Three clouds, one quarter: $135 billion of capex, $258 billion of debt, and the revenue that has to carry it

Every figure below traces to a filed document: Amazon's Q2 2026 earnings exhibit (8-K 0001018724-26-000024, filed 2026-07-30) and 10-Q (0001018724-26-000026, 2026-07-31); Alphabet's Q2 2026 exhibit (8-K 0001652044-26-000066, 2026-07-22); Microsoft's Q4 FY2026 exhibit (8-K 0001193125-26-323632) and FY2026 10-K (0001193125-26-323660, both 2026-07-29); Meta's Q2 2026 exhibit (8-K 0001628280-26-050596, 2026-07-29). Where a number comes from an earnings call rather than a filing, it is labeled as such. No street consensus is quoted. Research, not investment advice.

The three biggest clouds reported the same ninety days within eight days of each other in late July. Each print was covered on its own; the comparison mostly was not. Read side by side, the three filings answer a question the headline numbers dodge: the build-out is being paid for in three different currencies — Amazon's is borrowed money, Alphabet's is freshly issued stock, Microsoft's is the operating cash of a business that spends less than it earns. The revenue lines that have to carry that spending are growing at three different speeds too.

This piece walks the ledger in order: where the revenue comes from, what the cloud segments earn, what the capex did to cash, how each company funded the gap, what is signed but not yet recognized, and one line in the income statement that has quietly changed what "net income" means at all three.

Where the revenue comes from

Amazon ($AMZN) is still, by revenue, a retailer: North America and International stores are 79 percent of sales. But the operating income split tells the opposite story. AWS earned $16.6 billion of the company's $27.5 billion, a 39.4 percent segment margin, on 21 percent of revenue. The exhibit calls the 37 percent AWS growth rate "its fastest growth in 18 quarters" and puts the segment on a $169 billion annualized run rate. Amazon also files two AI numbers most of its peers do not: an AWS AI business that "exceeded a $25 billion annual revenue run rate, growing triple-digit percentages year-over-year," and a chips business at the same $25 billion mark.

Alphabet ($GOOGL) has the cleanest advertising engine of the three — Google advertising was $81.6 billion, up 14 percent, with Search & other up 17 percent and YouTube ads up 13 percent — and the fastest-growing cloud: Google Cloud revenue rose 82 percent to $24.8 billion and its operating income more than tripled, from $2.8 billion to $8.8 billion, a 35.6 percent segment margin. The exhibit attributes the acceleration to "enterprise AI Solutions and enterprise AI Infrastructure, as well as core GCP services." It does not put a dollar figure on AI revenue. The subscriptions, platforms and devices line, easy to overlook, was $12.9 billion, up 15 percent — larger than YouTube ads.

Microsoft ($MSFT) is the only one of the three where cloud is the largest segment. Intelligent Cloud was $39.3 billion, up 32 percent; across segments, "Microsoft Cloud" revenue was $59.3 billion, up 27 percent. The exhibit gives Azure a growth rate — 43 percent — and, for the fiscal year, a milestone: Azure "surpassed $100 billion for the first time." It does not give Azure a quarterly dollar figure, and there is no AI revenue run-rate in the July exhibit or the 10-K; the last one filed was $37 billion, up 123 percent, in the April Q3 exhibit (0001193125-26-191457). Productivity and Business Processes did $37.8 billion, up 14 percent; More Personal Computing shrank 4 percent to $12.9 billion.

So: three companies, three mixes. Amazon's profits already lean on the cloud far more than its revenue does. Alphabet's cloud is small but compounding fastest and now earns a real margin. Microsoft is the only one where the cloud is simply the business.

What the capex did to cash

This is the table that the earnings-day coverage tends to skip. Capex here is the cash-flow line ("purchases of" or "additions to property and equipment"); finance-lease additions, which all three report separately, are excluded.

Combined, the three spent $134.9 billion on property and equipment in one quarter, up from $71.7 billion a year earlier. Every one of them roughly doubled.

Two of the three could not fund that from operations. Alphabet's own free-cash- flow reconciliation shows negative $5.9 billion for the quarter, against $24.5 billion in the September 2025 quarter; Amazon does not file a quarterly FCF but its trailing-twelve-month figure is a $7.6 billion outflow, versus an $18.2 billion inflow a year earlier, and the exhibit attributes the swing to "a year-over-year increase of $66.1 billion in purchases of property and equipment" that "primarily reflects investments in artificial intelligence." Microsoft's operating cash flow of $55.4 billion covered its $35.8 billion of additions with room to spare, and it returned $10.2 billion to shareholders in the quarter on top.

The depreciation column is the one to watch next year. Amazon's D&A is already $20 billion a quarter, up a third; Alphabet's depreciation of property and equipment is up 42 percent. Capex becomes an expense with a lag, and the lag is now short enough to show.

Who is paying for it

Amazon's long-term debt nearly doubled in six months. The 10-Q puts senior notes outstanding at $132.1 billion. The cash flow statement shows $67.0 billion of long-term debt proceeds in the first half against $2.8 billion of repayments. This is the plainest financing of the three: borrow, build.

Alphabet did something the other two did not. The exhibit's "Equity Capital Raise" note discloses that in June the company "issued a combination of Class A stock and Class C stock and mandatory convertible preferred stock for aggregate net proceeds of $49.6 billion, to be used for general corporate purposes, including capital expenditures to scale AI infrastructure and global compute." Add $20.3 billion of notes and the balance-sheet cash line rose by $116 billion between December and June. A company that reported $112 billion of net income raised $70 billion from outside investors in the same ninety days — which only makes sense once you see the negative free cash flow above and the capex run-rate behind it.

Microsoft's balance-sheet debt fell. Long-term debt is $31.1 billion, total debt $40.3 billion, and the cash flow statement shows no new issuance. The caveat we documented in the shadow-debt map still applies: the 10-K carries $66.6 billion of finance-lease liabilities under "other liabilities," and a much larger schedule of data-center leases signed but not yet commenced. On the filed debt line, though, Microsoft is the only one of the three funding the build-out from operations.

The order books: what is signed but not yet revenue

Amazon's number is the one that moved. A year earlier the same footnote read "approximately $195 billion" with a 4.0-year life; it is now $496 billion with a 6.4-year life. The 10-Q names the two contracts behind the jump: an expansion of the OpenAI commitment "by $100.0 billion over 8.0 years" in Q1, and an expansion of the Anthropic commitment "by more than $100.0 billion over 10.0 years" in Q2, both including "contractual obligations related to the performance of AWS chips." The footnote also carries its own warning: revenue timing "will be driven by customer usage," which "can extend beyond the original contractual duration and commitment." An order book is not a schedule.

Microsoft's $678 billion is the largest filed figure, and it grew $51 billion in the single quarter since March. Alphabet's backlog is the one we cannot cite from a filing yet; the Q2 10-Q was not indexed in our filings corpus at the time of writing, so the $514 billion stands as management's spoken number.

Taken at face value, roughly $1.2 trillion of cloud demand is signed across Amazon and Microsoft alone — the clearest filed evidence that the $135 billion quarter of capex has a buyer. The same three footnotes are also where concentration lives: two AI labs account for at least $200 billion of Amazon's book.

The line that changed what net income means

All three income statements now carry a number that has nothing to do with selling cloud.

• Amazon's net income was $62.6 billion, more than triple last year's $18.2 billion. The exhibit states that it "includes non-operating pre-tax other income of $53.4 billion, primarily from our investments in Anthropic." Operating income was $27.5 billion.
• Alphabet's net income was $112.2 billion on $40.8 billion of operating income. "Other income reflected a net gain of $98.0 billion, primarily the result of net unrealized gains on our equity securities."
• Microsoft's net income rose 31 percent on a GAAP basis but 22 percent non-GAAP; the exhibit lists "a $3.2 billion gain from our investment in Anthropic" among the discrete items and excludes the OpenAI investment impact from its non-GAAP figures.

Together that is more than $150 billion of pre-tax gains in a single quarter, generated by the rising private-market valuations of AI labs that are, at the same time, among the largest customers in the order books above. Amazon's 10-Q Anthropic commitment and Amazon's Anthropic gain are two sides of one relationship. This is not a comment on whether the marks are right; it is a reminder that the headline EPS at two of the three is now mostly a valuation event, and that operating income — $27.5, $40.8 and $40.6 billion — is the number that can be compared across quarters.

The buyer's side of the same trade

Meta ($META) filed its quarter the same evening as Microsoft, and it is the useful control group: a company spending at hyperscaler scale without a cloud business to rent the capacity out. Revenue was $60.8 billion, up 28 percent, with advertising at $59.4 billion; ad impressions grew 14 percent and average price per ad 12 percent. Capital expenditures including finance-lease principal were $31.08 billion, operating cash flow $31.86 billion, and free cash flow $784 million. Operating margin fell from 43 to 31 percent. Long-term debt was $83.66 billion. The full-year capex outlook was narrowed to $130–145 billion.

Meta's ads grew faster than Google's in the quarter (27 versus 14 percent) and it still ended with a quarter of near-zero free cash flow, because on Meta's ledger the AI build-out is pure cost until new products monetize it. On Amazon's and Microsoft's ledgers the same industry-wide spending shows up as a $496 billion and $678 billion order book. Same capex cycle, opposite side of the invoice.

What a filed reader takes from this

Growth is real at all three, but its shape differs. AWS at +37 percent is the biggest cloud accelerating; Google Cloud at +82 percent is the smallest compounding fastest and now earning a 36 percent margin; Azure at +43 percent is the one that has already crossed $100 billion a year.

The funding question is answered differently at each. Amazon borrowed (long-term debt $65.6 billion to $128.9 billion in six months). Alphabet sold stock ($49.6 billion, plus $20.3 billion of notes) and ended with $242 billion of cash. Microsoft paid from operations and let its debt line fall. Which of those you prefer is a question about cost of capital and dilution, not about AI.

Free cash flow has gone negative at two of three. Alphabet's −$5.9 billion is filed; Amazon's −$7.6 billion is the trailing-twelve-month figure. Neither company describes this as a problem, and the order books argue it is an investment rather than a leak. But the depreciation lines — up 31 to 42 percent — are how that investment will arrive in the income statement, and they are already moving.

Net income is no longer the comparison metric. $53.4 billion at Amazon and $98.0 billion at Alphabet came from marking AI-lab stakes, not from selling compute. Compare operating income, segment operating income, and cash.

Read the footnote, not the headline. Amazon's $496 billion, Microsoft's $678 billion and the customer names inside them are in a Note 1 and an exhibit table respectively. They are also where the two-customer concentration sits. The numbers that decide whether $135 billion a quarter is prudent or reckless are filed; the debate mostly is not.

We never say buy or sell. Every figure above links to the page and accession number it came from; the valuation pages for each company show the seven-method fair-value range and cite the same filings. Research, not investment advice.