AMZN: One Buy, One Avoid, Same Conviction

On October 2, 2026, four investors read one fact sheet on Amazon (AMZN). Investor A returned Watch at conviction 6, Investor B Avoid at 8, Investor C Watch at 6 and Investor D Buy at 8. A name needs three Buy or Strong Buy verdicts out of four to join a basket, so one Buy leaves Amazon out of every basket. The session turned on one question. What does a dollar of the roughly $220 billion 2026 capital program earn, when the sheet's incremental return on invested capital row reads not available?

The day before, the same committee looked at Microsoft and produced no Buy at all: Microsoft committee review, October 1, 2026.

What the price of Amazon requires

At a $2.71T market cap, Amazon trades at 37.3x trailing after-tax operating income of $72.55B. The sheet discounts at 9.79%, which is the 10-year yield plus 4.5 points, and assumes 3.0% growth after year ten. On those terms the price requires after-tax operating income to grow 15.1% a year for ten years. On trailing net income the requirement drops to 6.7% a year. That line runs 86% above after-tax operating income, though, and the sheet itself flags it as overstating recurring earnings. On free cash flow the sheet gives no answer: trailing free cash flow is minus $11.62B, so the required growth cannot be computed. Over the last four fiscal years, operating income compounded 33.9% a year. Free cash flow growth over the same span is not computable. This is what the price assumes, not a forecast.

The debates on Amazon

1. Pre-sold capacity, or a build nobody can check yet?

The question. Is the build-out capacity that customers have already bought, or spending whose return cannot be checked?

What the sheet shows. Trailing capex is $173.03B, which is 2.00x depreciation and amortization. Trailing free cash flow is minus $11.62B. 2026 capex guidance is about $220B (CRN, July 30, 2026). The incremental return on invested capital row reads not available. Against that, AWS grew 36.8% at a 39.4% margin in the quarter to June 30, 2026. Anthropic has committed more than $100B of AWS spend over a decade (Amazon, April 20, 2026).

Where the seats landed. Investor D: "In reality, AWS capacity additions are pre-sold under contractual agreements". Investor A finds the key figure missing: "The whole bull case rests on $200B+ a year earning more than its cost of capital, and that is the one figure the filings do not let me see." Investor A also questions where the contracted demand comes from: "Some of this revenue is funded, in part, by Amazon's own capital." Investor B: "The bull case requires cash returns on new investment that the supplied figures do not yet verify." Investor C accepts the demand but not the capex path: "The capex is currently finding demand, which is the only fact that keeps the loop honest."

What settles it and when. The first test is the 10-Q for the quarter to September 30, 2026, due by November 9, 2026. If AWS growth falls below 20% while capex stays above $200B, Investor A moves to Avoid. Below 30%, Investor C moves to Avoid. Below 18.0%, Investor D's thesis breaks. If trailing free cash flow in the same filing is worse than minus $30B, Investor C moves to Avoid. A gatekeeper decision from the European Commission is reported for November 2026 but unconfirmed. It would test whether the switching costs behind the AWS margin hold.

The chair's question to Investor D. Which line in the 10-Q due November 9, 2026, would show the Anthropic commitment arriving as revenue, and not just as capacity built in advance of it?

2. Which earnings number is the business?

The question. Is Amazon's earnings power the $135.28B headline or the $72.55B recurring figure?

What the sheet shows. Trailing net income is $135.28B. Trailing after-tax operating income is $72.55B. The sheet notes that net income runs 86% above it. That gap is the difference between a 20.0x P/E and 37.3x on after-tax operating income. In the quarter to June 30, 2026, net income was $62.65B and operating income $27.46B. The sheet does not say where the non-operating items come from.

Where the seats landed. Investor A: "The reported trailing P/E of 20.0x is misleading." Investor C: "The headline 20x P/E flatters a true 37.3x." Investor B: "capitalizing the headline earnings figure would overstate recurring earning power." Investor B then goes further and normalizes to $39.60B, a 68.4x multiple on the market cap. Investor D also uses the 37.3x line and still buys, on the grounds that "fundamental earnings power is expanding substantially faster than the implied valuation hurdle." No seat defended the 20.0x figure.

What settles it and when. No seat set a breaker on the non-operating line itself. The 10-Q due November 9, 2026, will put another quarter's net income next to operating income. Investor C names the downside case: "If non-operating marks, most plausibly investment revaluations, reverse, reported earnings fall with no operational deterioration." The longer tests are Investor A's ROIC floor of 11% in the fiscal 2026 10-K, and Investor B's $180.7B normalized earnings bar in the fiscal 2029 10-K.

The chair's question to Investor A. His entry needs about 22x after-tax operating income with trailing free cash flow already positive. Investor B's base puts recurring earnings nearer $39.60B than his $72.55B. Which base is the business, and what observation splits them?

The chair's question to Investor B. If the next filings bring net income back in line with operating income, does a five-year average that includes the fiscal 2022 loss still describe the business?

3. Can a company outgrow a hiking cycle?

The question. Does the rate turn matter for a company whose operating income has compounded through the last one?

What the sheet shows. Over 30 days, fed funds rose 25bp to 3.88%, the 10-year rose 54bp to 5.29%, and the high-yield spread widened 59bp to 3.24%. Short interest is 0.76% of shares outstanding, at 2.77 days to cover. The VIX is 15.48. Against that, operating income compounded 33.9% a year over four fiscal years, and that span includes fiscal 2022, when operating income fell to $12.2B.

Where the seats landed. Investor C: "The bond market is tightening while the equity market prices perfection". On positioning, Investor C adds: "the bear case is unpositioned." Investor D points to $122.99B of liquid assets against $2.75B of principal due in the next 12 months, which Investor D reads as "insulating it from immediate refinancing strain while maintaining internal funding capacity." Investor C also ranks the two pressures: "rate impulses mean-revert, capital cycle trajectories persist."

What settles it and when. The AWS figure in the 10-Q due November 9, 2026, comes first. Investor C moves to Avoid below 30%. Investor D's breaker sits at 18.0%. Investor C's path to Buy needs three things together: AWS at or above 36.8%, 2027 capex guidance held near $220B, and a sustained decline in fed funds. The earliest point at which all three can be observed is the fiscal 2026 results release. If the European Commission confirms the reported November 2026 designation with anti-lock-in obligations, Investor C moves to Avoid.

The chair's question to Investor C. If AWS growth prints at or above 36.8% in the 10-Q due November 9, 2026, while the 10-year holds at 5.29% or higher, which of the two signals does the verdict follow?

The fact sheet every investor read

Five fiscal years (SEC EDGAR, 10-K rows, years to December 31)

Segments, quarter to June 30, 2026 (10-Q filed July 31, 2026)

The six months to June 30, 2026, show the same pattern. AWS was 21% of segment revenue at a 38.6% margin and produced 60% of segment operating income.

How Amazon makes money

The committee's analyst, a fifth voice that does not vote, splits Amazon into three segments. North America and International sell retail to consumers, who pay per order, plus services to third-party sellers and advertisers. AWS rents computing, storage and AI infrastructure to businesses and governments. AWS is about a fifth of revenue and about three fifths of operating income. In the latest quarter, each AWS revenue dollar carried about 39 cents of operating profit, against about 8 cents in North America and 4 cents in International. The analyst's summary: "retail supplies scale and cash flow, and AWS supplies most of the profit."

Depreciation drives cost. It rose from $34.4B in 2021 to $65.8B in 2025, so a growing share of cost is fixed, which gives operating leverage. Capex runs at 2.00x depreciation. The analyst notes that the build now uses more cash than operations generate, and debt and leases fund the gap.

Several figures are missing from the sheet: per-order and per-server economics, the subscription mix inside retail, and incremental returns on invested capital.

The risks to Amazon

Permanent impairment, the analyst's top five

The next two to five years, the analyst's top four

The analyst names the Taiwan foundry as the most underappreciated risk. The fact sheet has no data on it at all. Yet it sits beneath the segment that carries most of Amazon's profit and beneath the whole capex build. The Synopsys partnership pushes Amazon further into custom silicon that depends on that one foundry.

The analyst recommends the next fact sheet collect: chip sourcing by foundry and country, the share of AI capex spent with a single accelerator vendor, memory spend and supplier count, contracted against planned power, revenue by country and currency, AWS remaining contract value and customer concentration, the Anthropic stake's carrying value and any gains booked in non-operating income, the status of the US antitrust case, the EU designation and the India and China rules, labor and driver-classification cases, leadership succession, and the current total of purchase commitments.

Investor A, quality and long-term value

Investor A sees two strong businesses at a price that leaves no room for error. His words: "I would gladly own this business. At 37x recurring after-tax earnings, with the 10-year yield at 5.29%, the price gives me no margin of safety." The thesis rests on the owner's cash: "The open question is what the owner is paid."

The scenarios weight bear 25%, base 50% and bull 25%. The base case puts year-three free cash flow at $40B to $70B as capex levels off. The kill criterion is AWS year-over-year revenue growth below 20% in the 10-Q for the period to September 30, 2026, due by November 9, 2026. Why he could be wrong: "My caution could simply be the old mistake of underpaying for a great business." The ideal entry is about 22x after-tax operating income, and only once trailing free cash flow is clearly positive and capex runs below 1.5x depreciation and amortization.

The expected return line, as written: "About 3% to 8% a year". The seat adds: "The midpoint is close to the 10-year yield of 5.29%, which is no margin of safety."

Verdict: Watch. Conviction 6 of 10.

Investor B, deep value and margin of safety

Investor B starts from a conservative earnings base and finds the market far above it. Normalized earnings are $39.60B, the lower of the five-year average and the latest full year. At 15x that gives a value of $594B against a $2.71T market cap. That value is 78.1% below the market, which leaves no margin of safety. Tangible book, which would set a floor, reads not available. "Amazon may be a good business, but it is not a good defensive investment at this valuation."

The scenarios weight bear 25%, base 50% and bull 25%, with base-case year-three free cash flow of $43.6B. The kill criterion is an AWS operating margin below 30% in the 10-Q due November 9, 2026, which "would warrant a lower conservative multiple." The valuation breaker is normalized earnings of $180.7B, the level needed to support today's market cap at 15x, tested in the fiscal 2029 10-K. Why he could be wrong: "The five-year average may materially understate a permanently improved business." The ideal entry is 10.5x conservatively normalized earnings or less.

Verdict: Avoid. Conviction 8 of 10.

Investor C, global macro and reflexivity

Investor C reads Amazon as a self-reinforcing loop. AWS numbers justify the capex raise, the capex feeds suppliers, and partner commitments come back as contracted demand. "That is reflexivity in its textbook form, and it is still in its positive phase." Then the macro turned. Over 30 days, fed funds rose 25bp, the 10-year rose 54bp to 5.29%, and high-yield spreads widened 59bp. Short interest is 0.76% and the VIX is 15.48. "I do not short working reflexivity, and I do not buy crowded duration into a hiking impulse."

The scenarios weight bear 25%, base 50% and bull 25%. Base-case year-three free cash flow is $0 to $10B, the lowest of the four seats. AWS growth below 30% in the 10-Q due November 9, 2026, moves the seat to Avoid. A move to Buy needs three things together: AWS at or above 36.8%, 2027 capex guidance held near $220B, and a sustained fall in fed funds. The first point to observe them is the fiscal 2026 results release, which the last four releases place in early February 2027. "I may be applying a macro frame to a company that has escaped the macro frame." The ideal entry is roughly 25x trailing after-tax operating income. The alternative is a high-yield spread near 4.5% in a liquidity event that hits the equity without touching AWS fundamentals. He calls that spread his own judgment, not a sourced level.

Verdict: Watch. Conviction 6 of 10.

Investor D, growth and fundamental stock picking

Investor D cast the session's only Buy, the first from any seat in eight days. AWS re-accelerated to 36.8% growth at a 39.4% margin, which he reads as structural operating leverage. In Investor D's view the capacity "is already being monetized through massive multi-year enterprise commitments." The demand proof is Anthropic's commitment of more than $100B of AWS spend. The supply proof is the Synopsys silicon partnership and the 20-year, 690MW nuclear agreement.

The scenarios weight bear 25%, base 55% and bull 20%. Investor D is the only seat to put more than half on the base case. Base-case year-three free cash flow is $55B. The kill criteria are AWS growth below 18.0% or AWS margin below 30.0% in the 10-Q for the period ending September 30, 2026 or later, or trailing operating cash flow below $110B in the fiscal 2026 10-K. The critical environment: "A negative shift in enterprise compute appetite would strand hundreds of billions in capital infrastructure, turning the investment verdict from Buy to Sell." Why he could be wrong: enterprise AI adoption could hit an architectural ceiling and leave Amazon with over $220B of rapidly depreciating datacenter assets. The ideal entry is 28.0x after-tax operating income or lower, or enterprise value below 20.0x trailing operating income.

Verdict: Buy. Conviction 8 of 10.

What would change the count on Amazon

What we take from it

Conviction measures how sure a seat is of its own reading. It says nothing about whether anyone else shares that reading. The two most confident seats in this session both scored 8 of 10, read the same sheet line by line, and came out at opposite ends. If the threshold weighed conviction, the chair would decide how much a confident voice counts, and that is a thumb on the scale. The published threshold counts verdicts and nothing else. A Buy at 8 is one vote, exactly as a Buy at 2 would be. The record shows why that matters. Since September 24, Investor A has returned Watch on seven sessions running, Investor B Avoid at conviction 8 on all seven, and Investor C Watch on every one of them. Investor D is the only seat that moves. Three steady seats outvote one ardent one.

As of October 2, 2026, BagsCapital manages no outside capital and holds no position in Amazon.