The Honest Machine

Amazon is the only company in Big Tech that ever admitted its servers wear out faster — and paid double its own estimate for saying so. This quarter the honesty got expensive: AWS grew its fastest in eighteen quarters, and trailing free cash flow went negative anyway. We value the two machines with the clock told straight.

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 Amazon's own SEC filings and is as-of dated; model figures are labelled as estimates and separated from filed fact throughout. Earnings figures marked "ex-gain" strip the $53.4bn Anthropic-related gain disclosed in the Q2 release (method in §2 and §7); the maker of the AI used to prepare this note is Anthropic, and that conflict is flagged where the gain is discussed and in §8. 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; §7 documents sources, formulas and limitations.

Executive summary

Fair value $142 per share against the $271.58 close of 31 July 2026 — a 48% gap. On the reported-schedule basis $158; bull blend $201; 12–18 month evaluation window.

• The boom is real and the market agreed. AWS grew 37%, its fastest in eighteen quarters; operating income rose 43%; the stock rose 15.3% on the print to within 3% of its 52-week high.
• The bill is in the same filing. Trailing capex $169bn (+64%), trailing free cash flow −$7.6bn against +$18.2bn a year ago, capex guidance raised to ~$220bn, long-term debt doubled in six months.
• Amazon is the only company that ever shortened a server clock — and is still the most distorted of the four. The 2025 cut cost double its own guide; the house re-mark still takes FY2025 EPS $7.17 to $6.56, because the dose rules: 2.34×, the highest in Big Tech.
• Strip the mark that is not the machine. $53.4bn of the headline $62.6bn of Q2 net income is a pre-tax gain on Anthropic — the maker of the AI used to prepare this note (§8) — and we remove it everywhere.

Where to find the work: the re-mark is §3, the two-machine valuation §4, the case against us §5.

The story so far

On 17 July we published The Depreciation Question — a record hardware build expensed on schedules that flatter earnings, roughly $221bn deferred out of 2026–28. On 30 July, The First Invoice scored it against the Q2 prints: four-for-four on accelerating depreciation, every clock held. Amazon's chapter role: the honest machine — the only company of the six that ever shortened a server life, and the one whose honesty has proved most expensive. It reported Q2 after the close on 30 July and filed its 10-Q on 31 July, confirming both that the shortened clock held and that segment depreciation grew 42%, with AWS alone at 67%.

§1 Two machines, one invoice

Amazon is two businesses wearing one ticker. The first is a store: the retail operation (North America plus International), which over the last twelve months sold $627bn of goods and services and kept $39.0bn — about 6 cents on the dollar. The second is a utility: AWS, which rents computing the way a power company sells electricity, and over the same twelve months turned $148bn of sales into $54.7bn of operating income — 37 cents on the dollar. AWS is 19% of Amazon's revenue and 58% of its operating profit. When you buy the stock, you are mostly buying the utility; the store comes attached.

And one coincidence in this quarter's filings is the whole series in a single number: AWS's annualized revenue run rate is $169bn — and Amazon's trailing-twelve-month capital expenditure is $169bn. The invoice for building the utility has caught up with everything the utility currently sells.

The utility is booming. AWS grew 37% in the June quarter — its fastest in eighteen quarters, per the release — with the AI business and the in-house chips business each above a $25bn annualized run rate, both growing at triple-digit percentages year on year (the release's phrase). And utilities have one iron law: to sell more power, you must first build the plant. That is the invoice this series tracks. Amazon's trailing-twelve-month capital expenditure ran $88bn → $103bn → $116bn → $128bn → $147bn → $169bn across the last six quarters — up 64% in a year, an increase the release attributes, in its own words, "primarily... investments in artificial intelligence." On the earnings call, capex guidance for 2026 rose again, to roughly $220bn from ~$200bn (press-sourced; §7). The result sits in one line of the same filing: trailing free cash flow, +$18.2bn a year ago, −$7.6bn now. The best quarter AWS has had in four years did not cover the cost of building it.

Source: Amazon Q2-2026 Form 8-K Ex-99.1 (0001018724-26-000024, filed 2026-07-30), Supplemental Financial Information: TTM purchases of property and equipment net of proceeds, and TTM free cash flow as Amazon defines it (Ex-99.2). Data as of 2026-07-31.

One ratio carries this series: capex 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 runs 0.07× and its depreciation schedule is a rounding error. Amazon ran 1.70× in FY2025. On the trailing twelve months — with the Anthropic gain stripped — it now runs 2.34×, the highest of the four hyperscalers, walking toward Oracle's 3.26×, the ratio at which our July note found earnings 26% overstated. On our full-year 2026 estimates the dose reaches ~2.5×. When a company spends $2.34 for every $1 it earns, the depreciation schedule is not an accounting detail. It is the income statement.

§2 What the quarter actually said

Data shows a genuinely strong operating quarter. Net sales rose 20% to $200.6bn — North America +16% to $116.2bn, International +15% to $42.2bn, AWS +37% to $42.2bn. Operating income rose 43% to $27.5bn, a record 13.7% operating margin; AWS segment income rose 64% to $16.6bn at a 39.4% segment margin. Advertising grew 26% to $19.8bn. Third-quarter guidance calls for net sales of $197–202bn (+9–12%, with a Prime Day calendar shift the release quantifies at nearly 400 basis points of optical drag) and operating income of $22.5–26.5bn against $17.4bn a year ago. Stock-based compensation actually fell 8% in the quarter, to $6.0bn — a real credit in a series where peers' SBC is climbing. One flattering one-off to note against the record margin: the 10-Q discloses roughly $640m of IEEPA tariff refunds in the quarter, booked mostly as a reduction of cost of sales in North America.

The headline net income is another matter — and the reader deserves the split. Reported Q2 net income was $62.6bn ($5.75 per diluted share), versus $18.2bn ($1.68) a year ago. The release states it plainly: "Second quarter 2026 net income includes non-operating pre-tax other income of $53.4 billion, primarily from our investments in Anthropic." Anthropic is also the maker of the AI used to prepare this research — the conflict is disclosed in §8 and we flag it here, at first mention. We strip the gain everywhere: at the 21% statutory rate (the filing does not disclose the gain's own tax line; the H1 deferred-tax swing of $30.5bn indicates most of the associated tax is deferred), ex-gain net income is roughly $20.5bn, up 13% — a real number a fifth the size of the headline. Operating income, which contains no investment marks, is the clean line, and it is the one we quote: +43%. The distinction matters beyond hygiene: the mark is an asset that already exists on the balance sheet (we value it, once, in §4); it is not earnings power, and at 2.34× dose the earnings power is the thing the capex has to be measured against.

The funding of the build has changed shape fast. Six months ago Amazon carried $65.6bn of long-term debt; at June 30 it carried $128.9bn, after $67.0bn of new issuance in the half — and quarterly interest expense has more than doubled year on year ($1.31bn vs $0.52bn). Property and equipment on the balance sheet grew $89bn in six months, to $446bn. Another $20.6bn was added in the half to equipment bought but not yet paid for (the balance, ~$48bn by our arithmetic from the year-end $27.0bn plus the disclosed increase, is an estimate). Operating cash flow of $161.4bn TTM (+33%) is the genuine engine funding all of this — but for the first time in the modern history of the company, it is not enough: free cash flow is negative, and the marginal dollar of buildout is now borrowed.

The store, meanwhile, quietly improved. North America's operating margin held at 7.9% (TTM 7.4%, up from 7.0%), International reached 4.1%, delivery speeds set records, and advertising — the store's high-margin passenger — grew 26%. Nothing in this note argues the machine is broken. The argument is about what the machine's output costs, and what its reported earnings mean while the cost is running through the longest permissible pipes.

§3 The corrected entry: the only honest clock, re-marked anyway

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

Between 2020 and 2024, Amazon extended its server clock three times — and then did the thing no one else in Big Tech has ever done. Servers went from three years to four (effective January 2020), four to five plus networking five to six (January 2022 — worth a disclosed $3.6bn of depreciation relief and $2.8bn of net income, $0.28 per share, in 2022), five to six (January 2024, guided at $3.1bn of 2024 operating income; the FY2024 10-K put the realized effect at $3.2bn of depreciation relief and $2.5bn of net income — $0.23 per share). Each extension raised reported profit with no change in cash. Then, in its FY2024 10-K (February 2025), Amazon reversed: a Q4-2024 useful-life study concluded that a subset of servers and networking equipment should be cut from six years back to five, "due to the increased pace of technology development, particularly in the area of artificial intelligence and machine learning" — and the company took $920m of accelerated depreciation on servers it retired early, on top. Thirteen disclosed extensions across the group; one reversal. This is the reversal. Credit where due: it is the single most honest depreciation decision in this series, made against the company's own reported earnings, in the asset class where honesty costs the most.

Then the honesty cost double what it guided. The February 2025 guide anticipated a ~$0.7bn hit to 2025 operating income — computed, as the filing states, only on assets "included in 'Property and equipment, net' as of December 31, 2024." The FY2025 10-K records what actually happened: an increase in depreciation of $1.4 billion and a $1.0bn cut to net income, because the realized figure also captured assets "acquired during the year ended December 31, 2025" — and 2025 was the biggest server-buying year in Amazon's history ($172.5bn gross servers at year-end, up from $113.2bn). The mechanism is the series in miniature: Amazon kept buying the very assets it had just decided wear out faster, so the honest correction doubled between the estimate and the invoice. The same filing block, it should be recorded, also extended heavy warehouse equipment from ten years to thirteen (+$0.9bn) — lives lengthen where the estimate is soft and shorten only where the technology forces the issue.

Source: Amazon Forms 10-K FY2021–FY2025 (0001018724-22-000005, -24-000008, -25-000004, 0001018724-26-000004), useful-life policy notes and change-in-estimate disclosures, quoted amounts as disclosed; the January 2020 change per the filing record cited in the July note. Data as of 2026-07-31.

Here is the paradox this chapter exists to price: the company with the shortest clock still has the most overstated earnings of the four hyperscalers. Our July model re-marks every company onto a three-year server clock — our estimate of the economic life of AI-cycle compute, argued from NVIDIA's two-year architecture cadence. On that clock, Amazon's FY2025 diluted EPS of $7.17 re-marks to $6.56 — net income 9.3% lower (a model estimate, not a restatement) — a bigger gap than Microsoft's 4.9%, Alphabet's 4.4%, and larger even than Meta's 9.0% on the group's longest clock. Not because Amazon's schedule is aggressive — at five years for the subset it is the most conservative in the group — but because the dose rules: 1.70× capex per dollar of income in FY2025, now 2.34×. An honest clock on double the hardware still defers more than a stretched clock on half of it. Truth-telling has a price, and it wasn't enough.

Forward, the deferral grows with the buildout. On the model's assumptions — 45% server share, capex held at $200bn a year, which is now below the ~$220bn guided on this week's call, making every figure here conservative — the current schedule defers roughly $14.6bn of depreciation out of 2026, $21.3bn out of 2027 and $22.3bn out of 2028: about $58bn across the window, pre-tax (model estimates). Marked against our FY2026 estimates, ex-gain EPS of ~$7.94 reads ~$6.86 on the house clock. And the acceleration is no longer a forecast: the Q2 10-Q's clean split shows depreciation on property and equipment up 42% in the quarter ($13.9bn vs $9.8bn) — against 20% revenue growth — with AWS's allocated depreciation up 67%, exactly the pattern the July model said must appear as the record vintages land, in exactly the segment the 10-K said it would land in.

Source: Amazon Forms 10-K (PP&E depreciation and amortization expense incl. finance-lease amortization, FY2019–FY2025); FY2026E = H1 actual ($26.7bn, Q2-2026 10-Q, 0001018724-26-000026) plus a TON618 H2 estimate. Red marker: FY2026E plus the July model's 2026 deferred slice ($14.6bn, server-only, pre-tax — a model estimate). Data as of 2026-07-31.

One honesty note, stated because the reader should know which validations are real — and this one now carries a correction. The July model was checked against every disclosed life-change benefit it could be checked against, with every Amazon input fixed from the filings before the comparison was run — a genuine out-of-sample test, unlike the Meta leg (which contains a fitted parameter, disclosed in that note). The model rebuilt the 2022 extension's benefit at $2.80bn pre-tax, against the filing's disclosed $3.6bn depreciation reduction — a −22% residual, wider than the −10% the July note's validation table reported: that table compared the model against $3.1bn, a figure that belongs to the January-2024 change, not 2022. We report the wider residual rather than the flattering one, and the July table's correction is logged. The residual's direction also cuts against us generously: a model that under-reconstructs Amazon's disclosed benefit understates, rather than overstates, the distortion this note prices.

Two open items closed on press day — and both closed our way. This note flagged two checks pending the Q2 Form 10-Q; it was filed July 31, as this note went to press, and we read it. First, the useful-life note: no change in estimate is disclosed — under GAAP a change must be disclosed in the quarter it occurs, so silence is confirmation. The five-year subset held; there is no second correction, in either direction. Second, the clean depreciation split the release bundles: PP&E-only depreciation grew 42% in the quarter — faster than the 31% broad line — and AWS's net property and equipment grew from $190bn to $264bn in six months, on $90bn of first-half additions. The second payment is not a model projection anymore. It is a filed number, compounding at 42%.

§4 What the two machines are worth when the clock tells the truth

Method selection. Amazon is a leveraged, non-dividend-paying growth equity in mid capital-structure shift (debt doubled in six months), with two segments that deserve different multiples. We value the equity three ways and weight two: a consolidated multi-stage FCFE model (60% weight), and a sum-of-the-parts comps method — AWS and the retail complex valued against their own peer sets on FY2027E after-tax segment operating profit — run on both the reported schedule and the re-marked house clock (re-marked carries the 40% weight; reported is shown for the spread). The investment book, dominated by the Anthropic stake, is valued once, as an asset, in both methods — never as earnings.

House conventions, nothing softened. Cost of equity 10.84% (CAPM: 4.67% ten-year Treasury per FRED, July 29; beta 1.37 computed from two years of daily closes against SPY; 4.5% equity risk premium). Tax 19%. SBC stays expensed and the 10,903M diluted count is held flat. The FCFE add-back uses PP&E-only depreciation — content and lease amortization are excluded because their cash costs recur inside operating income. Terminal capex comes from the steady-state identity (depreciation + growth × capital intensity), not from optimism: 10.5% of revenue in the base case. Every scenario funds the trough with modelled debt and pays interest on it.

The 2026 base year is built from the company's own numbers: H1 actuals plus the Q3 guide midpoint plus a fading Q4 give $823bn of revenue (+15%) and $106.5bn of operating income (+33%); ex-gain net income ~$86.6bn (EPS ~$7.94 ex-gain; ~$6.86 re-marked). Capex ~$218bn net (the ~$220bn raised guide, press-sourced, less sale proceeds). Model-basis free cash flow to equity before borrowing: roughly −$74bn; even on Amazon's own more generous definition, the first half alone was −$24.9bn. The dose: ~2.5×.

Probability-weighted FCFE value: $118 per share (including the investments leg below). The interim years are the story: in the base case, cumulative 2026–28 free cash flow to equity is negative even after $155bn of modelled new borrowing. The equity holder pays the invoice for three years before the model lets them collect anything.

The SOTP lens — and the $41 the clock is worth. On FY2027E we value AWS at 22–30× after-tax segment operating profit (Microsoft trades ~25× consolidated forward earnings growing half as fast; the bull multiple matches pure AI-infrastructure names), the retail complex — including advertising — at 14–20×, and add the investment book: cash and securities net of debt, plus the Anthropic stake and other investments at Amazon's own reported carrying value net of accrued deferred tax — roughly $180bn, or ~$16.50 per share (a model estimate; band $150–210bn). We apply no independent mark to Anthropic in either direction — the conflict in §8 is exactly why. On the reported schedule the SOTP is $219 per share (probability-weighted across the three scenarios). Re-mark AWS's 2027 operating profit for the $21.3bn the current schedule defers (the 10-K states the life changes "primarily impacted our AWS segment") and the same weighting gives $178. The $41 spread is what the five-to-six-year clock is worth inside the market's favourite segment — and note that the reported-basis SOTP, at $218, already sits below the pre-print close.

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 the $271.58 close:

We believe the shape is the finding: even the bull case — AWS supercycle intact, margins at records, capex glide realised — lands 26% below the close. The market's enthusiasm for this print is an enthusiasm for the reported clock at a discount rate we cannot reproduce.

The reverse question, which is where the disagreement actually lives. Work backwards from $271.58 through the base engine, one variable at a time, and the price requires: a 26.4% six-year revenue CAGR (base: 9.1%; Amazon has not sustained that since it was a tenth of today's size), or a 7.05% cost of equity (a ~1.8% equity risk premium), or 7.9% perpetual growth. Most striking: no terminal capex assumption reproduces the price at all — even machines that are literally free forever leave the base engine roughly $70 short of the close. The market price is not resting on a heroic single input; it rests on Street conventions (SBC added back, soft discount rates) plus the reported depreciation schedule plus real AWS optimism, stacked. Remove any layer and the price does not survive.

Sensitivity, honestly displayed. Terminal capital intensity is worth ~$51/share across 8.5%→14% of revenue on the FCFE leg; each 100bp of cost of equity is worth ~$19; terminal growth 2.5%→4.0% spans ~$19. Our fair value is most exposed to the discount-rate convention — at a 9.8% cost of equity the blend rises to roughly $155–160, still 33% below the close. The single most valuable disclosure Amazon could make for this model is a segment-level depreciation schedule for AWS; it does not provide one.

§5 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 33.6% annualised (beta 1.37); earnings-adjacent 10% single-day moves are live, as this print showed. Concentration: AWS is 58% of operating income; this is, functionally, a semiconductor-cycle-adjacent utility with a store attached.

The steelmanned bull case, at full strength. First, the demand is contracted, not hoped for: management says 2027 AWS capacity is largely reserved and some 2028 capacity is already spoken for (press-sourced) — if true, today's capex converts to revenue at utility-like certainty, and our terminal-capex fear prices a risk the order book has already retired. Second, the two $25bn engines are new: AI and chips businesses each growing at triple-digit rates from a $25bn run rate is a second AWS being born inside the first; Trainium at scale would also cut the capex bill per unit of compute — attacking our thesis at its root. Third, the honest clock cuts our way: Amazon already runs the group's shortest server life; if five years is right, the re-mark's residual is only the dose, and the reported-basis SOTP ($219) is the fair anchor — still below the close, but the gap is a discount-rate argument, not an earnings-quality one. Fourth, the store is inflecting: NA margin 7.4% TTM and rising, advertising compounding at 26%, and SBC falling. If 2027 capex guidance arrives with a stated glide path while AWS holds 35%+, our base case is wrong and we say so now: that world is the bull blend, $201.

What would change our mind, with dates.

Levels, against the $142 fair value: at the $271.58 close, near the 52-week high, the gap to fair value is at its widest and the valuation argument is fully against the price; near $155 it thins; at $107 (a 25% margin of safety to fair value) the arithmetic turns constructive if AWS growth and the store's margin arc are intact. We would concede outright on a 2027 capex guide with a stated glide path below ~20% of revenue while AWS sustains 30%+ growth — in that world the bull engine is the right one. The Fund holds no position in AMZN, and this note does not recommend initiating a short: a 2.3%-of-float short base, a 33% vol regime and a capex cycle with no reliable clock make carry expensive in both directions.

§6 Where this note sits in the series

This is the Amazon satellite of The Depreciation Question (July 17) and its Q2 scoreboard The First Invoice (July 30). In that framework Amazon is the honest machine: the only company of the six that ever shortened a server life — and ate double its own guide for it — yet still the most distorted of the four hyperscalers on the house clock, because it runs the highest dose, now 2.34× and rising. The scoreboard's Q2 verdicts (depreciation accelerating four-for-four, the spend confirmed, doses rising everywhere, every clock held) are the out-of-sample evidence this valuation stands on. The siblings each carry one face of the same invoice: Microsoft (“The $116 Billion Question”) held its clock while the bill arrived fastest; Meta (“Paying Twice”) is compounding up the dose curve; Alphabet (“The Benchmark”) is the yardstick; Apple (“The Control Group”) proves the mechanism from the origin. The series re-scores itself every earnings season.

§7 Sources & Method

Every filed figure is drawn from Amazon's SEC filings via EDGAR, pulled and verified on July 31, 2026; market data from verified end-of-day sources on the same date. Model figures come from the July v3 vintage model, published 17 July 2026 and unchanged since. 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/AMZN/2026-07-31-055904).

Filings and issuer documents relied upon. Q2-2026 Form 8-K Exhibits 99.1/99.2, filed 30 Jul 2026 (0001018724-26-000024) — all Q2/H1/TTM statements, segments, the supplemental TTM capex and free-cash-flow series, share counts, Q3 guidance, the Anthropic-gain sentence (quoted verbatim in §2), and Amazon's free-cash-flow definition (Ex-99.2). FY2025 Form 10-K, filed 6 Feb 2026 (0001018724-26-000004) — the useful-life policy table (servers and networking five to six years; heavy equipment ten to thirteen), the realized 2025 life-change effect ($1.4bn depreciation / $1.0bn net income / $0.10 per share, quoted verbatim in §3), PP&E components (servers gross $113.2bn→$172.5bn; construction in progress $71.7bn), equipment acquired but not yet paid ($27.0bn), PP&E depreciation expense (FY2023–25: $30.2/32.1/41.9bn), the FY2025 one-time charges, and the Anthropic investment disclosures (notes at $45.8bn fair value and preferred at ~$14.8bn as of 31 Dec 2025; $69.1bn total investment book per MD&A). FY2024 Form 10-K, filed 7 Feb 2025 (0001018724-25-000004) — the $0.7bn guide for the 6→5 change and its in-service-assets-only basis, the "expect to continue to acquire" sentence, the $920m early-retirement charge, the heavy-equipment extension, and the 2024 5→6 change ($3.2bn/$2.5bn/$0.23). Earlier life changes per the 10-K series cited in the July note. Q2-2026 Form 10-Q, filed 31 Jul 2026 (0001018724-26-000026) — the useful-life note (no change in estimate disclosed; the five-year subset held), PP&E-only depreciation by segment (Q2 consolidated $13,869M vs $9,766M; AWS $8,076M vs $4,844M; H1 $26,703M), segment net PP&E ($190.1bn→$263.8bn for AWS), segment net additions ($90.1bn AWS, H1), and the ~$640m IEEPA tariff-refund disclosure. Per the house data-hygiene rule, no annual capex series is taken from 10-Q XBRL (10-Q trailing windows pollute annual flows); annual flows come from Forms 10-K only.

Market data and rates. Prices are adjusted daily closes from Massive Market Data: anchor close $271.58 (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 +$36.08, +15.3%, from the $235.50 close of 30 Jul). 52-week intraday range $196.00–$278.56; year-to-date +17.7% from the $230.82 year-end close. Market capitalisation ~$2.93tn on 10,783M shares outstanding. Beta 1.37 computed from two years of daily closes against SPY; 30-day realised volatility 33.6% annualised, same series. Risk-free rate 4.67%, ten-year constant-maturity Treasury (FRED DGS10, 29 Jul 2026). The +15.3% print-day move is the official close-to-close change, not a press-sourced indication; earlier overnight estimates in the press (~9–10%) understated it. Earnings-call items — the ~$220bn 2026 capex guide (raised from ~$200bn), Andy Jassy's capacity-constraint remarks, and CFO Brian Olsavsky's 2027-capacity comment — are press-sourced (CNBC; Investing.com earnings-call coverage; Yahoo Finance, all 30–31 Jul 2026): Amazon publishes a webcast at amazon.com/ir but no transcript, and each such item is labeled where used. The $220bn guide enters the model only as the 2026 capex input, clearly labeled; every re-mark figure conservatively retains the July model's $200bn.

Formulas and conventions. Ex-gain net income = reported − disclosed pre-tax non-operating gain × (1 − 21% statutory); applied to Q2 ($62.647bn − $53.415bn×0.79 ≈ $20.5bn) and TTM ($135.281bn − $79.818bn×0.79 ≈ $72.2bn); both are estimates — the filing does not disclose the gain's own tax line, and the H1 deferred-tax swing (+$30.5bn) indicates the tax is largely deferred. Dose = capex (net of proceeds, Amazon's own reported basis) ÷ net income (ex-gain where the window contains disclosed gains), same window. FY2026E revenue = H1 actual + Q3 guide midpoint + Q4 at +13%; operating income = H1 + Q3 guide midpoint + Q4 at +23%. FCFE = net income (ex-gain, 19% tax) + PP&E-only depreciation − capex net − after-tax net interest + net borrowing, discounted at Ke 10.84% (CAPM: 4.67% + 1.37 × 4.5%), mid-year convention from a 31 Jul 2026 valuation date, terminal Gordon growth 3.5% (bear 3.0%). Content and lease amortization are excluded from the add-back because their recurring cash costs sit inside operating income; working capital is held neutral (conservative for Amazon's negative cash-conversion cycle). Terminal capex = terminal depreciation + terminal growth × net-PP&E-to-revenue (base: 8.5% + 3.5%×~55% ≈ 10.5% of revenue); the bear case deliberately holds capex above the identity — that is the bear thesis. SBC is never added back and the 10,903M diluted count is held flat. SOTP: segment NOPAT = segment operating income × (1−19%); AWS multiples 22/26/30×, retail 14/17/20×; the re-mark deducts the July model's 2027 delta ($21.28bn pre-tax, model capex $200bn — below the current guide, hence conservative) from AWS operating income per the 10-K's "primarily impacted our AWS segment." The investments leg (~$180bn central) = year-end book $69.1bn + H1 P&L marks $69.0bn + H1 OCI marks (~$52bn pre-tax, derived from the comprehensive-income statement) + ~$35bn estimated new investments − ~$45bn accrued deferred tax; a derived estimate with a ±$30bn band, carried identically in both methods and never marked independently.

Limitations, stated. (1) The 21%-statutory gain-strip is an approximation; the true tax on the Anthropic marks is not separately disclosed. (2) FY2026E PP&E depreciation ($57.5bn) is built from the 10-Q's H1 actual ($26.7bn, +42%) plus a house H2 estimate; the H2 portion is unfiled. (3) The 2026 capex input rests on a press-sourced guide; the filing-verified alternative ($200bn, the prior guide) would raise our fair value by roughly $6/share — the direction of the uncertainty favours the company. (4) The investments leg depends on Amazon's own Level 3 marks of a private company that is also this note's AI maker (§8); we neither raise nor cut it. (5) The 4.5% equity risk premium is an analyst assumption with high leverage on the output; the sensitivity table prices the disagreement. (6) Segment SOTP operating profits are analyst allocations of a consolidated forecast; Amazon discloses no segment depreciation schedule. Each chart carries its source beneath it.

§8 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 Amazon (AMZN), Microsoft (MSFT), Alphabet (GOOGL), Meta Platforms (META), Apple (AAPL), 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 separately reports 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. In addition, the sum-of-the-parts method in §4 carries Amazon's investment book, which is primarily its Anthropic stake, at Amazon's own reported carrying value: we apply no independent mark in either direction. 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.