Alphabet Spent $1.15 of Capex for Every Dollar of Cash Flow. I De-Cumulated 260 Quarters to See Who Else Has.

TL;DR: AI circular financing is the loop where a supplier funds its own customer's purchases, and the version you can actually verify sits in the cash flow statement, not the press release. I pulled the SEC's XBRL data for five hyperscalers on 2026-08-15 and de-cumulated 260 quarters. Alphabet's Q2 2026 was $39.07B of operating cash flow against $44.92B of capex, a negative free-cash-flow quarter and the only one in the 49 quarters the XBRL record covers for them. Amazon has had 14 such quarters and Microsoft has had zero in 61, so this is not one story. The number worth tracking is capex divided by operating cash flow: for Alphabet, Microsoft, Meta and Amazon combined it went from 69% in Q2 2025 to 96% in Q2 2026.

The bubble argument usually arrives as narrative. It can be made out of filings instead, and filings are free.

What did I actually compute?

Two us-gaap concepts per company from the SEC's XBRL companyconcept API, which needs no account but returns 403 unless your User-Agent carries a contact email. Operating cash flow is NetCashProvidedByUsedInOperatingActivities for all five. Capex is PaymentsToAcquirePropertyPlantAndEquipment for Alphabet, Microsoft, Meta and Oracle.

Amazon is the exception and it matters. It stopped using that tag after Q1 2017 and reports the same line under PaymentsToAcquireProductiveAssets. Quietly swapping concepts would corrupt the comparison, so every Amazon capex figure below uses that tag, the gross purchases line.

Two definitional warnings. First, cash flow statements are cumulative within a fiscal year. Alphabet's Q2 10-Q shows $84,859M of operating cash flow and $80,598M of capex, but those are six-month numbers, and neither the $39.07B nor the $44.92B appears anywhere in the filing. You get them by subtracting the Q1 10-Q's $45,790M and $35,674M.

Second, free cash flow is not a GAAP measure. Mine is operating cash flow minus the gross capex tag. The SEC's own non-GAAP guidance calls that the typical calculation while warning it "should not be used in a manner that inappropriately implies that the measure represents the residual cash flow available for discretionary expenditures." My figures will not match the companies' own. Meta reported Q2 2026 free cash flow of $784 million because it adds finance-lease principal payments to capex; I get $1.75B because I do not. Amazon reported a trailing-twelve-month outflow of $7.6B on a definition that nets equipment-sale proceeds.

Does the de-cumulation hold up?

This is the step most likely to be wrong, so here is the check. Every derived quarter must sum back to the reported year-to-date total, and Q1 needs no subtraction. Across 163 fiscal-year chains, 0 failed. Then an independent test: filers sometimes tag a discrete quarter alongside the YTD one. Since January 2024 there were 46 such values, and 46 of 46 matched my derived number. Meta's press release is a third check: their $784M equals my derived $31.86B of operating cash flow minus their $31.08B capex.

Which companies have had a negative quarter?

Not the ones the narrative implies.

Company	Negative FCF quarters	Out of	Most recent
Alphabet	1	49 (from Q4 2013)	Q2 2026
Microsoft	0	61 (from Q4 FY2008)	never
Meta	2	62 (from Q4 2010)	Q2 2012
Amazon	14	37 (from Q4 2016)	Q2 2026
Oracle	9	51 (from Q4 FY2010)	Q4 FY2026

Alphabet's Q2 2026 is the only negative quarter in its XBRL history, and I am scoping that to what I checked: the 49 quarters the SEC's structured data covers, back to Q4 2013. Alphabet listed in 2004 and I did not pull the pre-XBRL filings, so "first ever" is not a claim I can make.

Amazon's 14 is the number that should calm anyone down. It went negative in Q1 of 2018, 2019, 2020, 2021, 2022, 2023, 2025 and 2026. That is what Amazon looks like during a buildout, and the company is still here. Oracle has gone negative in five of its last six. Microsoft has never done it in 61.

How fast is the ratio moving?

Faster than the level is alarming. Capex divided by operating cash flow, calendar 2024 against the most recent four quarters:

Company	2024	Last 4 quarters
Alphabet	42%	71%
Microsoft	44%	63%
Meta	41%	69%
Amazon	72%	107%
Oracle	53%	174%

All five moved the same direction, the part narrative cannot fake. For the four calendar-year filers combined, Q2 2025 was $128.48B of operating cash flow against $88.25B of capex: a 69% ratio, $40.23B of free cash flow. Q2 2026 was $171.76B against $165.05B: 96%, and $6.71B. Operating cash flow grew 34% year over year, a spectacular result. Capex grew 87%. That capex is also most of what US GDP growth currently consists of, on arithmetic almost nobody quoting it can reproduce.

Where is the missing money coming from?

Debt, at the two companies that tag it comparably. Amazon issued zero long-term debt across all of calendar 2024, $15.68B in 2025, and $66.99B in the first half of 2026 alone. Meta issued $29.91B in Q4 2025 and $24.91B in H1 2026. Alphabet and Microsoft return 404 on ProceedsFromIssuanceOfLongTermDebt, so they tag issuance differently and I am not comparing them here; Oracle's coverage is too sparse to use. Reporting four of five would be inventing a trend.

That is the honest answer to the circular financing question. The loop people worry about, a chip vendor underwriting its largest customer's purchases, is real but sits in the private layer around OpenAI, not in these five filings. What these show is more checkable: the hyperscaler buildout is still funded by operations and ordinary bond markets, and it has stopped being self-funding at the margin. That transition is what to watch, because a company funding capex from its own cash has infinite patience and one funding it from capital markets has a maturity schedule.

What does this not prove?

It does not prove anyone is in trouble. These are among the most cash-generative businesses ever assembled, and a negative quarter at Alphabet is a rounding event against $164.71B of operating cash flow in 2025. Nor does a high ratio mean the spending is wrong. Depreciation timing guarantees this shape: capex leaves the bank account the quarter the servers are bought, while the revenue they support arrives across their depreciable life, a life these same filers keep re-estimating in both directions. Any genuine buildout produces this pattern, and Amazon's 14 negative quarters prove it can be survived repeatedly.

It also does not prove the ratio keeps climbing. One quarter is not a slope, Alphabet's own Q2 2025 hit 81% before falling back to 49%, and quarterly capex is lumpy enough to make any single reading unreliable. And it says nothing about whether the demand is real, the argument the bubble debate turns on and which no cash flow statement can settle.

What number should you watch?

One ratio, four filings a year, no subscription. Take NetCashProvidedByUsedInOperatingActivities, subtract the prior quarter's cumulative figure to get the discrete quarter, do the same for the capex tag, and divide. Pull it from data.sec.gov/api/xbrl/companyconcept/CIK{10-digit}/us-gaap/{Tag}.json with your email in the User-Agent.

The threshold that matters is not zero. It is whether the ratio comes back down. Capex above 100% of operating cash flow is a bet that revenue arrives before the debt does, and it is survivable: Amazon has been above 100% in 12 separate quarters and is still here. What would change my mind about this cycle is four consecutive quarters above 100% at a company also issuing debt to cover the gap, because at that point the buildout is no longer financed by the business it is meant to serve. Alphabet has one such quarter. Amazon has averaged 107% across its last four. Oracle is at 174%.

The gap between those three numbers is why "is AI a bubble" is the wrong question to ask of a balance sheet. Prices keep ratcheting because the compute is expensive, and the compute is expensive because of the spending in that table. Whether it was worth it depends on revenue that has not arrived yet, the same open question underneath every pilot-failure statistic in circulation. The filings will tell you before the press releases do.