Q2 2026: the quarter AI capex reached the cash flow statement

The AI capex debate has been conducted for three years mostly in guidance numbers — forward-looking ranges announced on earnings calls, revised upward, argued about. Guidance is a statement of intent. The second quarter of 2026 is the first reporting period where the spending shows up in audited cash flow statements at a scale that changes the sign on the bottom line, and that is a different kind of evidence.

Alphabet reported negative free cash flow of $5.9 billion for the quarter. Amazon reported negative free cash flow of $7.6 billion on a trailing-twelve-month basis, against a positive $18.2 billion in the comparable period a year earlier. Meta’s quarterly free cash flow came in at $784 million — positive, but on $60.8 billion of revenue. These are not analyst estimates or adjusted metrics; they are figures the companies published in their own releases.

At the same time, two of the four headline profit numbers were driven overwhelmingly by things that are not operations. The gap between those two facts is the story of the quarter.

The reported numbers

Alphabet announced Q2 results on 22 July: consolidated revenues of $119.8 billion, up 24%, with Google Cloud accelerating to $24.8 billion on 82% growth and consolidated operating margin at 34%. Purchases of property and equipment were $44.9 billion for the quarter against $39.1 billion of operating cash flow — hence the negative free cash flow. Trailing-twelve-month free cash flow remained positive at $53.3 billion, which is the number to watch for whether the quarter was an inflection or a timing artifact. On the earnings call, CFO Anat Ashkenazi raised full-year 2026 capex guidance to $195–205 billion, up from a prior $180–190 billion, describing an environment of industry-wide supply constraints.

Microsoft’s fiscal Q4, reported 29 July, was the cleanest of the four: revenue of $90.0 billion up 18%, operating income of $40.6 billion up 18%, Microsoft Cloud revenue of $59.3 billion up 27%, and Azure growth of 43%, with Satya Nadella noting Azure revenue passed $100 billion for the first time. Capital expenditures were $35.8 billion in the quarter and $115.9 billion for the fiscal year, against $55.4 billion of quarterly operating cash flow. Microsoft is the only one of the four whose operating cash generation still comfortably exceeds its capital spending.

Meta reported the same day: revenue of $60.8 billion up 28%, net income of $15.8 billion down 14%, diluted EPS of $6.18 down 13%, and capital expenditures including finance-lease principal of $31.1 billion. Full-year capex guidance narrowed to $130–145 billion from a prior $125–145 billion — a raise at the floor, not the ceiling. Meta is the one company in the group where AI spending is now visibly compressing reported earnings rather than being absorbed by them.

Amazon closed the week on 30 July with net sales of $200.6 billion up 20%, operating income of $27.5 billion up 43%, and AWS at $42.2 billion on 37% growth — its fastest in eighteen quarters — with AWS segment operating income of $16.6 billion. Purchases of property and equipment were $54.2 billion in the quarter against $32.2 billion a year earlier. On the call, Andy Jassy raised 2026 cash capex guidance to roughly $220 billion, attributing part of the increase to higher memory chip costs and telling investors to expect free cash flow headwinds until the data centers come online and get monetized.

Summing the four quarterly capital-spending figures as each company reports them gives roughly $166 billion in a single quarter. The definitions are not identical — Meta includes finance-lease principal payments, Microsoft’s headline capex line does not, and trade coverage of Microsoft’s call put capital spending including finance leases nearer $41 billion for the quarter. Anyone comparing these four numbers without checking the definitions is comparing three different things.

The profit line is doing something unusual

Alphabet’s earnings per share rose 294% to $9.11. That number is almost entirely not about Google. The release attributes a net gain of $98.0 billion in other income “primarily the result of net unrealized gains on our equity securities.” Strip that and the underlying operating story — 24% revenue growth, 34% operating margin, Cloud accelerating — is strong but nothing like a quadrupling.

Amazon’s net income of $62.6 billion includes $53.4 billion of non-operating pre-tax other income, primarily from its investments in Anthropic. Amazon’s operating income for the quarter was $27.5 billion. The majority of the reported profit came from marking up a stake in a private AI company, not from selling anything.

This is not an accounting irregularity — both companies disclosed it plainly in the first page of their releases, and the accounting treatment is standard. But it produces a specific distortion in how the quarter reads. Headline profits at two hyperscalers were inflated by the appreciation of AI-sector equity, in the same quarter their cash flow was being consumed by AI-sector capital spending. Mark-to-market gains on AI holdings are financing the optics of an AI buildout that is draining actual cash. Those gains reverse if sentiment turns; the data centers do not un-build.

Anyone using Q2 2026 EPS to argue that the AI buildout is already paying for itself is, in two of four cases, citing unrealized valuation changes as evidence of operating return. The capex-versus-revenue gap argument does not get resolved by a number like that; it gets obscured by it.

Where the money is coming from

If operating cash flow no longer covers capital spending, the balance has to be financed. It is being financed.

Amazon priced a $25 billion bond on 7 July across eight tranches with maturities out to forty years, and told its underwriters it did not plan to issue further debt this year. FactSet’s analysis of hyperscaler financing, published 23 July, put the structural shift in one figure: incremental annual debt rose from 9% of capex in FY24 to 32% on a trailing-twelve-month basis by mid-2026, with aggregate total debt across the five largest operators reaching approximately $700 billion and lease-related commitments not recognized as balance sheet liabilities totalling roughly $820 billion.

That last number deserves attention, because it is the part of the buildout that does not appear in the capex line at all. Microsoft’s disclosed plan to move more future data center leases from finance leases to operating leases — alongside extending the estimated useful lives of data centers and office buildings from 15 to 25 years — is an accounting change with a real effect on how large the spending looks. Finance leases sit inside reported capital expenditures. Operating leases do not. Longer useful lives spread depreciation over more quarters. Neither change alters a single dollar of underlying commitment; both make the reported numbers smaller and smoother.

This is worth flagging without accusation. Both changes are defensible on their merits — server and building lifespans are an empirical question, and lease classification follows the contract terms. But an industry whose capital intensity is under scrutiny, adjusting the presentation of that capital intensity in the same quarter, is a thing readers should be able to see stated plainly.

What the operating businesses actually show

The bearish read is not that the underlying businesses are weak. They are not. Google Cloud at 82% growth, AWS at 37% and accelerating, Azure at 43%, Microsoft Cloud at $59.3 billion in a quarter — these are large businesses growing at rates that would be remarkable at a tenth the size. Amazon’s AWS segment alone produced $16.6 billion of operating income in the quarter. Demand for AI infrastructure is not hypothetical; three of the four explicitly said supply is the binding constraint, not demand.

The tension is one of duration. Capital is being deployed now, at a pace that has flipped free cash flow negative at two of the four, against revenue that arrives over the useful life of the assets. Jassy’s framing on the call — spend heavily and accept free cash flow headwinds until the capacity is online and monetized — is a coherent statement of that trade. It is also unfalsifiable in the short run, which is precisely why it should be tracked against outcomes rather than accepted or dismissed on rhetoric.

The measurable version of the question is: does trailing-twelve-month free cash flow at these companies recover as the 2025–2026 capacity comes online, or does the next guidance revision push the recovery out another year? Alphabet’s TTM free cash flow of $53.3 billion is the baseline to measure the next four quarters against. Amazon’s is already negative.

What changes the picture going forward

Three things would move this materially, and all of them are countable.

First, the direction of the next guidance revision. Every 2026 revision so far has been upward: Alphabet from $180–190 billion to $195–205 billion, Amazon to roughly $220 billion, Meta lifting its floor. A quarter in which one of the four guides down — for any reason, including component costs or capacity discipline — would be the first evidence that the spending has an internally recognized ceiling.

Second, whether debt share keeps climbing. Going from 9% to 32% of capex funded by incremental debt in under two years is the fastest-moving number in the group. If it keeps rising while free cash flow stays negative, the buildout becomes a credit story rather than an equity story, and credit markets price duration risk differently than equity markets price growth.

Third, whether the non-operating gains hold. Alphabet’s $98.0 billion and Amazon’s $53.4 billion are marks, not receipts. They were struck in a quarter when AI-sector valuations were high. If those marks compress, the same two income statements that showed 294% EPS growth will show the reverse, and they will show it while the capital spending continues on schedule. That asymmetry — gains that can reverse funding assets that cannot — is the specific fragility this quarter introduced, and it is the one worth watching through year-end.