revenue is $451.4 billion with $101.1 billion of free cash flow.

Step back one fiscal year and the acceleration is even more striking than the levels. The four quarters ending in March 2026 read, in sequence: $94.0 billion (+10%, with an $800 million tariff bill), $102.5 billion (+8%, gross margin 47.2%, tariffs $1.1 billion, Greater China still negative at -3.6%), then $143.8 billion (+16%), then $111.2 billion (+17%). Gross margin climbed a full 210 basis points across the arc, from 47.2% to 49.3%, even as quarterly tariff costs tripled toward a $1.4 billion guided run rate — pricing power absorbing a five-billion-dollar-a-year political tax without visible strain. A company whose fiscal 2025 grew 6.4% — the figure that anchored every "Apple is ex-growth" argument, including the one embedded in this series' earlier reports — is suddenly compounding at pandemic-supercycle rates. The whole bull-bear argument of Part III turns on a single question this arc poses: is the step-change a new trajectory or a pulled-forward crest? The production cuts say crest. The China numbers say trajectory. The multiple, at 37×, has already voted.

The engine is the iPhone 17 cycle, and the composition matters. Counterpoint data show the iPhone 17 as the world's best-selling smartphone model (about 6% of all units sold globally), with the 17 Pro Max and 17 Pro in second and third — and in Q1 2026 Apple topped global calendar-first-quarter shipments for the first time ever. The one failure in the lineup is instructive: the ultra-thin iPhone Air undershot and was cut from production early — buyers, it turns out, wanted cameras, battery, and value, not thinness. The demand driver was not AI. Survey data are unusually blunt on this: a July 2026 upgrade survey found neither Apple Intelligence, nor the new Siri, nor even the coming foldable meaningfully driving upgrade intent, and UBS's 7,500-person panel showed willingness to upgrade early for AI falling five points to 24%. What actually drove the supercycle was an aging installed base of 2.3 billion devices, substantive hardware upgrades (cooling, displays, cameras) at held prices, and a China market where the iPhone 17's value proposition landed — Q4 2025 China shipments up 28% to retake first place, Q1 2026 up another 42%.

Two caveats temper the victory lap. First, the July 1 reports of a 15% production cut on the iPhone 17 line are normal end-of-cycle behavior ahead of September's iPhone 18 launch (Chinese rivals cut 15–30% in the same window), but they mark the supercycle's crest. Second — and this is where Apple's story cables directly into our SK hynix report — management guided fiscal Q3 to +14–17% revenue but warned explicitly that memory costs turn "significantly higher" from June onward, pressuring product margins into the very quarter the AI buildout's DRAM squeeze arrives on consumer bills of materials. Apple, the one Mag 7 member not buying AI infrastructure, is about to pay the AI infrastructure tax anyway — through the memory market, to the benefit of the company we covered at 5.5× forward earnings. The four-sided market has a fourth side, and Apple is standing on it.

8. The $124 Billion Toll Booth — Under Siege in Three Courtrooms

Annualize the March quarter and Apple's services business is now a roughly $124 billion-a-year enterprise growing mid-teens at 76.7% gross margin. Its estimated composition: $30–35 billion from the App Store, $17–20 billion of licensing dominated by the Google search payment, $10+ billion from iCloud, $8–10 billion from Apple Music, $8–9 billion from AppleCare, and a few billion from TV+ and payments. If it were a standalone company, services would be roughly the size of Nike and Starbucks combined, at Ferrari margins — it is the single most valuable business line in the world on a margin-adjusted basis, and it is the true collateral behind the 37× multiple.

It is also the part of Apple under simultaneous legal siege on three fronts, and 2026 is the year the sieges reach their decisive engagements.

Front one: Epic and the App Store's pricing power. The April 2025 contempt ruling forced Apple to permit commission-free external payment links in the U.S. In December 2025 the Ninth Circuit softened this to permit a "reasonable fee" below the prior 27%, and on July 2, 2026 — three days before this report — the Supreme Court granted certiorari, with argument in the October 2026 term. Estimated annual revenue at risk runs $1–4 billion depending on how much volume leaks to external payment rails. The Supreme Court case is now the terminal battle over whether the toll booth's rate card is Apple's to set.

Front two: the Google payment. Judge Mehta's remedies in the search antitrust case, finalized December 5, 2025, were close to the best realistic outcome for Apple: Google may keep paying for default placement, but the deals must be non-exclusive and renewed annually. The roughly $20 billion annual payment — near-pure profit, plausibly 15%+ of Apple's pre-tax income — survives, but its contractual security has degraded from long-term arrangement to one-year renewable, and the DOJ's February 2026 appeal seeks harsher terms while Google's cross-appeal defends the arrangement. An annuity that reprices annually under appellate uncertainty is worth structurally less than the same annuity signed long — a receipt-quality downgrade, in this series' terms, even though the cash still arrives.

Front three: Europe and the states. The EU fined Apple €500 million in April 2025 under the DMA and has forced sideloading, third-party payments, alternative browser engines, and a fee structure whose minimum effective rate is now around 7% for developers routing around Apple — a live experiment in what App Store economics look like when regulators set the take rate. Texas's app-store age-verification law took effect January 1, 2026, and the federal Open App Markets Act was reintroduced in mid-2025. None of these individually dents a $124 billion segment; jointly they establish the direction of regulatory travel on every continent: the toll rate only compresses from here.

It is worth pausing on the asymmetry of stakes across the three fronts, because they are not equal. The European front is loud but small: the EU is perhaps 7% of App Store billings, the €500 million fine is a day and a half of free cash flow, and the region's new ~7% floor rate is an experiment whose early evidence suggests most developers stay inside Apple's rails anyway, because payment plumbing and discovery are worth something even when the law says they're optional. The Epic front is medium-sized and terminal: whatever the Supreme Court decides will be the permanent constitutional settlement of U.S. app-store economics. The Google front is the big one, and it is routinely misfiled as Google's problem rather than Apple's: the $20 billion arrives with no associated cost of revenue worth mentioning, meaning it may represent something like 15 cents of every dollar of Apple's pre-tax income, delivered annually by a counterparty that a federal court has ruled must be free to walk away every twelve months — and that is simultaneously the landlord of Siri's new brain. No other Mag 7 company has a single line item of remotely comparable size resting on so short a contractual leash held by so entangled a counterparty.

The investment translation: services is simultaneously Apple's crown jewel and the only Mag 7 profit pool whose pricing is being litigated in three jurisdictions at once. The bull case is that volume growth (subscriptions +14–16%) outruns rate compression for years. The bear case is that the multiple is capitalizing 76.7%-margin revenue whose margin ceiling is now set in courtrooms Apple doesn't control.

9. The Two Checks in the Mail

Return to the image this report opened with, because it deserves a full accounting — the $20 billion inbound check and the $1 billion outbound check are the cleanest lens on what Apple has actually become.

The inbound check says: distribution is the scarcest asset in consumer technology. Google — owner of the best search engine, the best models by many measures, and the Transformer patent lineage — concluded that all of it is worth little without placement on Apple's 2.3 billion devices, and pays accordingly. Even a federal antitrust judge, having found the underlying market illegally monopolized, declined to stop the payment, accepting that cutting it off would primarily injure Apple — a bystander — rather than fix search competition. The payment survived because Apple's position is that structurally load-bearing.

The outbound check says: frontier intelligence has become a commodity input, at least at the level Apple needs it. When Apple finally decided Siri required a state-of-the-art model, it did not spend $50 billion on data centers or acquire a frontier lab; it ran a bake-off (Anthropic reportedly priced higher; OpenAI was entangled with Microsoft), picked Gemini, and rented a custom 1.2-trillion-parameter model for roughly $1 billion a year — about three days of Apple's operating cash flow, or 0.3% of the collective annual capex of its hyperscaler peers. The model runs on Apple's own Private Cloud Compute hardware, Google never sees user data, and Apple retains the option to swap vendors — or slot in its own model, should its 150-billion-parameter internal effort ever catch up — behind an interface it fully controls.

Notice what the two checks jointly assert: the scarce thing rents for $20 billion, the abundant thing rents for $1 billion, and Apple is on the right side of both trades. That is the entire "certainty premium" thesis in miniature, and it is genuinely strong. The rejoinder — which we developed at length in the Microsoft report's estate-versus-platform-tax framework — is that renting your critical input builds no estate. Microsoft's OpenAI arrangement, for all its April 2026 humiliations, left Microsoft holding a 27% stake worth perhaps $135 billion. Apple's Gemini arrangement leaves Apple holding a receipt for services rendered. If conversational AI someday replaces rather than inhabits the smartphone interface — if the front door itself moves — Apple will discover that it spent the transition renting the one thing it needed to own. The two checks are a fortress today. The lease terms are annual.

10. The Absentee: Thirteen Billion Against Six Hundred Sixty

Now the heart of the cross-section, and this report's reason for existing within the series: the capex table.

In 2026, the AI buildout's principal financiers will spend approximately: Amazon ~$200 billion, Alphabet ~$185 billion, Meta ~$125 billion, Microsoft ~$120 billion — call it $660–690 billion across the hyperscale complex, the figure our SK hynix report treated as the demand side of the memory market and our Nvidia report treated as the revenue pool. Apple's fiscal 2025 capital expenditure was about $12.7 billion; 2026 plans run near $13 billion, with management promising only that growth will be "substantial." Fifty dollars of peer capex for every dollar of Apple's.

The composition of Apple's $13 billion makes the abstention even purer: much of it is tooling for device manufacturing, corporate facilities, and a modest Private Cloud Compute footprint built partly on its own Baltra inference silicon (Broadcom-partnered, TSMC N3E, mass production targeted for late 2026, data-center deployment in 2027). Apple's AI infrastructure strategy, stated plainly, is: rent third-party compute for training, run inference on its own devices' silicon where possible, serve cloud inference from a small owned footprint on chips it designed, and rent the frontier model itself from Google. It is the complete inversion of the Meta Compute doctrine — and of the entire receipts economy this series has documented.

Both readings of the abstention deserve their strongest form.

The bull reading: Apple has correctly identified that in every prior platform era, the money was made not by whoever owned the infrastructure but by whoever owned the customer. Apple did not build cellular networks to profit from mobile; carriers spent the hundreds of billions and Apple captured the margin. It did not build the cloud; it sells devices into it. The hyperscalers' $660 billion is, functionally, other people building the utility layer beneath experiences Apple will distribute — while Apple's $100 billion of annual free cash flow, unencumbered by depreciation bombs, buys back stock. Every dollar of peer capex that commoditizes intelligence makes Apple's distribution more valuable, exactly as the two checks demonstrate. "Smart spending beats big spending," as the bulls' shorthand goes, and the fact that hyperscaler free cash flow is being devoured by capex while Apple's conversion stays pristine (TTM FCF $101.1 billion on $112 billion of net income) is the receipts economy's own verdict.

The bear reading: this is what a franchise in run-off looks like when the run-off is disguised by excellent margins. Apple is harvesting — maximum monetization of an installed base, minimum investment in the next platform — and the abstention is not strategy but incapacity: the company tried to build frontier AI and organizationally failed (Section 5), tried to build the next device and failed (Section 6), and now rationalizes necessity as choice. If AI-native interfaces genuinely reorder consumer computing, the $660 billion isn't a subsidy to Apple's distribution; it is the siege engine being assembled outside the walls, and Apple will meet it with $13 billion and a lease.

We will adjudicate between these in Part III. But note the historical rhyme with October capex season — the shared signal every report in this series has pointed to. When the hyperscalers report their next capex guides in late October 2026, Apple reports the same week. One set of income statements will show the cost of buying the future; Apple's will show the cost of assuming you already own it. Rarely does the market get so clean a controlled experiment on one earnings calendar.

11. Handset Wars: The Empire's Borders, Surveyed

The device business that funds everything is, as of mid-2026, in its strongest competitive position in years on some borders and structurally exposed on others. A tour of the map:

The global crown, contested on a technicality. For full-year 2025 Apple led global smartphone shipments with about 20% share — its first annual win over Samsung since 2011. Q1 2026 splits by scorekeeper: Counterpoint gives Apple 21% and first place; IDC gives Samsung 21.7% and the crown. The discrepancy is methodological noise; the signal is that Apple and Samsung were the only two positive-growth vendors in a global market that shrank roughly 3%.

The premium fortress. In the $600+ segment where all the profit lives, Apple holds about 62% of unit sales and rising. Nobody is within artillery range; Samsung is second and gaining only slowly. This — not aggregate share — is the number the income statement runs on.

China, the whipsaw theater. The two-year arc: Apple fell to 15% share and sixth place in parts of Q3 2025 under Huawei's resurgence, then the iPhone 17 landed and shipments rose 28% in Q4 2025 (retaking first place at 22%) and another 42% in Q1 2026 (19%, second behind Huawei's 20%). The recovery is real, hardware-led, and won without Apple Intelligence — which remains unavailable in China after the Alibaba partnership's regulatory odyssey (CAC review delays, a March 30 accidental launch pulled within hours, trade-war postponements). China simultaneously delivers Apple's fastest regional growth (+33% in the first fiscal half) and its deepest structural dependency: roughly three-quarters of iPhones are still assembled there, against a backdrop where tariff costs ran $800 million, then $1.1 billion, then a guided $1.4 billion across three successive quarters — annualizing above $5 billion.

India, the escape route under construction. Indian factories assembled about 55 million iPhones in 2025 (+53%), roughly 23–25% of global output, headed for 28% in 2026; iPhones became India's single largest export good ($23 billion), with Tata's share of those exports rising from 13% to nearly 40%. The plan: most U.S.-bound iPhones from India by the end of 2026. The friction: Indian assembly still imports Chinese components and costs 5–8% more. The diversification is genuine and, by our reading, the most consequential supply-chain migration in consumer electronics history — but it relocates final assembly, not the component ecosystem, and it does nothing about TSMC's Taiwan concentration, which remains the un-hedgeable tail risk beneath every name in this series.

The cumulative tariff arithmetic, for the record: roughly $3.3 billion of realized cost from April through December 2025, an annualized run rate now above $5 billion, and plausible cumulative drag approaching $20 billion by decade's end if the current regime persists — real money even at Apple's scale, and notably the only line item in this entire report where Apple's cost structure resembles its Mag 7 peers' exposure to policy rather than to markets. The India build-out is the hedge, but a partial one: Indian assembly still imports the Chinese component ecosystem it sits on top of, runs 5–8% more expensive, and covers the United States lane specifically — the rest of the world's iPhones remain overwhelmingly a Chinese product.

The Washington settlement. Apple bought its tariff peace explicitly: a $500 billion U.S. investment pledge in February 2025, raised to $600 billion over four years that August, in exchange for semiconductor-tariff exemption under the "build in America and you're exempt" formula. Call it what the structure suggests — a negotiated tax, paid in capex commitments and press conferences, and cheap at the price against a >$5 billion annual tariff bill that would otherwise compound.

12. The Threats That Haven't Arrived

Every bear case on Apple eventually reduces to one sentence: something replaces the iPhone. The 2026 field report on the candidates:

OpenAI's hardware program — the $6.5 billion acquisition of Jony Ive's io — has slipped from "device in late 2026" to a camera-equipped smart speaker no earlier than February 2027, with a palm-sized screenless device, earbuds, and a pen in the pipeline and a Qualcomm-partnered AI phone explored for 2028. The program is real, lavishly funded, and led by the iPhone's own designer; it has also shipped nothing, and its first product is a speaker — a category Apple, Amazon, and Google saturated a decade ago.

Meta's glasses are the one AI-native form factor with actual escape velocity: 7+ million units through EssilorLuxottica in 2025, tripling year over year, with capacity discussions at 20–30 million annually. Glasses today complement rather than replace the phone — every pair pairs with one — but the category owns a strategic beachhead Apple has repeatedly delayed contesting (its display-less glasses: late 2027; its display glasses: 2029). This is the first wearable category of the modern era where Apple will arrive third, behind Meta and likely Samsung.

The threat already inside the category deserves equal billing with the speculative ones, because it ships today in nine-digit volumes. Samsung's Galaxy S26, launched in February 2026 with Gemini integrated at operating-system depth, does things no iPhone yet does: autonomous multi-step actions inside third-party apps, real-time call translation, and a choice of assistants — Google's, Samsung's, or Perplexity's — as the system default. Coverage framed the S26 as a preview of what Google's own vision of an AI phone looks like, with Apple's rebuilt Siri arriving eight months later running a smaller custom cut of the same underlying model family. So far the competitive damage is undetectable — Apple holds 62% of the premium segment and just printed its two best quarters ever — which is itself evidence for the thesis that AI features do not currently move handset buyers (the same surveys that show tepid iPhone upgrade intent show no S26 AI stampede either). But the structural point stands: in the scenario where assistant quality does start deciding premium handset purchases, Apple's assistant runs on its rival's model, one generation behind its rival's own deployment of it, under a contract its rival can reprice. That is a manageable position, not a commanding one.

The graveyard already filled once. Humane's AI Pin raised $230 million, shipped under 10,000 units, and sold for parts to HP at $116 million. Rabbit's R1 moved about 100,000 units into a wave of returns. The market tested "AI device replaces smartphone" and returned a verdict swift enough that OpenAI itself now frames the smartphone as AI's central vehicle for years to come. Perplexity's CEO put the bull case for Apple in one line: the stronger AI gets, the more valuable the iPhone becomes — because AI needs a sensor-rich, always-carried, payments-enabled terminal, and 2.3 billion of them already exist.

Our assessment, consistent with the series' receipts discipline: the smartphone-replacement thesis currently has no receipts — no shipped category-killer, no demand evidence, two bankruptcies, and a slipping timeline — while the smartphone-absorbs-AI thesis has the iPhone 17 supercycle as a $200-billion-a-year exhibit. Threat timelines matter, though: the window where this could change is 2027–2028, precisely when Apple's own glasses and foldable roadmap either lands or doesn't. The bear case is not wrong; it is unfunded and early.

Part III: Synthesis

13. What the Closed Scissors Mean: Anatomy of the Certainty Premium

Throughout this series we have used one valuation instrument more than any other: the gap between trailing and forward P/E — the scissors — as the market's own estimate of how much the present is expected to change. Nvidia's wide scissors (29.8× trailing, 15.3× forward) price enormous earnings growth and enormous doubt about its durability. SK hynix's grotesque scissors (23× trailing, 5.5× forward) price a boom the market is certain will die. Tesla's inverted telescope (~380× trailing, ~196× forward) prices a transformation that has not begun.

Apple's scissors: 37.3× trailing, 32.1× forward. Nearly closed. The market expects Apple's earnings to grow about 14% and then... to keep being Apple. No cycle mortality, no transformation, no cliff, no takeoff. It is the only valuation in the group that prices stasis — and it prices that stasis at the highest sober multiple in mega-cap tech, richer than Nvidia on trailing earnings, richer than Microsoft, nearly double Meta.

The supporting metrics fill out the picture without changing it: a PEG ratio of 2.49 (paying two and a half turns for each point of expected growth — the group's least favorable), price-to-sales of 10, EV/EBITDA of 28.4, and a return on equity of 141% that is really a monument to the buyback's balance-sheet compression rather than to operating magic. Against its own history, Apple traded between 10× and 18× earnings for most of the 2013–2019 period, re-rated to the high 20s and 30s during the pandemic, and has simply never de-rated since — the multiple survived the 2022 rate shock, the 2024 China scare, and the 2025 AI-laggard narrative essentially intact. Meanwhile the Mag 7's aggregate premium to the other 493 S&P companies sits near a decade low. The market has spent two years compressing the price of growth while leaving the price of certainty untouched, and Apple is where all of that certainty premium pooled.

Decompose what the 37× actually capitalizes and the premium becomes legible, if not necessarily payable:

The bear decomposition of the same 37×: it capitalizes a toll rate under litigation on three continents (Section 8), a $20 billion licensing annuity renewable annually under appeal (Section 9), a China franchise both booming and hostage (Section 11), a memory-cost margin squeeze arriving this quarter (Section 7), fifteen years of AI organizational failure patched by a rental (Section 5), and a product pipeline whose last invented category is eleven years old (Section 6) — at a multiple that assumes none of these ever bites, priced against mid-single-digit normalized growth once the iPhone 17 supercycle crests, which the 15% production cut says is now.

Both decompositions are true. The scissors being closed means the market has chosen the first and priced the second at approximately zero. In a series about receipts, Apple's multiple is the strangest document of all: it is a receipt the market wrote to itself, certifying that the one company that bought nothing and promised nothing is the one company whose future requires no proof.

14. The Confession Tape: What Berkshire's Exit Says

This series has repeatedly used capital-markets behavior as confession — SK hynix listing its ADR at the top, Meta's buyback going to zero, Microsoft's April amendment, Tesla's trillion-dollar comp package as an official sum-of-the-parts. Apple's confession tape belongs to its most famous shareholder.

Warren Buffett's Berkshire Hathaway held roughly 905 million Apple shares entering 2024 — at peak, close to half of Berkshire's entire equity portfolio. Across 2024 he sold about two-thirds of the position in three escalating tranches; the trimming continued through 2025. As of the Q1 2026 13F, Berkshire holds about 228 million shares worth $57.8 billion — a 75% reduction from peak, though still the firm's largest single holding at 22% of the portfolio. Buffett's public explanation (tax rates) was never sized to the sale. The plainer reading: the investor who defined "wonderful company at a fair price" concluded that at 30–37× earnings with single-digit normalized growth, Apple had become a wonderful company at a price that no longer was — and he sold it the entire way up.

The sequencing of the sales sharpens their meaning. Berkshire's first big tranche — 115 million shares — went in the first quarter of 2024, when Apple traded near 26× earnings; the largest tranche, 390 million shares, followed that summer near 30×; the trimming continued as the multiple pushed through the mid-30s. At every stage, the sale price looked expensive against Apple's growth and cheap against where the stock subsequently went — which is to say Buffett was not calling a top, he was refusing a valuation regime. The refusal is the datum. The man who bought the position at an average cost near $35 a share, against roughly $294 today, was never going to be hurt by trimming; what he declined to do was underwrite, with fresh capital-equivalent conviction, the proposition that a 6%-growth company (as it then was) deserved a mid-30s multiple indefinitely. The market has spent every day since taking the other side.

Two footnotes complicate the confession usefully. First: Greg Abel's first quarter as Berkshire's CEO — Q1 2026 — is the first quarter in nine that Berkshire didn't sell Apple. The new management stopped the exit with a $58 billion position intact. Second: the sale's opportunity cost has so far been modest — Apple underperformed the market through 2025, and 2026's relative resilience (best Mag 7 performer at -8.4%) began only after the AI complex cracked. Buffett's exit was neither vindicated nor punished; it was, precisely, a refusal to underwrite the certainty premium. The market's judgment and Buffett's diverge on exactly one variable — what stasis is worth — and that variable is the entire investment question this report keeps arriving at from different directions.

15. The Handover: A Hardware Engineer Inherits the AI Era

On April 20, 2026, Apple announced that Tim Cook, 65, will step down as CEO on September 1, 2026, becoming executive chairman, with hardware engineering chief John Ternus, 50, a 25-year Apple veteran, taking over — the first CEO transition since 2011, landing one week before the iPhone 18 and Apple's first foldable launch.

The 2011 precedent is the obvious template, and its lesson cuts both ways. Cook was doubted as a visionless operator and delivered a thirteenfold return by monetizing rather than inventing. The market's implicit bet on Ternus is a second repetition: that Apple's value lies in disciplined execution atop an unassailable installed base, and a hardware engineer who shipped every Apple Silicon transition is the right steward. The structural difference is what the new CEO inherits. Cook inherited a product at the beginning of its S-curve (the iPhone had sold 73 million cumulative units in 2011; it sells 230+ million a year now) and rode penetration. Ternus inherits the mature plateau — plus the three courtroom sieges, the China dependency, the rented brain, and the question Cook's operational brilliance deferred for fifteen years: whether Apple can still originate a platform.

Ternus's actual record repays a closer look than the succession coverage gave it. He joined Apple's product design group in 2001 and has fingerprints on essentially every hardware line of the modern era — iPads from the beginning, AirPods, the Mac's Apple Silicon transition, and the iPhone since the X generation; he presented the M-series migration keynotes and has increasingly fronted the iPhone launches themselves. What the record shows is flawless execution within categories Apple already owned; what it cannot show, because the opportunity never arose, is whether he can call a new category into existence or force a services-and-AI agenda through an organization whose software and services chiefs (Federighi, Cue, and the new AI leadership) will now report to a hardware man. The bench around him also thinned at precisely the wrong altitude: the operations succession (Khan), finance succession (Parekh), and AI succession (Subramanya) are all first-year incumbents, meaning Apple will enter its most consequential product year since 2007 — foldable, new Siri, Supreme Court term, memory-cost squeeze — with a leadership team whose collective tenure in role is measured in months, supervised by a chairman who spent fifteen years making every final call himself. Executive chairmen who built thirteen-fold returns do not always find it easy to stop deciding.

The surrounding bench has turned over almost entirely in eighteen months: CFO Luca Maestri out (January 2025, Kevan Parekh in), COO Jeff Williams — long the heir presumptive — retired (July 2025, Sabih Khan in), AI chief Giannandrea out (December 2025, Amar Subramanya in), Cook ascending to chairman. This is the most concentrated leadership reset of any Mag 7 company this cycle, executed while the AI talent market strip-mines Apple's research bench at $200 million a head. The transition risk is real but conventional; the deeper signal is the selection: offered a choice of futures, Apple's board picked the man who builds the devices. The estate, they are saying, is the hardware. Everything else — including intelligence itself — can be sourced.

16. The Dashboard: Five Signals With Dates

As with every report in this series, we close the analysis with falsifiable markers rather than adjectives. Five signals, five dates, all within nine months:

1. July 30, 2026 (approx.) — fiscal Q3 earnings. Watch the memory-cost warning materialize or not in product gross margin (guide: 47.5–48.5%, versus 49.3% just printed). This is the hynix squeeze arriving on Apple's P&L, and the first quantified evidence of what the AI buildout costs its one abstainer. Watch, same week, the hyperscaler capex guides — the controlled experiment of Section 10.

2. September 2026 — the double launch. iPhone 18 Pro and the first foldable iPhone debut at expected higher prices, one week after Ternus takes office. The foldable is Apple's first genuinely new hardware form since 2015 and its first product priced against survey data showing tepid upgrade intent. A strong foldable cycle re-arms the hardware story; a soft one leaves the supercycle's crest behind with nothing in front but services math.

3. Fall 2026 — iOS 27 ships the Gemini-powered Siri to 2.3 billion devices. The claimed 92% task-success rate meets reality at civilization scale. This is the single largest consumer AI deployment ever attempted, and its reception decides whether "rent the brain" reads as masterstroke or stopgap — and whether the $20 billion inbound check gains a defensive moat (a Siri good enough that users never leave Apple's surfaces) or loses one.

4. October 2026 — Apple v. Epic at the Supreme Court. The terminal ruling on App Store pricing power, with $1–4 billion of annual high-margin revenue directly at stake and the toll booth's rate-setting sovereignty at issue. Watch alongside it the D.C. Circuit's schedule on the Google-payment appeal — the two courtrooms jointly govern perhaps a quarter of Apple's pre-tax income.

5. Q2–Q3 2026 13Fs — does Abel hold? If Berkshire's pause becomes a durable floor, the most famous skeptic's exit is over and the certainty premium has survived its most credentialed dissent. Renewed selling reopens the question at exactly the moment the growth comps get hard.

17. The Verdict: The Company That Charges Admission to the Future

Seven audits, one cycle, and the ledger now closes where it must — on the company that declined to participate.

Line the series up by how much verifiable future each company has purchased or sold. SK hynix sold its future in signed, take-or-pay contracts years forward and trades at 5.5× forward earnings. Nvidia sells the shovels against $600 billion of documented buildout at 15.3×. Microsoft, Alphabet, and Amazon bought their futures with $500 billion of combined capex against $1.4 trillion of contracted backlog, at 20–25×. Meta bought its future with capex so heavy it zeroed the buyback, at 16× forward. Tesla purchased nothing verifiable, promised everything, and trades at 196×. And Apple — Apple neither bought nor promised. It spent $13 billion maintaining the machine, rented the one input it lacked for $1 billion, returned half a trillion dollars to its owners over five years, and was awarded 32× forward earnings for the demonstration that it didn't need to try.

The certainty premium is not irrational. It capitalizes real things: the deepest moat in consumer technology, the cleanest cash conversion in the group, a services annuity whose like does not exist elsewhere, and — 2026's own lesson — the safest place to stand when an AI trade deflates, as this year's relative performance table shows. Apple at 37× is expensive the way a Zurich vault is expensive. You are not paying for growth; you are paying for the one balance sheet in the complex guaranteed to be intact whichever way the $660 billion experiment resolves.

But the audit obliges us to state the exposed flank with the same clarity. Every component of Apple's stasis is, right now, in motion: the toll rate is before the Supreme Court, the Google annuity renews annually under appeal, the China engine runs on a geopolitical fault line, the margin meets the memory squeeze this quarter, the CEO changes in eight weeks, and the brain of the product is leased from the adversary. Stasis priced at 37× requires all six of those to resolve quietly. The scissors are closed because the market believes nothing will change. The dashboard above lists five dated occasions, inside nine months, on which something could.

Our series began with a company that nearly died three times and now sells the most oversubscribed future in the industry at the cycle's lowest multiple. It ends with a company that nearly died once, twenty-nine years ago, and now sells the most assumed future in the industry at the cycle's highest sober multiple. Between hynix's 5.5× and Apple's 37× lies the market's entire 2026 theory of evidence: signed futures are discounted because cycles die; assumed futures are capitalized because installed bases don't. If the next twelve months teach the market anything new, it will be about the second clause — and the tuition, either way, will be paid in the multiple.

The machine still prints. The courts convene in October. The engineer takes over in September. The scissors, for now, stay closed.

Sources: Apple Newsroom (quarterly results, April 30, 2026; Ternus succession, April 20, 2026; U.S. investment program, August 2025; Giannandrea retirement, December 2025); SEC filings (FY2025 10-K, FY2026 10-Qs); CNBC (Q1 FY26 earnings, January 29, 2026; Q2 FY26 earnings, April 30, 2026; Gemini-Siri agreement, January 12, 2026; Google antitrust remedies, December 5, 2025; Apple delays Siri, March 7, 2025; $4 trillion milestone, October 28, 2025); Bloomberg (Rockwell Siri reassignment, March 20, 2025; Pang departure, July 2025; India production share, March 2026; Meta glasses capacity, January 2026); Counterpoint Research (global and China smartphone share, Q1 2026; premium segment H1 2025; best-selling models 2026); IDC (Q1 2026 shipments); MacRumors (iPhone 17 production cuts, July 1, 2026; Gemini-powered Siri, April 2026; Epic case coverage; China Apple Intelligence timeline); 9to5Mac (Q2 FY26 earnings; Apple glasses roadmap, May 31, 2026; Google Safari-deal appeal, May 2026); AppleInsider (Microsoft 1997 investment retrospective; Kuo on Baltra, January 2026; iPhone 18 modem reports, June 2026); Ninth Circuit opinion (Epic v. Apple, December 11, 2025); Supreme Court docket (certiorari granted, July 2, 2026); European Commission (DMA decisions); Fortune (Cook succession, April 20, 2026); Forbes/GuruFocus/Seeking Alpha (Berkshire 13F tracking through Q1 2026); UBS Evidence Lab and MacTech upgrade surveys (July 2026); Yahoo Finance (market data as of July 3, 2026); TechCrunch (Titan cancellation, February 27, 2024); The Register/Tom's Guide (Vision Pro shipment estimates); Axios (OpenAI device timeline, January 2026); TIKR/Barchart/Motley Fool (valuation and performance comparisons); Variety (services revenue coverage). Figures reflect company disclosures where available and analyst/researcher estimates where noted; estimates are identified as such in the text.

This report is produced by Aya Invest (invest.aya-ai.org), an AI-assisted equity research project. For information and research purposes only. Not investment advice.

FAQ

Why doesn't Apple join the AI capex arms race?

Apple spends about $13B a year on capex while Amazon, Alphabet, Meta and Microsoft will spend a combined $660-690B in 2026 — roughly fifty to one. Its strategy: rent frontier intelligence (a custom 1.2-trillion-parameter Gemini model for ~$1B/year), run inference on its own device silicon and a small Private Cloud Compute footprint built on its Baltra chip, and let others build the utility layer beneath experiences it distributes. Bulls call it discipline — in every prior platform era the money went to whoever owned the customer, not the infrastructure. Bears call it harvesting: maximum monetization of an installed base, minimum investment in whatever comes next.

What is the Apple-Google Gemini deal for Siri?

Announced January 12, 2026: a multi-year agreement under which Google supplies a custom 1.2-trillion-parameter mixture-of-experts Gemini model — about eight times the scale of Apple's own 150B cloud model — to power the rebuilt Siri, for roughly $1 billion a year. The model runs entirely on Apple's Private Cloud Compute servers; Google never touches user data. The rebuilt "Siri AI" was unveiled at WWDC on June 8, 2026 and ships with iOS 27 this fall, with Apple claiming a 92% multi-step task success rate versus the old Siri's 58%.

Is Apple's 37x P/E justified?

The bull decomposition: a ~$124B/year services annuity at 76.7% gross margin growing mid-teens, 62% of the premium handset segment, $101B of trailing free cash flow with the group's cleanest conversion, a buyback retiring 2-3% of the float annually, and zero exposure to the AI capex cycle's downside. The bear decomposition of the same multiple: App Store pricing before the Supreme Court in October 2026, a ~$20B Google payment renewable annually under appeal, a China franchise both booming and hostage, memory costs guided "significantly higher" from June, a CEO transition in September, and an AI brain leased from a rival. The report's verdict: 37x priced for stasis, while all six components of the stasis are in motion.

© Aya Research · About · invest.aya-ai.org · For information and research purposes only. Not investment advice.