ReportOwnershipAnnika Sorensen10 min read
ReportOwnership

Amazon.com, Inc., Form 8-K filed 29 October 2026, Exhibit 99.1; Form 10-Q for the quarter ended 30 September 2026, filed 30 October 2026.
Operating income and net income moved in opposite directions inside the same quarter, and only operating income measures the business. Operating income rose $10.7bn to $28.1bn, with AWS operating income of $17.8bn against $11.4bn. Net income fell $0.5bn to $20.7bn from $21.2bn, and Amazon’s own explanation is the whole of the argument: the decrease is “driven primarily by pre-tax gains of $9.5 billion in third quarter 2025 included in non-operating income (expense) from our investments in Anthropic”. The comparison was against a mark taken a year earlier. The same note takes a larger one now: $62.9bn of upward adjustment on private-company equity in the nine months to 30 September 2026, of which $62.8bn was the nonvoting preferred stock in Anthropic. A CFO reading a supplier’s earnings as a measure of the supplier’s health is reading, in part, a valuation of the supplier’s customer, and that customer has filed to list.
| Line and basis | Where it is recognised | Prior figure | Latest figure | What moves it |
|---|---|---|---|---|
| Anthropic nonvoting preferred stock, at 31 December 2025 and 30 September 2026 (approximate, as the issuer prints it) | Balance sheet, Level 3, remeasured through other income (expense), net | $14.8bn | $92.5bn | Each Anthropic funding round sets a new observable price |
| Convertible notes issued to Amazon, at fair value, same two dates; principal outstanding $6.2bn | Balance sheet at fair value | $45.8bn | $146.2bn | The next funding round, or the listing |
| Unrealized gain on those notes, same two dates | Accumulated other comprehensive income, outside net income | $39.5bn | $140.3bn | Recognition in other income when the notes convert |
| Upward adjustment recorded in other income, nine months to 30 September 2025 and 2026 | Other income (expense), net | $7.2bn | $62.8bn | Observable changes in price, which the issuer ties to “Anthropic’s fundings” |
| Net discrete tax expense attributable to those adjustments, same nine-month periods | Income tax provision | $354m | $15.2bn | The revaluation is taxed when it is recognised, with no sale |
Amazon Form 10-Q for the quarter ended 30 September 2026, filed 30 October 2026, investment note and income tax note; Anthropic, PBC Form S-1/A filed 2 November 2026 for the $6.2bn of principal outstanding.
The note is unusual for how little it leaves inferred. The stock is 164 million nonvoting preferred shares at 30 September 2026, each convertible into ten shares of nonvoting common stock, carried at fair value through other income rather than held at cost, so every funding round becomes an earnings event. Amazon names the inputs and calls them Level 3: observable changes in prices, the rights and obligations of the stock, and a discount for lack of marketability. At 31 December 2025 the preferred stood at approximately $14.8bn; nine months later it stood at approximately $92.5bn. The line in the accounts is called equity investments in private companies, and on the issuer’s own attribution it is one company’s preferred stock that moved the profit line.
The registration statement supplies the mechanism, and the mechanism is specific. Anthropic, PBC applied to list its Class A common stock on the Nasdaq Global Select Market under the symbol ANT, and the amendment filed on 2 November 2026 leaves the estimated price range blank. On completion the notes convert into Class N common stock, sized “without causing it to exceed the applicable ownership cap”, and the amount that cannot convert at the offering is dealt with in the next sentence: “the remaining balance of such notes will automatically convert, subject to Amazon’s ownership cap, into shares of our Class N common stock at the applicable conversion price at the end of each fiscal quarter following this offering”. Where a balance still does not convert before maturity, “such balance will be repaid in cash at such maturity date”. Two consequences follow. How much of a $140.3bn unrealized gain becomes income in a single quarter is a decision Amazon holds, because the cap “may be waived at our election”, and not a price the market sets. And whatever the cap keeps out of stock becomes a cash claim on a company that warns in the same document that it “may not have enough available cash on hand” to repay its convertible notes.
After the offering the mark keeps moving, against a different reference. The notes are revalued against a quoted price rather than a funding round, and the residue converts at the closing price at the end of each fiscal quarter, so a line that produced $62.8bn of other income in the nine months to 30 September 2026 becomes a quarterly remeasurement. The figures are worth holding together. At 31 December 2025 the notes carried $45.8bn at fair value with $39.5bn of unrealized gain in accumulated other comprehensive income; at 30 September 2026 they carried $146.2bn with $140.3bn. The tax on the part that has reached profit is already charged: $15.2bn of net discrete tax expense in the nine months, attributable in the issuer’s words primarily to these adjustments, against a gain nobody has sold. The $140.3bn that has not reached profit is waiting on a price.
Amazon is not alone in disclosing that the counterparty is joined at both ends. Microsoft’s Form 10-Q for the same quarter states that OpenAI is “an equity method investee” and “a related party as defined in Accounting Standards Codification Topic 850”, then discloses “revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments, of $10.0 billion” for the three months to 30 September 2026, with accounts receivable from OpenAI of $7.2bn against $6.0bn at 30 June 2026. The revenue is real and it is earned on commercial terms. The qualifier that matters to a reader using it as a demand signal is “inclusive of revenue-sharing payments”, because a numerator that contains the return of the investor’s own capital is not a measure of what an independent buyer will pay.

Anthropic’s registration statement states the buyer’s problem in the registrant’s own words, and it puts it in the risk factors rather than the notes. “Amazon, Google, and Microsoft, through their respective affiliates, have made significant equity investments in our company.” In the same paragraph: “These CSPs are also our customers, purchasing our intelligence service for internal use or for their own products and services.” On the consequence the filing does not hedge: the distribution relationships “create dependencies, give them visibility into our pricing and commercial terms across the platforms they operate, and may influence their incentives regarding pricing, compute allocation, service quality, sales prioritization, and compensation structures”. An enterprise that ran its concentration review on the model lab and its exit review on the cloud has been testing one counterparty twice, and the lab’s own filing says so.
Our reliance on a limited number of CSPs and suppliers, some of which have strategic and investment relationships with us and are our customers, creates complex dynamics that could give rise to conflicts of interest and adversely affect our access to compute.
Size the relationship from the commitments, not from the equity. Amazon’s 10-Q reports that in Q2 2026 “AWS and Anthropic announced an expansion of the strategic collaboration and existing multi-year commitment by more than $100.0 billion over 10.0 years, which includes contractual obligations related to the performance of AWS chips”, and that in Q1 2026 the OpenAI arrangement was expanded by $100.0 billion over 8.0 years on top of an existing $38.0 billion, on the same chip clause. Both are announced expansions of multi-year commitments, not booked revenue, and Amazon does not call them take-or-pay. What they establish is direction. The customer’s commitment obliges it to consume the supplier’s silicon, the supplier’s reported profit gains from the customer’s valuation, and the buyer signing the same shape of commitment sits on neither ledger.
By 31 January 2027 the board resolves the first question below, and it is a scope decision rather than an opinion: whether every AI commitment whose supplier holds equity in, or has lent to, the party delivering the models sits on one line of the counterparty register. The CFO’s register entry, the CIO’s re-run concentration test and the General Counsel’s portability schedule are the three artefacts that answer it, all due the same day, and the other questions on this list are what the board asks once that scope is fixed.
The number to hold is not $62.8bn. It is the conversion price in the next amendment to the registration statement, because that is the figure that decides how much of the $140.3bn stops being one company’s estimate about another and becomes income a market sets every day. Amazon’s filing says which number its profit depends on. The operating business earned $28.1bn, the reported figure was lower than a year earlier, and the difference is a holding that is about to be priced in public.
Share this
Related