1. Investment Thesis
Amazon is a retail distribution network, a seller-services marketplace and an infrastructure platform with different capital needs. AWS profit growth is real, but the investable question is whether new infrastructure produces cash after replacement spending. A low headline P/E following private-investment gains is not proof of cheap operating assets.
- Retail logistics and Prime reinforce purchase frequency; advertising monetizes existing purchase intent.
- AWS and custom silicon can improve economics, but hardware refresh cycles and power constraints consume capital.
- Separate core enterprise value from liquid assets, financing claims and risky private-investment optionality.
Source
2. Financial History
FY2011-2025 captures the transition from low-margin merchandise sales to a services-heavy mix. FY2025 revenue reached $716.924B and operating income $79.975B, yet company-defined FCF fell to $11.194B. Improving margins and weakening cash conversion can occur together when investment expands faster than earnings.
- The 15-year financial window is not a complete IPO-to-present audited database; the filing inventory extends further where available.
- Pre-2022 EPS and shares are normalized for the 20-for-1 split. Original filing values are otherwise retained, not silently restated.
- Gross margin excludes fulfillment and technology costs from cost of sales; ROE uses year-end equity, not average equity.
Source
3. Latest Quarter
Q2 2026 revenue rose about 20% to $200.606B and operating income reached $27.461B. The $62.647B net income includes unusually large non-operating investment income. The operating result is the cleaner anchor for an earnings-power discussion; stripping a pretax gain directly out of after-tax profit would be misleading.
- AWS revenue grew 36.7%; the $169B annualized run rate is not reported full-year revenue.
- Q3 sales guidance is $197-202B and operating income $22.5-26.5B, issued July 30; Q3 results are not yet in this report.
- Prime Day timing and foreign exchange affect the Q3 growth comparison. Seasonality makes straight-line quarterly extrapolation unreliable.
Source
4. Segment Economics
The three reported segments are North America, International and AWS. Q2 AWS generated $16.621B of the group's $27.461B operating profit, despite representing roughly one fifth of sales. Geography-based retail segments already include advertising and seller services: those channels cannot be added again as a fourth segment.
- Retail productivity depends on delivery density, inventory placement and fulfillment cost per unit, not just GMV.
- Seller-services revenue is not seller GMV; advertising revenue does not disclose standalone operating profit.
- The dashboard uses disclosed segment profit and channel revenue. Prime subscriber count, take rate and cohort CAC remain N/A where not disclosed.
Source
5. Moat and Competition
Amazon's advantage combines selection, delivery reliability, merchant demand and enterprise switching costs. These advantages do not guarantee constant pricing power. Retail competes with local stores and other marketplaces; AWS competes with hyperscalers, specialized clouds and customer-owned infrastructure.
- Microsoft Azure and Google Cloud are qualitative AWS competitors; no unsourced market-share or peer-multiple table is used.
- Delivery speed can increase retention while raising fulfillment cost. Ads can monetize demand but excessive ad load can degrade discovery.
- Watch operating margins and cash return on new investment rather than assuming scale alone proves an economic moat.
Source
6. AI Infrastructure
Trainium, Graviton, Bedrock and the broader AWS stack offer several ways to serve AI demand. The key risk is duration mismatch: equipment is purchased up front while utilization, pricing and customer funding evolve. Management's AI and chips run-rate disclosures overlap AWS and can overlap one another; they are not incremental consolidated sales.
- RPO of $496B has a 6.4-year weighted remaining life; backlog is neither cash collected nor revenue guaranteed next year.
- Contracts with major AI labs link compute demand and strategic financing. Stress-test counterparty funding rather than equating commitments with risk-free receipts.
- The model assumes total investment of $225-235B in 2026, including noncash additions; this is our estimate, not a verified latest capex guidance quote.
Source
7. Cash Flow Quality
Company FCF is operating cash flow less property purchases net of asset-sale proceeds and incentives. It is not cash available after all lease principal or strategic investments. Positive operating cash flow can coexist with negative distributable cash, particularly when SBC, vendor timing and infrastructure growth are substantial.
- TTM company FCF is -$7.604B; after finance-lease and financing-obligation principal it is -$9.511B, still before strategic investments.
- H1 adds $2.128B of finance-leased equipment and $20.620B of unpaid-equipment increases beyond cash purchasing; these are distinct timing and financing disclosures.
- Historical aggregate D&A includes operating-lease and content components. The enterprise model does not blindly add that aggregate to EBIT; SBC stays an economic cost.
Source
8. Balance Sheet and Claims
Liquidity must be assessed alongside funded debt, financing leases and additional commitments. The DCF bridge uses debt carrying value rather than debt face value. Operating lease payments are already costs in the projected operating profit, so subtracting the full operating lease liability as well would mix incompatible valuation conventions.
- June cash plus marketable securities totals $122.988B; restricted cash is excluded. Known subsequent OpenAI funding reduces the pro-forma liquid pool by $21.3B.
- Long-term debt carrying amount is $132.224B; finance-lease liability is $13.451B. Financing obligations use the note's rounded approximately $9.315B total.
- Uncommenced lease payments and purchase commitments are future nominal obligations, not additional current debt at full nominal value. Their operating and investment implications remain risks.
Source
9. Capital Allocation and Investments
Anthropic and OpenAI positions create potentially valuable upside and concentrated financing exposure. Accounting marks are not distributable cash, and preferred-stock rights, conversion conditions, lockups and future funding affect realized returns. Model the investment portfolio outside core operating cash rather than capitalizing investment gains as recurring EBIT.
- Anthropic preferred shares and notes total about $190.4B at June carrying/fair values. Q2 includes $50.5B upward preferred-stock marks; note gains can sit in OCI.
- OpenAI funded cost reaches $50B after the disclosed subsequent payment. The model uses that cost, not an invented current market valuation.
- 25/50/75% recovery scenarios and a 20% proceeds reserve are deliberate haircuts, not fair-value certifications. The potential remaining $15B Anthropic facility is deducted in all cases.
Source
10. Ten-Year Operating Forecast
The model grows North America, International and AWS separately, then applies segment margins. It explicitly charges full investment and incremental working capital against after-tax operating earnings. A revenue growth story only creates equity value when incremental returns exceed financing and replacement costs.
- Base 2026 sales assumptions are $460B North America, $170B International and $178B AWS; they are not management segment guidance.
- AWS growth fades toward 12% and its margin toward 32% by 2035 in the base case, allowing for competition and hardware-heavy mix.
- Full-investment intensity fades to 14% of sales while non-operating-lease depreciation reaches 10.5%. These unverified assumptions are the central model risks, not hidden outputs.
Source
11. Valuation and Sensitivity
The reference price is the October 2, 2026 close of $251.52, not a live quote. Discount unlevered operating cash with WACC, add pro-forma liquid assets, deduct financing claims once and add haircut investment optionality. Show core value and investment value separately so a private-asset mark cannot disguise a weak operating valuation.
- Historical market capitalization uses split-adjusted close times annual diluted weighted shares: a comparability proxy, not exact year-end equity capitalization.
- P/E is omitted in loss years; negative FCF yield is retained rather than relabeled. Recent net profit is distorted by investment marks.
- The first year includes only an estimated quarter of cash still prospective at October 4. The terminal calculation keeps reinvestment and adjusts working capital to terminal growth.
Source
12. Risks and Thesis Breakers
The most important downside is not a single weak quarter but a sustained gap between infrastructure spending and cash generation. Power availability, equipment useful lives, AI customer funding, retail competition and regulation can jointly impair returns. A model requiring lower investment intensity is vulnerable if growth needs continuous elevated replacement spending.
- Reconsider the thesis if AWS margin falls while capex stays elevated and new capacity fails to lift cash generation over several reporting periods.
- Watch lease commencements, equipment payment catch-up, dilution and strategic funding rather than only headline OCF growth.
- Security outages, antitrust remedies, tariffs and changes in seller/customer behavior can weaken otherwise durable platforms.
Source
13. Monitoring and Conclusion
Amazon's operating franchise is strengthening, but the stock is an investment in future cash returns rather than reported investment gains. The scenario range is a decision framework, not a price target or recommendation. Update the model when Q3 results and investment funding disclosures arrive, keeping the same cash and claim conventions.
- Quarterly checklist: AWS growth and margin; retail segment profit; net cash capex; noncash equipment additions; OCF less all investment.
- Annual checklist: replacement lives, SBC and share count, lease/debt maturities, realized investment proceeds and cash-tax normalization.
- No personalized investment advice. Historical results, unaudited interim accounts and analyst forecasts have distinct evidentiary status.
Source