Amazon reported the fastest AWS growth since 2021 on the same print where its trailing twelve months of free cash flow went negative. Revenue came in at $200.61 billion, up 20 percent and above the $196.47 billion estimate; AWS posted $42.2 billion, up roughly 37 percent, a fifth straight acceleration. [1] [2] Trailing free cash flow, which was an $18.2 billion inflow a year ago, is now a $7.6 billion outflow, against $54.2 billion of quarterly capital spending. [1]
The cash-conversion question has its second company. The July 29 account of Meta keeping $784 million of $31.9 billion as its AI buildout ate the quarter asked whether the AI buildout still leaves anything behind; Amazon's answer is that it currently leaves less than nothing on a trailing basis. The market's answer, by contrast, was a shrug and a bid: shares rose 9 to 10 percent after hours. [1]
The grading is case by case, and this week's curve is steep. Microsoft's $331 billion year with Azure past $100 billion was rewarded; Meta's margin collapse was punished; Amazon's negative cash flow was applauded. The growth ranking explains some of the dispersion — Google Cloud at plus 82 percent, Azure at 43, AWS at 37 — but none of the cash logic. [2]
The beat that is mostly a mark
GAAP earnings per share of $5.75 requires an immediate companion: adjusted EPS of $1.97 against a $1.82 estimate. The distance between the two figures is a $53.4 billion pre-tax gain, primarily a mark on Amazon's Anthropic investment. [1] Nearly all of the GAAP beat is the market repricing an asset Amazon already held, not operations producing new cash. The adjusted $1.97 is the operating receipt; the $5.75 is the receipt plus the appraisal.
That appraisal discipline matters more this quarter than last because the cash ledger underneath it flipped. A company can defend negative free cash flow during a buildout if the buildout is converting — and Amazon's evidence for conversion is the backlog, which hit $496 billion, with the AI and chips units each passing a $25 billion annual run rate. [2] Backlog is a promise ledger: signed demand, unbilled and undelivered. It answers the question "will anyone buy this capacity" without answering "at what margin, against what memory bill, in which year."
The guidance the rally skipped
The third-quarter guide came in below consensus — $197 to $202 billion against a $204.1 billion estimate, a miss management attributed to the Prime Day calendar shift. [1] [3] Forecasts are not results, and a calendar-shift explanation is the kind that auditors of management rhetoric should note and file. On the call, executives raised 2026 capital-spending guidance from $200 billion to roughly $220 billion, blaming higher memory costs. [1] [3]
The memory line deserves attention beyond Amazon. When two of the largest buyers of compute on earth — Meta on its call, Amazon on this one — both name memory pricing as the pressure on their capex plans, the AI buildout's binding constraint is migrating from chips to the components around them. The hundred-year-flood language belongs to Apple's call later the same evening; Amazon's version is a $20 billion guidance bump. [3]
What the two frames each miss
CNBC's frame — AWS booming, capex up, stock soars — is accurate and incomplete: it underplays that the GAAP beat is mostly an investment mark and that the trailing cash figure crossed zero. [1] The fin-twit split runs the other direction in both halves: the backlog camp treats $496 billion of signed demand as the answer to the capex question, while the cash-flow camp notes Amazon is the second hyperscaler this season, after Alphabet, to post negative free cash flow outright. [1] [2] The first camp underplays that backlog is promises, not cash. The second underplays that AWS's acceleration is a real operating receipt beneath the flattered EPS.
A reader of only one frame misses the hinge fact: the market is now grading AI quarters on growth quality and ignoring cash conversion exactly until it doesn't. Meta found the cliff Wednesday. Amazon, on Thursday, was waved past it with a worse cash number and a better growth line. The difference between those two verdicts is the year's most consequential pricing question, and July 30's print did not answer it — it just made the question more expensive to get wrong.
-- THEO KAPLAN, San Francisco