Alphabet, Amazon, Meta, Microsoft and Oracle took in $253.4 billion from new long-term debt in their latest nine reported months, equal to about 45% of the $560.4 billion they spent on property and equipment, by our arithmetic from their 10-Q and 10-K cash flow statements. In the same months a year earlier, the ratio was about 17%: $46.5 billion of debt proceeds against $280.9 billion of capex.
Capex doubled, and debt proceeds rose about 5.5 times. The $207.0 billion increase in borrowing is equal to about 74% of the $279.5 billion increase in capex, by our arithmetic. A post on X by @pequityresearch on 11 November put the figure lower: “Debt issuance YTD from hyperscalers represents ~36% of capex spend”. The post does not give its inputs, so we rebuilt the ratio from the filings.
Business Pill · Capex and Depreciation Explained
An evergreen Business Engineer explainer on capital spending and depreciation. It teaches the general idea only and does not discuss these filings.
The key insight: Across Alphabet, Amazon, Meta, Microsoft and Oracle, gross long-term debt proceeds came to $253.4 billion against $560.4 billion of capex in the latest nine reported months, about 45%, against about 17% a year earlier, by our arithmetic from their cash flow statements. Amazon and Alphabet were near two-thirds; Microsoft reported no debt proceeds in either window.
Update (12 Nov): the bond-market side, from PIMCO.CNBC reported on 12 November that AI-related borrowers account for almost half of year-to-date gross issuance of non-financial U.S. investment-grade debt with maturities beyond 15 years, citing a PIMCO note dated 10 November. CNBC also cited Goldman Sachs forecasts of around $250 billion of dollar-denominated bonds from AI hyperscalers this year, and estimates of $300 billion to $570 billion of debt issuance for the wider AI ecosystem, including chipmakers, utilities and data center developers.
Amazon’s 10-Q shows $115.1 billion of proceeds from long-term debt in the nine months to September, against $175.7 billion of purchases of property and equipment, about 66%. In the same months of 2025, Amazon took in $746 million of long-term debt, about 1% of its capex. Amazon also took in $20.7 billion of short-term debt and repaid $19.6 billion; we leave short-term borrowing out of the ratio.
Alphabet reported $85.2 billion of proceeds from issuance of debt, net of costs, against $132.0 billion of capex, about 65%, up from about 60% a year earlier. It also sold stock: $42.9 billion of common stock and $19.1 billion of mandatory convertible preferred stock in the nine months. Its 10-Q describes a June public offering with $20.5 billion of net proceeds and a concurrent $10.0 billion private placement to an affiliate of Berkshire Hathaway. Together, Amazon and Alphabet borrowed about two-thirds of their capex, $200.3 billion of $307.6 billion.
Gross long-term debt proceeds as a share of purchases of property and equipment, January to September 2025 and 2026 (Oracle: December to August), from the companies’ 10-Q and 10-K cash flow statements. Window sums and ratios are our arithmetic.
Oracle, Meta and Microsoft
Oracle took in $28.2 billion from senior notes, term loans and other borrowings over its last three fiscal quarters, December to August, against $63.6 billion of capital expenditures, about 44%, up from about 33%. In the June to August quarter, its cash flow statement shows $19.9 billion from stock sold through an at-the-market program and $4.2 billion of debt repaid. Our December to August figures are built from four Oracle filings.
Meta reported $24.9 billion of proceeds from long-term debt against $83.1 billion of capex in the nine months, 30%, after none in the same months of 2025. Microsoft is the outlier: its fiscal 2026 10-K shows proceeds from issuance of debt of 0, and it repaid $4.85 billion of debt from July to September, while its January to September capex came to about $106.0 billion. Our earlier piece found Microsoft alone had positive free cash flow in the latest quarter.
How We Matched the Windows
Alphabet, Amazon and Meta report on the calendar year, so their January to September figures come straight from the 10-Qs. Microsoft’s fiscal year ends on 30 June: we added its January to March quarter, its April to June quarter, which is the 10-K total less nine months, and its July to September 10-Q. Oracle’s fiscal year ends on 31 May, and its latest 10-Q runs to 31 August, so its window is December to August. The 2025 windows are built the same way.
The ratio uses gross long-term debt proceeds as each company labels them, against purchases of property and equipment, or capital expenditures in Oracle’s statement. New finance leases are not in either figure. The bond deals priced since 29 October fall after these windows, among them $25 billion from Microsoft and $15 billion plus €9 billion from Meta, as listed in five hyperscaler bond deals from the term sheets.
What the Cash Flow Line Leaves Out
Lease commitments do not pass through the debt line. Meta’s 10-Q lists about $355.73 billion of leases that had not yet commenced at 30 September. It puts its maximum exposure to loss in unconsolidated data center ventures at $68.4 billion, up from $46.0 billion at the end of 2025; $41.5 billion of that is residual value guarantees. Microsoft lists $364.6 billion of leases, mainly for datacenters, that had not yet commenced. We added up these commitments across the five in Big Tech’s $3.2T off-book bills.
Update, 12 Nov 2026, 20:05Z: Wells Fargo on the bond market. Bloomberg published a story at 19:45 UTC headlined “AI ‘Supply Shock’ Pushes Bond Market to Limit, Says Wells Fargo”. The article is paywalled and we have not seen its figures, so we do not report any. The headline points to borrowing of the kind this piece measures, though the story may cover wider AI-linked issuance. By our arithmetic from the filings, the five companies took in $253.4 billion of new long-term debt in nine months, against $46.5 billion a year earlier.
The Structural Read
As we read the cash flow statements, the build-out moved from being paid for mostly out of operating cash to being paid for in large part with new debt and, at Alphabet and Oracle, new stock: Alphabet sold $42.9 billion of common stock and $19.1 billion of mandatory convertible preferred stock in the nine months, and Oracle $19.9 billion of stock through its at-the-market program from June to August.
@pequityresearch on X, 11 Nov 2026
“Debt issuance YTD from hyperscalers represents ~36% of capex spend”
Three Implications
FOR THE TOTAL $253.4 billion of gross long-term debt proceeds against $560.4 billion of capex, about 45%, by our arithmetic; including Amazon’s $20.7 billion of short-term debt proceeds would make it about 49%.
FOR THE CHANGE Capex doubled from $280.9 billion, while debt proceeds rose about 5.5 times from $46.5 billion, by our arithmetic.
FOR THE OUTLIERMicrosoft’s fiscal 2026 10-K shows proceeds from issuance of debt of 0, and its July to September 10-Q shows $4.85 billion of debt repaid.
The framework says: “the capex race is no longer a corporate spending cycle. It is a sovereign-scale financing problem dressed in tech-company clothing — and the financial architecture is starting to invert.”
As we read it, a move from about 17 cents to about 45 cents of new long-term debt per capex dollar in a year is that inversion showing up in the cash flow statements. The framework predates these filings and does not discuss them.
Capex (purchases of property and equipment) and gross long-term debt proceeds, billions of dollars, January to September 2026; Oracle December 2025 to August 2026. Source: the companies’ 10-Q and 10-K cash flow statements; window sums and ratios are our arithmetic.
What Is Not Established
The ratio sets two cash flows side by side; it does not show that borrowed dollars paid for any particular asset, since cash is fungible. Amazon’s capex figure is before $3.5 billion of proceeds from property and equipment sales and incentives. Including Amazon’s short-term debt proceeds would lift the combined ratio to about 49%.
We did not read the analysis behind the @pequityresearch figure, so we cannot say whether the gap comes from its windows, its debt measure or its companies. The companies were not contacted for this piece.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Across the five, new long-term debt equalled about 45% of capex in the latest nine reported months, against about 17% a year earlier, by our arithmetic from the cash flow statements. Amazon and Alphabet borrowed about two-thirds of their capex; Microsoft borrowed nothing in either window.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Every figure comes from the cash flow statements and notes in the companies’ 10-Q and 10-K filings on SEC EDGAR: Alphabet, Amazon and Meta for the nine months to 30 September; Microsoft from its 10-Qs for the quarters to 31 March and 30 September 2026 and its fiscal 2026 10-K; Oracle from its 10-Qs for the quarters to 30 November 2025, 28 February and 31 August 2026 and its fiscal 2026 10-K.
Debt proceeds are gross long-term proceeds as each company labels them; capex is purchases of property and equipment, or capital expenditures for Oracle. Window sums and ratios are our arithmetic. The 36% figure is from a post on X by @pequityresearch, whose inputs we did not see. The companies were not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
