Act 01 · The filingThree records in nine days, one single trigger

Between July 22 and July 30, 2026, Alphabet, Microsoft and Amazon reported their quarterly results. All three posted records, together $210 billion in net income in a single quarter. Amazon's press release contains one sentence that almost nobody quoted.

Amazon, Q2 2026 · one sentence from the press release

„Second quarter 2026 net income includes non-operating pre-tax other income of $53.4 billion, primarily from our investments in Anthropic.“[01]

Of $62.6 billion in reported profit, $53.4 billion is not revenue. No customer paid for it, not a cent changed hands.

Amazon's over the same trailing twelve months was minus $7.604 billion. A year earlier it was plus $18.184 billion[01].

„Record profit“ and „earned no money“ appeared in the same document.

Both are true. It is an accounting rule, not a contradiction.

What a paper gain is, in one picture

Imagine you bought a house three years ago. Yesterday your neighbour sold an identical one for twice the price. On paper yours is now worth twice as much. You have received nothing, you cannot spend it, and if the market turns, the gain is gone.

The only difference to Amazon: you do not have to report this as income on your tax return. A US corporation has to put it in its income statement. That is a .

The neighbour here is called . On May 28, 2026, $65 billion in fresh capital went into Anthropic at a valuation of $965 billion[05]. Amazon had paid $13 billion into the same company over three years[08] and now marks that stake up to the new price. The difference is the profit.

Who put up those $65 billion is the more interesting question. The answer is in Act 03.

All of this is legal, audited and standard. It only has two properties that the word „profit“ does not suggest: you cannot spend it, and it can reverse.

Chart 01 · Reported and received
Three companies, the same quarter, the same stake in the background.
reported net income of which paper gain, after tax quarter ending June 30, 2026
Alphabet · reported net income
$112.2bn[02]
Up 298 percent against the same quarter last year.
of which paper gain, after tax
$77.1bn
69 percent of profit. Free cash flow in the quarter: minus $5.855bn
A single line item: „gain on equity securities, net“ of $99.031 billion. Alphabet's press release gives only the total. The value of the individual stake appears only in a separate filing[06].
Amazon · reported net income
$62.6bn[01]
Up 245 percent against the same quarter last year.
of which paper gain, after tax
around $41.6bn[△]
66 percent of profit. Free cash flow, trailing twelve months: minus $7.6bn
Pre-tax it is $53.415 billion. Amazon itself writes „primarily from our investments in Anthropic“. The after-tax figure is derived, Amazon does not state it.
Microsoft · reported net income
$35.8bn[03]
Up 31 percent against the same quarter last year.
of which paper gain, after tax
around $3.1bn[△]
9 percent of profit. Free cash flow in the quarter: plus $19.6bn
$3.2 billion from Anthropic, stated that way in the press release and without a tax figure, plus $0.48 billion from OpenAI, which Microsoft already reports as an after-tax effect. Microsoft is in the same business and books a fraction. Why, is in Act 02.
Aggregate Q2 2026 · Amazon and Alphabet combined
reported profit
$174.8bn
sum of reported net income
free cash flow
−$13.5bn
same quarter: Alphabet −5.9, Amazon −7.7[△]
$152.4 billion in revaluation gains sit pre-tax in the two income statements (53.4 plus 99.0). After tax around $118.7 billion remains, a good two thirds of reported net income. A „booked to cash“ ratio cannot be formed, because the denominator is negative. Including Microsoft, the three together reported around $210 billion in net income and around $122 billion in paper gains after tax, or 58 percent[△].
Sources: Amazon earnings release Q2 2026 (July 30, 2026)[01] · Alphabet earnings release Q2 2026 (July 22, 2026)[02] · Microsoft FY26 Q4 (July 29, 2026)[03]. Amazon's cash flow figure is the trailing twelve month value, Alphabet and Microsoft show quarterly values. After-tax figures for Alphabet and Microsoft from their respective footnotes, for Amazon derived from the 10-Q[△].

Act 02 · The ruleOwn more, book less

Microsoft is the control case: the company holds stakes in both large AI labs. In the same quarter in which Amazon books $53 billion, Microsoft's entire non-operating income line shows a plus of $3.444 billion[03]. A year earlier the same line showed minus $1.707 billion.

Chart 02 · Microsoft's paradox
The smaller stake produces the bigger paper gain. By almost a factor of seven.
① The small position · Anthropic
Paid inaround $5bn, invested in November 2025[07]
Stakewell below the influence threshold
RuleASC 321, remeasured at every funding round
Resultplus $3.2bn in the quarter[03]
② The large position · OpenAI
Paid in$11.9bn of $13.0bn committed, as of June 30, 2026[16]
Stakearound 25 percent „on an as-converted basis“, above the threshold[16]
RuleASC 323, equity method under HLBV, value does not count
Resultplus $0.48bn in the quarter[03]
Microsoft has put more than twice as much money into OpenAI as into Anthropic, and press reports value that position at around $135 billion[17]. It still contributes one seventh of what the small bet contributes. The reason is not in the business, it is in the rulebook.
Sources: Microsoft FY2026 Q4 press release and GAAP reconciliation (July 29, 2026)[03], which states verbatim „These include a $3.2 billion gain from our investment in Anthropic“. Commitment, funding, ownership and the HLBV method from the 10-K annual report for fiscal year 2026[16]. Important: the $135 billion is an estimated market value of the stake, not a balance sheet value. Under the equity method Microsoft's books carry a far lower figure, namely the amounts paid in less the losses absorbed.

The 20 percent threshold

Under US accounting rules, the size of a stake decides what may enter the income statement at all. The line sits at around 20 percent of voting rights: above it, an investor is presumed to exercise significant influence[13].

Below the threshold, applies. Because a private company has no running market price, the so-called kicks in: the stake is carried at cost and only adjusted on an observable price event. A new funding round is exactly such an event. This is where Amazon's and Alphabet's Anthropic stakes sit.

Above the threshold, applies, the equity method. The investor then books not the value of the stake but its share of the investee's earnings. At an AI lab burning billions, that is more of a loss channel, and increases in value do not show up at all.

The larger the stake, the less of its appreciation may reach the profit line.

The same rule hands Amazon and Alphabet their record numbers: both deliberately keep Anthropic below the threshold. Amazon's stake runs through non-voting preferred stock, Alphabet's sits at around 14 percent.

All three paper gains have the same trigger: Anthropic's Series H on May 28, 2026 falls into the quarter of all three companies. A single price event at a firm none of them controls moves three income statements. Not one of them received a dollar for it.

Was it always like this? No, the rule only applies since 2018

Until the end of 2017, stakes without a market price fell under the cost method: the holding sat at cost and stayed there. Increases in value did not appear in the income statement, they only became profit when the stake was actually sold, that is, when money moved.

In January 2016 the US standard setter FASB changed this. ASU 2016-01 abolished the cost method, equity investments have since been carried at fair value, and changes in value run through the income statement. For calendar-year companies this applies from fiscal 2018 onwards. The measurement alternative comes out of the same change[14].

Had Anthropic closed the same round in 2017, Amazon's income statement would show a zero here.

That the change would make reported profits noisier was known at the time. EY wrote for clients in 2016 that the cost method was no longer available for equity securities without readily determinable fair values, and that it followed:

„The new guidance could significantly increase earnings volatility for some entities, especially those that hold significant investments in equity securities they classified as AFS before adopting ASU 2016-01.“EY, Technical Line on ASU 2016-01[14]

EY had listed holdings primarily in mind there, whose swings in value had until then bypassed net income and gone straight to equity. Today's record numbers come from the other branch of the same change, the holdings without a market price. The warning was still right on the point that matters here: reported profit has swung ever since for reasons that have nothing to do with the business.

The $53.4 billion is not a trick and not a grey area, it is the prescribed consequence of a rule that has only existed for eight reporting years. It becomes visible only now, because before there were no private companies whose valuation triples within twelve months while corporations sit on double-digit billion stakes in them.

The same event, three levels of openness

All three book from the same occasion, but they differ in how clearly they say where the profit came from.

CompanyWhere the number sitsIs the stake named?
Amazon Press release, third bullet, in plain words: „non-operating pre-tax other income of $53.4 billion“[01] Yes, verbatim „primarily from our investments in Anthropic“
Alphabet Press release as an aggregate line „Gain on equity securities, net“ of $99.0bn, after-tax effect in a footnote[02] No. The value of an unnamed holding sits in a separate filing, the attribution comes from Bloomberg[06]
Microsoft Press release, in the section on discrete items: „These include a $3.2 billion gain from our investment in Anthropic“. The smaller OpenAI effect of $480m additionally sits in its own reconciliation table[03] Yes, both holdings by name. Microsoft is the only one that even publishes a profit figure adjusted for the stake
None of these three variants breaks a rule. What is prescribed is the line, not the name behind it. The difference is still striking: Alphabet, which reports the largest paper gain of all at $99 billion, is the only one of the three that does not say where it came from. Anyone who wants to know has to read a separate filing and take the attribution from a news agency.

Act 03 · The curveA year ago it was two billion

The numbers in Act 01 are a still image. The movement says more: the same item, six quarters in a row, at the same two companies.

Chart 03 · Six quarters, the same line
From 2.4 to 152.4 billion. The paper gain does not grow, it jumps.
Amazon Alphabet paper gain pre-tax, billions of dollars
Q1 2025
12.5
Q2 2025
2.4
Q3 2025
20.9
Q4 2025
3.5
Q1 2026
52.6
Q2 2026
152.4
Across all six quarters combined: $244.3 billion in paper gains. Of that, $152.4 billion, or 62 percent, falls into the last quarter alone. Against the same quarter a year earlier that is a factor of 63.
Amazon: line „Other income (expense), net“, Alphabet: line „Gain (loss) on equity securities, net“, each read directly from the earnings releases of those six quarters[01][02][15]. Pre-tax values, because only those are disclosed for all six quarters. The two lines are not identical: Alphabet's is a pure equity item, Amazon's is somewhat broader and contains, alongside the markups on holdings in private companies (around 50.5 of the 53.4 billion in the last quarter), smaller valuation items such as warrants and currency effects[△].

First, the number jumps rather than grows. Q2 and Q4 2025 sit at two to three and a half billion, Q1 and Q3 in between at twelve and twenty-one. That is not a business trajectory, it is a calendar: the entry only appears when a new funding round takes place. Earnings with this pattern cannot be extrapolated, which is what every forecast attempts.

Second, the order of magnitude is new. In the quarter ending June 30, 2025, after-tax paper gains made up around five percent of net income at Amazon and around four at Alphabet. A year later it is two thirds. Anyone treating that as the normal state is mistaking one year for a trend.

And where did the money come from that set the price?

Outsiders value Anthropic higher, three companies mark their stakes up alongside. That is incomplete: Anthropic broke down in the Series H announcement where the $65 billion came from.

Chart 04 · Where the $65 billion came from
Just under a quarter of the money that set the new price came from the companies booking gains on it.
Financial investors
around $50bn
Led, according to Anthropic, by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, co-led by Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ and XN, alongside Blackstone, Brookfield, Fidelity, Jane Street, Lightspeed, MGX, T. Rowe Price and Temasek among others. This is fresh outside equity, and it sets the price.
Hyperscalers, $5bn of it Amazon
$15bn
Previously committed amounts from the same companies that subsequently report paper gains on the outcome of the round. Google does not appear in the round as a funder but through a commitment for five gigawatts of TPU capacity with Broadcom. Who provided the other $10 billion of the $15, Anthropic does not say.
„It also includes $15 billion of previously committed investments from hyperscalers, including $5 billion from Amazon.“ Anthropic, Series H announcement, May 28, 2026[05]
Breakdown from the Series H press release[05]. The residual of around $50 billion is a difference calculation, Anthropic does not disclose the individual amounts from financial investors[△].

Which means the house analogy from Act 01 no longer quite holds. The neighbour did not simply sell his house for more. Part of the purchase price came from you.

A good three quarters of the round is money from investors with no connection to Amazon, Google or Microsoft. The $965 billion price was not made by the hyperscalers. It did not come about without them either. What stands out is the ratio of money in to profit booked: Amazon has paid $13 billion into Anthropic and writes $53.4 billion in profit out of it in a single quarter. Four times everything ever transferred.

Act 04 · The consequencesWhy this is more than a footnote

One could object: fine, analysts strip this out anyway. That is the best counterargument, and Gil Luria of D.A. Davidson makes it[04]. Three points remain nonetheless.

First: the increase consists almost entirely of this one item

The percentage of profit is the milder view. Calculated against the increase, it looks different. The prior-year quarter also already carried paper gains. Stripping them only out of the current quarter compares an adjusted year with an unadjusted one and depresses the result. The calculation below therefore adjusts both periods.

Chart 05 · The increase in profit, with and without
More than 90 percent of the increase at Amazon and Alphabet is this one item. At Microsoft, half.
Metric, in $bnAmazonAlphabetMicrosoft
Net income, prior-year quarter18.228.227.2
of which paper gain, after tax0.91.0−1.6
adjusted, prior year17.327.228.8
Net income Q2 202662.6112.235.8
of which paper gain, after tax41.677.13.1
adjusted, now21.035.132.7
Share of the increasearound 92 %around 91 %around 54 %
reported growth+245 %+298 %+31 %
for reference: company's own adjustmentnonenone+22 % (OpenAI only)
Growth without paper gainsaround +22 %around +29 %around +14 %
Own calculation from the reported figures in the three earnings releases[01][02][03], both periods adjusted[△]. Alphabet states its after-tax figure itself, for Amazon and Microsoft it is derived. Microsoft's quarter is FY26 Q4 and ends on the same day. On the range: if only the Anthropic markup is stripped out at Amazon instead of the whole line, the result is around plus 32 rather than plus 22 percent. Both definitions are defensible, the one used here is the more conservative. Method and calculation in the making-of.

Plus 22 and plus 29 percent are respectable numbers for companies of this size. They are simply something other than plus 245 and plus 298.

Microsoft shows that this runs in both directions. The company reports plus 31 percent and looks like the solid counterexample to the other two. In the prior-year quarter, though, its books carried a loss of $1.575 billion from the OpenAI stake[03], which pushed the comparison base down.

Microsoft voluntarily publishes figures adjusted for the OpenAI stake alongside the official ones and arrives at plus 22 rather than plus 31 percent[03]. That adjustment leaves the Anthropic gain in place, which was the larger of the two in the quarter. Strip that out as well and around 14 percent remains. The company with the smallest paper gain is therefore the only one of the three whose adjusted growth sits below its reported growth.

Second: it now distorts the entire index

Data provider ran the numbers for the quarter to see what happens when these items are removed. Earnings growth across the entire S&P 500 was around 48 percent. Without the equity gains at Alphabet and Amazon it would have been around 29[04].

Two positions in two balance sheets thus explain nearly two fifths of the earnings growth of the 500 largest listed US companies. Anyone paying into a broad index fund has this in their return, without a single additional dollar having been earned anywhere.

Third: it lands in a quarter in which money got tight

Paper gains would be a side note if the cash flow next to them held up. It does not. Amazon's capital expenditure over the last twelve months came to $169.0 billion against $161.4 billion in operating cash flow, or around 105 percent[01]. Among the large providers Amazon is the only one above one hundred, the others sit between 57 and 63 percent according to an independent analysis[10]. Long-term debt doubled within six months from $65.6 billion to $128.9 billion[01].

And Alphabet, a company with around $144 billion in net cash, issued shares in June 2026: $49.6 billion net, plus a programme for another $40 billion[02]. In the same quarter in which it reports $99 billion in paper gains, it raises fresh equity.

The profit sits in the income statement. The money came from the stock market.

ClosingWhat becomes of it

The question that stays open

„Does the paper ever turn into money?“

Outcome A · Bullish

The paper gains become real.

Anthropic confidentially filed for an IPO on June 1, 2026. If it prices at or above the Series H level, the entry turns into a market price at which shares can actually be sold. Amazon's $13 billion would then be worth real billions, and the accounting line would merely have been right earlier than the cash register. The tangible side supports this outcome: data centres are at capacity, AI chips have months of lead time, and the contracts behind it are signed.

Outcome B · Bearish

The paper gains go up in smoke.

If the valuation falls at the next round or at the IPO, the same mechanism runs backwards. Same line, same rulebook, opposite sign, and without a single customer ordering less. The drop is known: $244 billion in paper gains from six quarters, none of which ever passed through a cash register and therefore cannot be defended by one either. Anyone who read the gains as earning power will read the loss as collapse. Both would be equally wrong.

Outcome C · the middle path, and the most likely one. The paper gains stay, and reported profit stops working as a metric. This is already happening: analysts strip the items out, LSEG publishes two numbers side by side. What remains is a figure you can only read with a footnote.

The next data point has a date: Nvidia reports its second quarter on August 26, 2026. In the quarter before, $15.9 billion of $58.3 billion in net income came from equity gains, a good 27 percent, and its holdings of non-marketable equity securities nearly doubled within three months from $22.3 billion to $43.4 billion[11]. The $152 billion in this article is therefore probably not the full number for this quarter.*

Disclaimer · Not investment advice · Disclosure

* This site is run by Max Fraunhofer as a private journalist and holds no BaFin authorisation under § 32 KWG (German Banking Act). This article is journalistic analysis, not investment advice within the meaning of § 1 (1a) sentence 2 no. 1a KWG / § 2 (2) no. 4 WpIG, and not an investment recommendation within the meaning of § 85 WpHG in conjunction with Article 20 of the Market Abuse Regulation. All figures come from public press releases, SEC filings, earnings call transcripts and media reports (see source list). Where evidence is only indirect, this is marked in the text and in the source list. Past performance is not an indicator of future results. Anyone making investment decisions on the basis of this article does so at their own risk and should consult a licensed financial or tax adviser.

Disclosure (conflict of interest): At the time of publication the author holds direct equity positions in, among others, Alphabet, Amazon, Microsoft, Meta, ASML and TSMC, as well as broadly diversified index funds (including MSCI World) through which indirect holdings in other companies mentioned in this article exist. There are no direct positions in Nvidia or Oracle; there are no short positions in any of the securities mentioned.

Making-of · How this article came about
Making-of · as of August 5, 2026

An agent showed that my starting question was wrong before I could write it down

I started with the thesis that a considerable part of the cloud growth at AWS and Google Cloud was inflated by paper gains. One agent had the explicit job of refuting that, and it delivered: equity gains sit in „other income“ and cannot, under segment reporting rules, reach cloud revenue at all. The 82 percent growth at Google Cloud and the 36.7 at AWS are real, invoiced revenue. I would have measured the wrong object. The article was rebuilt from „inflated revenue growth“ to „inflated reported profit“, and that is the story that holds.

The pipeline behind this run
PHASE 01
Six research agents in parallel
Amazon figures, Alphabet figures, the other side (Anthropic and OpenAI), off-balance-sheet structures, accounting methodology, plus an explicitly adversarial run against my own thesis.
PHASE 02
Thesis rebuilt
The adversarial agent knocked out the revenue part. Three objects were separated: group profit, segment revenue, funding. Only the first carries this article.
PHASE 03
Microsoft as the control case
A second run for Microsoft alone. The result was the paradox: $5bn in Anthropic produces $3.2bn in paper gains, $11.9bn in OpenAI produces 0.48. That established the explanation via accounting method.
PHASE 04
Fact-check against primary sources
Two independent verification runs. The earnings release PDFs from Amazon and Alphabet were read directly, not through secondary sources. One figure was corrected, several were downgraded in evidence strength.
PHASE 05
Reviewers and gates
Article fact-check, UI and SEO review, legal review because listed companies are named in a risk context, German copy-editing pass, typography gate.

Method behind the plausibility calculations [△]

1. Amazon's paper gain after tax, around $41.6bn. Amazon reports the pre-tax figure ($53.4bn), not the after-tax one. The 10-Q documents the reported tax expense at $15.9bn for the first half and $4.1bn for the first quarter. The difference of around $11.8bn is the second quarter's tax share. 53.4 minus 11.8 gives around 41.6. Amazon does not state this figure itself.

2. Microsoft's paper gain after tax, around $3.1bn. Microsoft gives two figures that do not share a basis: the OpenAI effect at $480m explicitly as an effect on net income, that is, after tax, and the Anthropic gain at $3.2bn with no tax figure. For the $3.2bn, Microsoft's effective tax rate for the quarter of 18.8 percent is applied here ($8,281m tax expense on $44,047m pre-tax income), giving around $2.6bn. Together around $3.1bn. Microsoft does not state this combined figure itself, and it does not say whether the $3.2bn is meant pre-tax or after tax. If it were an after-tax figure, adjusted growth would be around 12 rather than 14 percent.

3. Growth without paper gains, around 22, 29 and 14 percent. Both periods were adjusted here, not just the current quarter. This is the methodologically decisive point: the prior-year quarter also carried equity effects (Amazon $1.117bn pre-tax, Alphabet $1.286bn pre-tax, Microsoft a loss of $1.575bn after tax). Stripping them only out of the current quarter compares an adjusted figure with an unadjusted one. At Amazon and Alphabet this makes adjusted growth too low, at Microsoft too high, because a loss depressed the base there last year. An earlier version of this article adjusted only one side and therefore arrived at 16 and 24 percent.

4. Share of the increase in profit, around 92, 91 and 54 percent. Difference between the after-tax paper gains of both quarters, divided by the difference in net income. For Alphabet the after-tax figure for the current quarter ($77.1bn) is official, all others are derived. No source states these ratios.

5. The $50bn from financial investors in Chart 04. A difference calculation: $65bn total volume minus the $15bn of previously committed hyperscaler investments that Anthropic states explicitly. Anthropic does not disclose the individual amounts from financial investors, only their names. The figure is therefore a residual, not a sum of individual items.

6. The time series in Chart 03 is not a calculation but consists of values read off one line per company across six earnings releases[15]. It is deliberately pre-tax, because only the pre-tax figure is disclosed for all six quarters at both companies. The after-tax figures used elsewhere in the article are therefore not directly comparable.

7. The definition used at Amazon, and why it shifts the result. Amazon does not report a separate „paper gain“ item. Used throughout here is the line „Other income (expense), net“. It is somewhat broader than the pure Anthropic markup: for the second quarter of 2026, Amazon states around $50.5bn in markups on holdings in private companies, while the line as a whole is $53.415bn. The difference consists of smaller valuation items, mostly also without cash effect. The choice is not neutral: with the narrower definition, Amazon's adjusted growth would be around plus 32 rather than plus 22 percent, because less is stripped out. The article uses the broader line because its question is how much of the profit is not a payment, not how much of it is attributable to one single holding.

8. The free cash flow of minus $13.5bn. Alphabet's quarterly figure of minus $5.855bn appears in its reconciliation table. Amazon's release only discloses the trailing twelve month figure (minus $7.604bn). The quarterly figure of minus $7.7bn used here is calculated from the quarter using Amazon's own definition: operating cash flow 45.387 minus purchases of property and equipment 54.208 plus proceeds and incentives 1.132.

What this article explicitly does not know

Five places where the research has no complete answer: Microsoft gives no tax rate for the $3.2bn Anthropic gain, so the after-tax figure in Chart 05 rests on an assumption · Alphabet gives only an aggregate in its earnings press release and does not break out how much of the $99.0bn relates to Anthropic and how much to SpaceX (the attribution comes from Bloomberg, not from Alphabet) · Microsoft's exact percentage stake in Anthropic is not disclosed, only the amount invested · whether the reported tax expense on the paper gains becomes cash-effective or remains a deferred tax could not be established from the press releases, so the article states only the earnings effect and makes no claim about tax payments · the lease figures for all three companies come from secondary sources quoting the filings, a direct comparison with the original text was not possible due to access restrictions.

Also open: of the $15bn in hyperscaler money in the Series H, only the $5bn from Amazon is attributed by name. Anthropic does not say who provided the other $10bn.

Reviewer pass: two independent fact-check runs against primary sources before publication, UI and SEO review, legal review, German copy-editing pass, deterministic typography gate.

What the last verification run found. After the revision, a further run was made with the explicit brief to refute my own figures rather than confirm them. It found three genuine errors, corrected here: the article claimed Microsoft's Anthropic gain had only been mentioned verbally on the earnings call, when in fact it appears verbatim in the press release. It described $135 billion as the book value of Microsoft's OpenAI position, when that is a market value estimate and $11.9 billion is what has been paid in. And it truncated a quote from EY at a point where the omitted clause turns the statement in a different direction. All three places are marked above and in the text.

SourcesEvidence, filings, transcripts

Convention in this article: hard figures, quotes and external model outputs are marked in the text with a superscript source number. Clicking jumps straight to the entry. Three levels of evidence are visually distinguished:

[N] hard primary source (earnings release, SEC filing, official press release) [N°] analyst estimate, transcript quote or secondary report [] own plausibility calculation, method in the making-of
  1. 01Amazon · earnings release Q2 2026 (July 30, 2026) · Net income $62.647bn, other income (expense), net $53.415bn with the literal addition „primarily from our investments in Anthropic“. Free cash flow TTM minus $7.604bn against plus $18.184bn a year earlier. AWS $42.232bn at plus 36.7 percent. Long-term debt $65.648bn (Dec 31, 2025) to $128.894bn (June 30, 2026). Primary source read directly.
  2. 02Alphabet · earnings release Q2 2026 (July 22, 2026) · Net income $112.193bn, „gain on equity securities, net“ $99.031bn. Footnote verbatim: the gain increased the provision for income tax, net income and diluted EPS by $21.9bn, $77.1bn and $6.26 respectively. Free cash flow Q2 minus $5.855bn. Equity raise June 2026 of $49.6bn net plus a $40bn ATM programme. Google Cloud $24.768bn at plus 82 percent. Primary source read directly.
  3. 03Microsoft · FY2026 Q4 results (July 29, 2026) · Revenue $90.0bn, net income $35.766bn, Azure plus 43 percent, free cash flow around $19.6bn, other income (expense), net $3.444bn against minus $1.707bn a year earlier. Verbatim on the discrete items: „These include a $3.2 billion gain from our investment in Anthropic“. OpenAI effect plus $480m in the quarter as an effect on net income, against a loss of $1,575m in the prior-year quarter. Microsoft additionally publishes net income adjusted for OpenAI: $35,286m against $28,808m, or plus 22 percent.
  4. 04CNBC · „Big Tech's Anthropic and OpenAI stakes are distorting the corporate earnings picture“ (August 3, 2026) · Tajinder Dhillon (LSEG) verbatim: „Without those gains from just Alphabet and Amazon's private company stakes, aggregate earnings growth would be sitting around 29%“, against around 48 percent reported. Also contains Gil Luria (D.A. Davidson) on the point that analysts already strip these items out. Data provider analysis via a media report.
  5. 05Anthropic · Series H press release (May 28, 2026) · $65bn in fresh capital, $965bn post-money valuation. The price event that triggered the revaluations at Amazon, Alphabet and Microsoft in the same quarter. On the origin of the money, verbatim: „It also includes $15 billion of previously committed investments from hyperscalers, including $5 billion from Amazon.“ Co-led according to the announcement by Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ and XN, other named investors include Blackstone, Brookfield, Fidelity, Jane Street, Lightspeed, MGX, T. Rowe Price and Temasek. Google appears not as a funder but through a commitment for five gigawatts of TPU capacity together with Broadcom.
  6. 06Bloomberg · Alphabet's Anthropic stake at around $124bn (July 23, 2026) · Alphabet put the fair value of an unnamed holding in a private company at around $124.3bn as of June 30, 2026 in a separate filing. The attribution to Anthropic comes from Bloomberg, not from Alphabet. Secondary sources describe the filing type inconsistently (8-K or 10-Q), a direct comparison with the original was not possible.
  7. 07TechCrunch · Microsoft books $3.2bn from its Anthropic investment (July 29, 2026) · Context and classification: Microsoft invested around $5bn in Anthropic in November 2025. The investment amount itself comes from this reporting. Secondary report. The $3.2bn is not evidenced from here but from Microsoft's own press release[03], where it appears verbatim. An earlier version of this article wrongly presented the figure as a pure earnings call quote.
  8. 08Amazon · additional investment in Anthropic (April 20, 2026) · $5bn immediately on top of the $8bn paid so far, with up to $25bn more committed. Cumulative actually paid in: $13bn. In return, a compute commitment from Anthropic to AWS.
  9. 09Bloomberg · OpenAI valued at $852bn after completing a $122bn round (March 31, 2026) · Amazon the largest single investor at $50bn, Nvidia and SoftBank at $30bn each. According to the reporting, Amazon's share is structured in tranches, with around $15bn immediate and $35bn tied to conditions. Secondary report.
  10. 10Silicon Analysts · hyperscaler capex against cash flow (July 10, 2026) · Comparison across providers: Alphabet 63, Meta 61, Microsoft 57 percent capex to operating cash flow. Independent analysis by a specialist blog, not bank research. Amazon's own ratio in this article is calculated from the primary source, not taken from here.
  11. 11Nvidia · Q1 FY2027 results (May 20, 2026) · Revenue $81.615bn, net income $58.321bn, of which $15.9bn from equity gains ($13.4bn listed, predominantly CoreWeave, $2.6bn non-listed). Next report on August 26, 2026. Holdings of non-marketable equity securities $22.251bn to $43.364bn.
  12. 12Bloomberg · Microsoft reports over $130bn in new data centre leases (July 29, 2026) · Signed but not yet commenced lease obligations at around $329.1bn as of June 30, 2026, after $196.6bn as of March 31, 2026. Secondary source quoting the filing, no direct comparison possible.
  13. 13Deloitte DART · roadmap to ASC 323, equity method investments · The presumption of significant influence from around 20 percent of voting rights and its consequence for earnings recognition. Additionally PwC Viewpoint on ASC 321 and the measurement alternative for holdings without readily determinable fair values. Technical literature on the accounting standard, not a company statement.
  14. 14EY · Technical Line on ASU 2016-01, recognition and measurement of financial assets and financial liabilities · Effective date verbatim: „The guidance is effective for calendar-year public business entities beginning in 2018“, for fiscal years beginning after December 15, 2017. On the abolition of the cost method, verbatim: entities „no longer use the cost method of accounting as it was applied before they adopted ASU 2016-01 for equity securities that do not have readily determinable fair values. As such, the new guidance could significantly increase earnings volatility for some entities.“ Technical publication by an audit firm on the standard, not a company statement.
  15. 15Time series in Chart 03 · earnings releases for the six quarters, each read directly: Amazon Q1 2026, Q4 2025, Q3 2025 · Alphabet Q1 2026, Q4 2025, Q3 2025. Values for Q1 and Q2 2025 from the comparative columns of the same documents. Amazon: line „Other income (expense), net“ (2.749 · 1.117 · 10.186 · 1.177 · 15.647 · 53.415). Alphabet: line „Gain (loss) on equity securities, net“ (9.758 · 1.286 · 10.734 · 2.302 · 36.915 · 99.031). Alphabet additionally states the after-tax effect per quarter in its footnote, $1.8bn for Q4 2025 and $28.7bn for Q1 2026.
  16. 16Microsoft · 10-K annual report for fiscal year 2026 (as of June 30, 2026) · On the OpenAI stake, verbatim: „We have an investment accounted for under the equity method that represents an approximate 25% interest on an as-converted basis“ and „We have made total funding commitments of $13.0 billion related to our investment, of which $11.9 billion has been funded as of June 30, 2026.“ On the measurement method: „We calculate our equity method income or loss using the hypothetical liquidation at book value (‚HLBV‘) method because our liquidation rights and priorities differ from our underlying ownership interest.“ Primary source read directly.
  17. 17Constellation Research · Microsoft's OpenAI stake and its estimated value · Basis for the circulating order of magnitude of around $135bn for Microsoft's OpenAI position. Secondary source with a market value estimate. That value does not appear as such in Microsoft's balance sheet: under the equity method the stake is carried at amounts paid in less losses absorbed, not at a market value. An earlier version of this article wrongly described the $135bn as a book value.