The Control Group

Four of the Mag 7 now spend $1–2 on AI hardware for every dollar they earn, on depreciation clocks that flatter today's profits. Apple spends seven cents — and its capex is falling. We value the only untouched income statement in the group, and ask what it proves: a scarcity premium, or a ceded future.

TON618 Capital Research · As of July 31, 2026 · Report date 2026-07-31 · Data as-of 2026-07-30 close · Download PDF

Every filed figure is drawn from Apple's own SEC filings and is as-of dated; model figures are labelled as estimates and separated from filed fact throughout. Prices are as of the July 31, 2026 close — the first full session after the print. This note is a satellite of The Depreciation Reckoning series; §8 documents sources, formulas and the model's disclosed limitations.

Executive summary

Fair value $208 per share against the $308.91 close of 31 July 2026 — a 33% gap. Bull $271, base $208, bear $144 at 25/50/25; 12–18 month evaluation window.

• The quarter was exceptional. A record June quarter: revenue $109.4bn (+16%), EPS $2.02 (+29%), every product line and every geography growing.
• And it was delivered with capital spending falling. Nine-month capex went $9.5bn to $6.8bn — a dose of $0.07 per dollar of net income, against $0.87 to $2.34 at the four hyperscalers.
• The control experiment worked: the re-mark rounds to zero. Running the house three-year clock over Apple's own capex moves FY2026 EPS by $0.002 — six cents of comps value, against $82 at Meta. Apple's earnings are the cleanest in the Mag 7.
• The price is the entire argument. At ~35× our FY2026 estimate, with a September guide decelerating to 9–11% and gross margin walking down under a memory-cost shock, even the bull case sits 12% below the close.

Where to find the work: the control-group re-mark is §3, the AI strategy §4, the valuation §5, the case against us §6.

The story so far

On 17 July we published The Depreciation Question, which found that one ratio — capex divided by net income — ranks how overstated Big Tech's earnings are. On 30 July, The First Invoice scored it against the Q2 prints: depreciation accelerating at all four hyperscalers, doses rising, every clock held. Apple was deliberately excluded from that work, and that omission is this note's subject: it is the control group. Apple reported FQ3 after the close on 30 July, filed its 10-Q on 31 July, and the shares fell 7.4% to $308.91 — on the September guide and a memory-price shock, not on anything in the depreciation debate.

§1 The ledger row with nothing in it

One ratio carries the series: capital expenditure divided by net income — the dose. It measures how much a company spends on machines per dollar it earns, and in our July research it ranked the earnings distortion across Big Tech almost perfectly.

Apple was left out of that work deliberately. It was excluded because its accounting choice cannot matter: at seven cents of capex per dollar of income, its depreciation schedule could halve or double and earnings would barely notice. In an experiment, that is what a control group is for. The control ran its first trial on July 30 and behaved exactly as the framework predicts: a record quarter, printed while capex fell, with depreciation growing 16% against Microsoft's 56%. The dose-response curve holds at the origin.

Exhibit 1 — The empty entry

Source: Apple FQ3-2026 Form 10-Q (0000320193-26-000020); Microsoft FY2026 10-K; Meta Q2-26 8-K Ex-99.1; Alphabet Q2-26 10-Q; Amazon Q2-26 8-K Ex-99.1/99.2. Alphabet and Amazon net income ex disclosed non-operating investment gains at the 21% statutory rate (estimates). Overstatement percentages: TON618 July v3 model, FY2025 basis. Data as of 2026-07-31.

But a control group answers a narrow question. It proves the mechanism — it does not tell you what the control stock is worth. Apple's cleanliness admits two opposite readings. Reading one: scarcity premium — the only Mag 7 income statement with no depreciation wave baked into 2027–28 deserves to trade dear, because its EPS means what it says. Reading two: ceded future — the reason Apple has no AI infrastructure bill is that it is not building AI infrastructure; it is renting intelligence from the landlords next door, and a company that rents its most important input has quietly traded its future margin for today's clean ledger. This note makes both cases with the strongest evidence we can find (§4), and then lets the arithmetic decide (§5). We commit to the answer wherever it lands.

§2 What the quarter actually said

Data shows a genuinely exceptional quarter. Revenue of $109.4bn, up 16% — Apple's strongest June quarter ever, and its fastest growth in years — beat consensus (~$108.7bn, press-sourced). iPhone revenue was $54.3bn, up 22%, on strength in Pro models; Mac rose 29%; every geographic segment grew double digits, including Greater China at +22%. Services grew 12% to $30.7bn — a June-quarter record, though short of the ~$31.2bn consensus, and the growth drivers the 10-Q names are advertising and cloud services. Diluted EPS of $2.02 rose 29%, helped by roughly $0.11 of tariff refunds after the Supreme Court's February ruling striking down certain IEEPA tariffs; we treat that transparently as one-time (§5). Nine-month operating cash flow was $117.0bn — up 43% — against $101.5bn of net income: every dollar of reported profit came in as cash, with room to spare.

Data also shows the two clouds management itself put over the print. First, costs: gross margin of 50.1% includes about two points of tariff refunds — strip them and the adjusted margin was 48.1%, down from 49.3% in the March quarter (call figures; press-sourced). The culprit is memory. CEO Tim Cook — on what he announced was his final earnings call — described a “hundred-year flood” in memory pricing: Apple paid more for DRAM and NAND in each sequential quarter this year, expects to pay still more in the September quarter, and has “reluctantly” raised Mac and iPad prices in response. The 10-Q converts the anecdote to policy: supply constraints and rising component costs, and — the company's words — “the Company expects these trends to intensify.” Second, the guide: September-quarter revenue growth of 9–11% (against a ~12% consensus), including up to 2.5 points of currency headwind, with gross margin guided to 47–48% — including another point of tariff refunds, which puts the ex-refund margin near 46.5%. On the CFO's own framing, memory costs account for more than the entire slide in adjusted gross margin from March through the September guide. The market's answer was immediate: the stock, which closed at $333.43 before the release — within 3% of its all-time high, days after briefly crossing a $5 trillion market value — fell 7.4% the next session to $308.91, taking roughly $370bn of market value with it (guide and margin-path figures press-sourced from named carriers; the price is the official close, §8).

The capital-return machine, meanwhile, ran at full throttle. In the quarter: $25.8bn of buybacks and $4.0bn of dividends. Over nine months: $61.8bn repurchasing 215 million shares (~$287 average; the cash-flow statement's $62.1bn includes settlement timing), $11.8bn of dividends, an $8.1bn term-debt paydown and the final $8.8bn instalment of the 2017 repatriation tax. A new $100bn authorization was announced April 30 and a $10bn accelerated repurchase signed in May; $138bn of authorization remains. The diluted share count is down 1.6% in a year. Nine-month capex, for comparison: $6.8bn. Apple spent nine dollars buying its own shares for every dollar it spent on plant and equipment.

And a governance milestone the record should carry: July 30 was Tim Cook's last earnings call as CEO. John Ternus, Apple's hardware-engineering chief for the past five years, becomes CEO on September 1, with Cook moving to Executive Chairman (press-sourced, multiple carriers). The next print — and the first year of the AI strategy this note examines — belongs to new management.

§3 We ran the clock on Apple. It rounds to zero.

Everything in this section that is not from a filing is a model estimate, and is labelled as such.

The house exercise in this series re-marks each company's earnings from its chosen depreciation schedule onto a three-year server clock — our estimate of the economic life of AI-cycle compute, argued in July from NVIDIA's two-year architecture cadence. At Meta the re-mark removes 9% of FY2025 net income; at Oracle, 17% (rising to 26% in Oracle's fiscal 2026). Here is Apple's, computed with the same vintage methodology on Apple's own filed capex:

The result is not small because we chose a friendly assumption — we gave the re-mark its maximum dose by treating all of Apple's capex as short-lived compute. It rounds to zero for two reasons worth stating precisely, because together they are the cleanest statement of what this series is about:

• The dose is roughly thirty-fold below the worst of the hyperscalers'. Apple's nine-month capex of $6.8bn stands against $101.5bn of net income: 0.07×. Meta's June quarter alone ran 1.96×. Whatever schedule Apple picks, the flattery available is proportional to the spend, and the spend is a rounding error against the earnings.
• Apple's capex is flat — and a depreciation schedule can only flatter earnings while spend is growing. Apple has spent between $9.4bn and $12.7bn a year for six consecutive years (Exhibit 2). On flat spend, the vintage stack reaches steady state and annual depreciation equals annual capex no matter what life is assumed — a longer clock changes when each machine's cost lands, but with equal-sized vintages the overlaps cancel. The distortion our July note measured at the hyperscalers is not a product of long clocks alone; it is long clocks times capex growth. Microsoft grew capex 80% last year. Apple shrank it.

Exhibit 2 — The flat line

Source: left — Apple Forms 10-K, FY2021–FY2025 (payments for acquisition of PP&E, cash-flow statements; FY21–22 per the FY2023 10-K, 0000320193-23-000106); FY26E = 9M actual per the FQ3 10-Q plus house Q4 estimate. Right — latest reported depreciation growth per each company's filing, windows as labelled; Apple = 9M D&A per the FQ3 10-Q. Data as of 2026-07-31.

Three details from the filings complete the picture. First, Apple's own depreciation is so unimportant to its income statement that the FY2025 10-K discloses no useful lives at all — the entire policy is one sentence (“Depreciation on property, plant and equipment is recognized on a straight-line basis”), a disclosure posture none of the four hyperscalers could get away with. Second, PP&E depreciation expense has declined three years running — $8.5bn, $8.2bn, $8.0bn — while Microsoft's went $15.2bn → $22.0bn → $34.3bn. Third, the FQ3 10-Q, filed July 31, discloses no change in any depreciation estimate — under GAAP a change must be disclosed in the quarter it occurs, so the schedules held. That closes one of the three open policy checks our scoreboard carried, and it closes it the uneventful way a control group should.

The verdict of the control experiment is unambiguous: earnings quality is not Apple's problem. Its reported EPS is the closest thing to economic EPS in the Mag 7 — cash-covered, un-deferred, with no 2027–28 depreciation wave baked in anyone's spreadsheet. What remains is the question the experiment cannot answer: whether the reason for the clean ledger — renting the AI future rather than building it — is a strategy or an abdication. That requires evidence of what Apple is actually doing.

§4 Renting intelligence: what asset-light AI actually costs

The strategy is now explicit. At WWDC in June, on Tim Cook's final keynote, Apple introduced Siri AI — the largest overhaul of its assistant since 2011 — and named its supplier: software chief Craig Federighi said Apple has “partnered deeply with Google to leverage the technology behind the Gemini family of models to co-create the next generation of Apple Foundation Models.” Press coverage widely describes cloud queries running a custom Gemini-derived model inside Apple's Private Cloud Compute; Apple's own technical staff describe the arrangement more narrowly — models “refined using outputs from Gemini frontier models,” with Federighi insisting “the amount of Google Assistant we use is none” — and both framings coexist in the record. On-device tasks run Apple's own smaller models; the arrangement is reported at roughly $1bn a year and non-exclusive (all press-sourced, multiple named carriers; Apple's own 10-Q confirms the launch: “During the third quarter of 2026, the Company... introduced Siri AI”). Either way the strategic fact this note prices is undisputed: the frontier layer is rented, not owned. Siri AI ships in September with the new iPhones.

Now put the rent against the peers' mortgages. The four hyperscalers will spend roughly $630bn on capital expenditure this calendar year (July model basis, corroborated since by Microsoft's filed $115.9bn, Alphabet's ~$161bn half-year pace, Meta's $130–145bn guide and Amazon's $169bn trailing spend). Apple's reported price for frontier-model access is about $1bn a year — less than 1% of what any single hyperscaler spends building it, less than Apple pays out in dividends every five weeks. If intelligence is becoming a commodity input produced by four competing landlords with unprecedented capacity coming online, then renting it — on non-exclusive terms, with the supplier swappable — is not an abdication; it is the correct side of an overbuilt market. That is the strongest form of the scarcity-premium case, and the market's $5tn flirtation in July suggests it is the case the market believes.

But the filings show the rent bill is bigger than the headline, and growing where clocks don't reach. Apple's AI spending doesn't route through capex — it routes through the income statement, and this quarter it is visible in three places:

Exhibit 3 — Where the AI bill actually lands

Source: Apple FQ3-2026 Form 10-Q (0000320193-26-000020, filed 2026-07-31): cash-flow statement (capex), income statement and MD&A (R&D and its stated driver), Note 5 (intangible assets; Sep-2025 comparatives re-presented with the new intangibles line). Data as of 2026-07-31.

• R&D rose 33% in nine months, to $34.0bn — now 11% of quarterly revenue, the highest in Apple's modern history — and the 10-Q names the driver: “higher infrastructure-related costs, including investments in artificial intelligence, and headcount-related expenses.” Renting compute is opex. Note the earnings-quality irony: this is the cleanest possible AI accounting — fully expensed the quarter it is incurred, deferring nothing — the exact opposite of a stretched server schedule.
• Gross intangible assets jumped $13.3bn — 53% — in nine months, to $38.2bn, large enough that Apple broke intangibles out as a new balance-sheet line for the first time. There is no business-combination note; $5.1bn of the balance is classified current, implying short-lived assets amortizing quickly. The 10-Q does not itemize further. We read this as capitalized licensed IP and content — the categories Apple's own purchase-obligation note lists ($29.3bn of “supplier arrangements, licensed intellectual property and content, distribution rights”) — and we flag its composition as the single most interesting open item for the FY2026 10-K. If a multi-year AI licence sits in there, Apple's AI bill is being amortized, just off a different shelf than its peers'.
• A new risk factor says the quiet part. The 10-Q now warns that Apple's business “including its artificial intelligence and machine learning offerings... depends on access to sufficient computing resources,” that industry demand has left compute “constrained,” and that “in addition to its own data center infrastructure, the Company relies on third-party cloud service providers to meet these compute needs, and the Company may be unable to secure sufficient capacity on commercially reasonable terms, or at all.” A renter's clause, in Apple's own words: the roof over Siri belongs to someone else, and the rent can go up.

Is Apple quietly turning into a builder? The evidence says: at the margin, yes; at hyperscaler scale, no. Press reporting around the print puts Apple's full-year FY2026 capex at $13–14bn including expanded Private Cloud Compute capacity on Apple's own silicon, with Apple evaluating NVIDIA, Broadcom and Google TPU hardware for training (press-sourced, labelled; the filed nine-month figure is $6.8bn, down 28%, so any build is heavily back-loaded and modest). Even taking the press number at face value, the dose ends FY2026 near 0.10× — the empty entry stays empty. Private Cloud Compute on Apple silicon is inference at the edge of the network, not frontier training; it is capital-light by design.

The ceded-future case, at full strength — because it deserves it. First, the landlord is also the rival: Apple's cloud intelligence now runs on the models of the company whose $20bn-plus annual search payments to Apple are under active antitrust remedy — the 10-Q itself warns that a reversal of the DOJ-v.-Google order on appeal could revive remedies “prohibiting Google from offering the Company commercial terms for search distribution,” which would “materially adversely affect” that licensing revenue. Apple's two most important Google dependencies — the money coming in for search, and the intelligence coming in for Siri — can both be repriced by courts or by Mountain View. Second, renters don't compound: Meta's counter-argument to our own series is that owned compute earns “a significant premium” sold to others; if that is right, the four landlords are building the margin pool of the next decade and Apple is a customer in it. Third, the endpoint premium is an assumption, not a law: Apple's moat is that intelligence must reach users through devices Apple sells. If agents and AI-native hardware commoditize the handset interface even partially, the toll booth loses traffic — and Apple will not own the layer that replaced it. The clean ledger is real; the question is whether it is the balance sheet of discipline or of a company that opted out of the fight.

Where the arithmetic lands between the two readings: the scarcity premium is earned — the earnings are clean, the dose is durable near-term, and the strategy has a coherent economic logic — but it is also already paid, roughly twice over. That is §5.

§5 What the cleanest income statement is worth

Method selection. Apple is a mature, dividend-paying equity with a stable capital structure whose defining behaviour is the conversion of essentially all free cash flow into shareholder distributions — $77.5bn of buybacks and dividends in nine months. We value the equity directly with a multi-stage FCFE model (60% weight) and cross-check with relative multiples on forward earnings (40%). A dividend-discount model would be the textbook alternative for a dividend payer, but Apple's dividend is one-sixth of its total distribution; FCFE captures the whole channel. A sum-of-the-parts adds nothing — Apple reports one integrated business — and we say so rather than manufacture a third method.

House conventions, stated up front. Stock-based compensation ($10.5bn expensed in nine months; $26.7bn unrecognized) is treated as the real expense it is — never added back to cash flow. In the FCFE leg the diluted count (14,715M) is held flat: the buybacks are the distribution of the FCFE we discount, so shrinking the count as well would double-count them. In the comps leg, where EPS is the object, we model the shrink explicitly at 2.4% a year — the pace implied by the current program net of RSU issuance. Cost of equity is CAPM with nothing softened: 4.67% ten-year Treasury (July 29, FRED), beta 1.08 (computed, two years of daily closes against SPY), 4.5% equity risk premium — 9.53%. Tax 17.5% (the nine-month effective rate, which is rising). Terminal growth 3.5%.

The FY2026 base year is built from the guide, not hope: nine months actual plus a September quarter at the 9–11% guide midpoint with a 47.5% gross margin gives revenue of $477.1bn (+14.6%), net income of $130.0bn, and EPS of ~$8.83 — about $8.66 normalised for the disclosed tariff refunds ($0.11 booked in FQ3 plus ~$0.06 embedded in the FQ4 margin guide). At $308.91 the market pays 35.0× that number — a multiple Apple reached, for reference, at no point between 2007 and 2020.

Probability-weighted FCFE value: $182. Note what the bull case is: six more years of above-consensus growth, record margins, and the same 0.07× capital intensity — and the FCFE engine still cannot reach today's price, because at a 9.5% discount rate a $4.5tn valuation demands cash flows Apple's scale arithmetic cannot produce.

The comps lens. FY2027E EPS across the scenarios is $8.55 / $9.46 / $10.22 (share shrink modeled). We apply 20× / 26× / 32× — the floor, middle and ceiling of Apple's own post-2020 multiple band, with the base a premium to both the S&P (~23×) and Microsoft (25×) in recognition of exactly the earnings-quality scarcity this note documents. That yields $171 / $246 / $327, probability-weighted $247. The stock currently trades at 35× the base FY2027 estimate: the market is paying our bull multiple on our bull-ish earnings today.

And here is the series' own yardstick applied to itself. In our Meta note, applying the house clock moved the comps value by $82 per share — the multiple was only defensible if you believed the E. Apple's re-mark (§3) moves FY2027E EPS by $0.002 and the comps value by six cents. There is no earnings-quality discount to argue about here. Which is precisely why the conclusion is different in kind: at Meta the debate is about the E; at Apple the E is the best in the group and the entire debate is the P.

Exhibit 4 — The football field

Source: TON618 analysis; model.py in the build folder reproduces every figure from the Phase-0 snapshot. Data as of 2026-07-31.

The payoff shape, from $308.91:

We believe the shape is the finding: every branch, including the bull, sits below the price. In our Meta note the bull case at least round-tripped to the close; here it does not. When the most favourable scenario a disciplined model can defend still implies a negative return, the position embedded in the market price is not a scenario — it is a different discount rate, or a permanently higher multiple.

The reverse question, honestly posed. Work backwards from $308.91 through the base engine, one variable at a time, and the price requires: a 16.6% six-year revenue CAGR (base: 5.8% — Apple has not compounded revenue at 16% over six years since it was a fraction of today's size), or a 7.05% cost of equity (an equity risk premium under 2.5%), or 6.4% perpetual growth, or 24 points of additional EBITDA margin. In our Alphabet note, softening three Street conventions reproduced the market price almost exactly; we said so. Here they do not: even discounting at a generous 8.5% cost of equity lifts the blend only to roughly $230; reaching the $250s takes 8.0%. Apple's price needs the conventions and genuine optimism — a Siri-AI supercycle priced as permanent, or the index-weight bid treated as a fundamental. We can name those beliefs; we cannot underwrite them.

Sensitivity, honestly displayed. Each 50bp of cost of equity is worth roughly $17 of FCFE value ($220 at 8.5%, $158 at 10.5%); terminal growth 2.5%→4.5% spans $163–$210; ±2pp of through-cycle EBITDA margin spans $172–$193. No plausible single-input flex closes a $125 gap to the close — the disagreement with the market is structural, and we prefer to state that plainly rather than tune the model to the tape.

§6 The case against this ledger

Likelihood and magnitude scored 1–5 over 12 months; score is the product.

Market risk: 30-day realised volatility is 32.8% annualised (beta 1.08); a one-day 95% VaR near 3.4% and ±5% moves around prints must be assumed in sizing. Governance: single-class shares, no controller — capital allocation answers to shareholders, and the $138bn standing authorization says what the answer will keep being.

The steelmanned bull case, at full strength. First, the quarter was a refutation of the decay narrative: +16% at a $477bn run-rate, with every segment and every geography contributing, is the broadest growth Apple has printed in years — and the AI phones haven't even shipped. Second, the EPS algorithm is unmatched: clean earnings, ~100% cash conversion, and a 2.4% annual share shrink mean high-single-digit EPS growth requires only mid-single-digit revenue growth — forever, at 0.07× capital intensity. Third, every dollar of the $630bn buildout needs a distribution endpoint, and Apple owns the world's most valuable one; the market's willingness to pay $5tn in July is the market pricing the toll booth, not the tolls. Fourth, renting is optionality: non-exclusive terms in a world of four competing frontier landlords mean Apple buys intelligence at marginal cost while its rivals fight a capex war — if AI compute overbuilds, the renter wins outright. If the September launch turns the installed base over at +12% for two years with Services reaccelerating, our bull engine is the right one — and we concede now that even it does not justify $333, which is why our disagreement survives good news about the business.

What would change our mind, with dates.

Levels, against the $208 fair value: above $345 (the all-time high) the gap to fair value is at its widest; at the $308.91 close the valuation argument is fully against the price; near $250 it is largely exhausted; at $156 (a 25% margin of safety to fair value) the arithmetic turns constructive if the operating KPIs are intact. We would concede outright on two consecutive quarters of 12%+ revenue growth with ex-refund gross margin at or above 48% and Services growth at or above 16% — in that world the bull engine understates the machine and the multiple debate reopens. The Fund holds no position in AAPL, and this note does not recommend initiating a short: at 33% realised volatility, against the deepest passive bid in the market and a $138bn standing buyback, being early on a multiple argument is expensive carry in both directions.

§7 Where this note sits in the series

This is the Apple satellite of The Depreciation Question (July 17) and its Q2 scoreboard The First Invoice (July 30). In that framework Apple is the control group: the Mag 7 name whose dose (0.07×) sits at the origin of the distortion curve, proving the dose-response mechanism by its absence — the July model's re-mark, applied here with the same machinery, moves Apple's EPS by two-tenths of a cent. The satellites value the treated group: Microsoft (“The $116 Billion Question”), Meta (“Paying Twice”), Alphabet (“The Benchmark”), Amazon (“The Honest Machine”). Apple's verdict differs from theirs in kind, and the difference is the series thesis in one sentence: at the four, the question is whether the earnings are real; at Apple the earnings are real and the question is only what they are worth. The series re-scores itself every earnings season.

§8 Sources & Method

Every filed figure is drawn from Apple's SEC filings via EDGAR, pulled and verified on July 31, 2026; market data from verified end-of-day sources on the same date. The full Phase-0 capture, with per-feed timestamps, content hashes and the failure log, is archived in the snapshot manifest (ton618-tech-valuation/data/AAPL/2026-07-31-055857).

Filings and issuer documents relied upon. FQ3-2026 Form 8-K Exhibit 99.1, filed 30 Jul 2026 (0000320193-26-000018; quarter ended 27 Jun 2026) — quarterly and nine-month statements, segment and category revenue, balance sheet, cash flows, the Cook and Parekh result quotations, the tariff-refund quantifications (~2pp of gross margin; $0.11 of EPS), and the dividend declaration. FQ3-2026 Form 10-Q, filed 31 Jul 2026 (0000320193-26-000020) — PP&E and intangibles detail (Note 5), the absence of any change-in-estimate disclosure, share-repurchase and ASR detail (Note/Item 2: 215M shares for $61.8bn in nine months; May 2026 ASRs, $10.0bn), SBC detail ($26.7bn unrecognized), purchase obligations ($57.0bn manufacturing; $29.3bn other incl. licensed IP and content), the R&D driver language, the third-party-compute and DOJ-v.-Google risk-factor language quoted in §4, tariff/memory MD&A, and the Siri AI introduction. FY2025 Form 10-K, filed 31 Oct 2025 (0000320193-25-000079) — FY2023–25 statements, the one-sentence depreciation policy, the PP&E note (gross $125.8bn; PP&E depreciation expense $8.0/$8.2/$8.5bn for FY25/24/23), capital-return and share-count history. FY2023 Form 10-K, filed 3 Nov 2023 (0000320193-23-000106) — FY2021–22 capex for the vintage stacks. Peer figures: Microsoft FY2026 10-K (0001193125-26-323660); Meta Q2-26 8-K Ex-99.1 (0001628280-26-050596); Alphabet Q2-26 10-Q (0001652044-26-000071); Amazon Q2-26 8-K Ex-99.1/99.2 (0001018724-26-000024) — as compiled in our scoreboard note, same accessions.

Market data and rates. Prices are adjusted daily closes from Massive Market Data: anchor close $308.91 (31 Jul 2026), the official regular-session close on the first full day after the print, cross-checked against the Schwab Market Data API (net change −$24.52, −7.4%, from the $333.43 close of 30 Jul). 52-week intraday range $201.50–$344.57; year-to-date +13.6% from the $271.86 year-end close; market capitalisation ~$4.51tn on 14,609M shares outstanding (10-Q). Beta 1.08 computed from two years of daily closes against SPY; 30-day realised volatility 32.8% annualised, same series. Risk-free rate 4.67%, ten-year constant-maturity Treasury (FRED DGS10, 29 Jul 2026). Peer closes (30 Jul): MSFT $451.10, GOOGL $333.66, META $539.03; Amazon is marked at its own 31 Jul close of $271.58 in the series comparison.

Press-sourced items, labelled as such throughout. September-quarter guidance (revenue +9–11%; gross margin 47–48% incl. ~1pp tariff refunds; ~2.5pp FX), the adjusted-gross-margin path (49.3% → 48.1% → ~46.5%) and the Parekh memory attribution, Cook's “hundred-year flood” language and price-increase comments, consensus figures, the CEO transition (Ternus effective 1 Sep 2026), the Siri AI / Gemini arrangement (~$1bn/yr reported; Federighi keynote quotation), and the FY2026 capex estimate ($13–14bn) — from Yahoo Finance/GuruFocus call recaps, Fortune, SiliconANGLE, CNBC, 9to5Mac, MacRumors, MacObserver, Business Standard and TradingKey (29–31 Jul 2026). Apple posts a webcast replay at apple.com/investor but publishes no transcript, so call quotations cannot be verified against an issuer document; guidance figures enter the model only through the FY2026E build, as stated. Consensus figures are context, never inputs.

Formulas and conventions. Dose = capex ÷ net income, same window (Apple 9M FY2026: 6.799/101.464 = 0.067; FY2025: 12.715/112.010 = 0.114); Alphabet and Amazon doses ex disclosed non-operating investment gains at the 21% statutory rate, per the scoreboard's method — Amazon's gain is the one its release attributes “primarily” to its investments in Anthropic. Re-mark = half-year straight-line vintage stacks on filed FY2021–26 capex, five-year policy clock versus the house three-year clock, 100% of capex treated as short-lived (the maximizing assumption); FY2026 delta $0.04bn pre-tax, taxed at 17.5%. FY2026E = nine-month actuals + September quarter at the guide midpoint (revenue +10% on FY2025's derived Q4 of $102.5bn; gross margin 47.5%; opex +20%; tax 17.5%). Normalised EPS deducts disclosed tariff-refund benefits only ($0.11 FQ3 + ~$0.06 in the FQ4 guide). FCFE = net income + D&A − capex (ΔNWC ~0; net borrowing ~0 against $146.5bn of cash and securities vs $84.3bn of debt), discounted at Ke 9.53% (CAPM: 4.67% + 1.08 × 4.5%), fiscal-year mid-year convention from a 31 Jul 2026 valuation date, terminal Gordon growth 3.5% (bear 3.0%, bull 4.0%). SBC never added back; diluted count 14,715M held flat in the FCFE leg; comps EPS uses a 2.4%/yr net share shrink. Blend = 60% FCFE + 40% comps, by scenario, probability-weighted 25/50/25.

Limitations, stated. (1) The re-mark's five-year policy clock is our assumption — Apple discloses no useful lives; its last disclosed machinery range (pre-FY2022 filings) topped out at five years. The conclusion is insensitive: on Apple's flat capex the delta is ~$0.1bn under any policy life from four to seven years. (2) The FY2026E fourth quarter is built from call guidance that exists only in press accounts, since Apple files no guidance; the revenue guide midpoint ±1pp is worth about ±$0.9bn of net income. (3) The composition of the $13.3bn intangibles addition is not disclosed; our licensed-IP reading is an inference and is flagged as the FY2026 10-K's key open item. (4) The 4.5% equity risk premium is an analyst assumption with high leverage on the output; the sensitivity table prices the disagreement, and no defensible ERP closes the gap to the market price. (5) The anchor is the first post-print close, so it already contains the market's reaction to the September guide; against the pre-print $333.43 every implied return in this note would be roughly five points wider, and no conclusion changes. Each chart carries its source beneath it.

§9 Disclosures

Information only. TON618 Capital. This report is for information purposes only. Nothing here is an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and nothing here is personalized investment advice or a recommendation regarding any instrument.

Publisher's exclusion. All research is published solely as general, impersonal information of regular circulation. It is not tailored to the objectives or circumstances of any individual and is not issued in connection with compensation from any client. The Fund has no clients and distributes all research free of charge. On that basis it publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)).

Registration & conflicts. TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity. The Fund is a Bitcoin fund and may hold or transact in the securities or digital assets it discusses. This report discusses Apple (AAPL), Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), Meta Platforms (META), Oracle (ORCL) and NVIDIA (NVDA); the Fund holds no position, long or short, in any of them, and has no economic interest in the price of any security named here. The Fund receives no compensation from any party in connection with its research.

Use of AI. Artificial intelligence is used in the creation of this research. The AI used is developed by Anthropic; this note adjusts Amazon's dose ratio for Amazon's disclosed gain on its investments in Anthropic, and readers should weigh that connection — the Fund has no economic interest in Anthropic and the figure is quoted directly from Amazon's release. All methodology and data integrity are reviewed and approved before publication by TON618 Capital's Chief Investment Officer, Keyth Beck; errors may nonetheless occur, and readers should verify independently.

CFA. This report was prepared to align with CFA Institute analytical standards (methodology only). CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute. That reference describes the analytical framework applied; it does not imply the report was prepared, reviewed, or authored by a CFA charterholder, and the report is not issued, reviewed, endorsed, certified, or approved by — nor affiliated with — CFA Institute.

Risk & feedback. Past performance is not indicative of future results. Digital assets and equities are volatile and may result in total loss of capital. Corrections and feedback are welcome — please direct them to CIO Keyth Beck at keyth@ton618capital.com.