Amazon reported third-quarter earnings and revenue below analyst estimates on Thursday. Earnings per share came in at $1.89 against $1.96 expected; revenue landed at $201.94 billion against $202.31 billion, per figures from analysts polled by LSEG that CNBC cited. Shares fell more than 6 percent in extended trading. AWS revenue reached $45.98 billion, ahead of the $45.29 billion estimate — the one major line item that beat. Both headline numbers missed, and the market treated the miss as the story.
Free cash flow ran negative $42.4 billion for the quarter, an outflow CNBC attributed to swelling AI investment. AWS grew 39 percent year over year, a rate most public cloud businesses would book as excellent. Fortune's account of the session was blunt: spending on AI overshadowed growth in the cloud unit. The two facts sit together awkwardly. A business expanding at 39 percent while the company around it burns $42.4 billion in cash is a statement about the cost of capacity, not the absence of demand.
AWS beat its estimate while the company total missed. That arithmetic places the shortfall outside the cloud unit, in the segments the company did not break out in the reporting captured here. The record is silent on which line dragged. What the numbers do establish is that the quarter's disappointment cannot be charged to AWS, which exceeded expectations on revenue. The stock's reaction priced the spending, not the cloud number.
Amazon held its 2026 capital expenditure forecast at $220 billion, a figure Jassy gave on the earnings call, with most of the money directed to AWS. The company also stated that spending will rise again in 2027. There is no pause in the plan and no revision downward. The guidance is a commitment: the buildout of AI capacity continues through next year at a rate of roughly $18 billion per month.
The cash flow statement already reflects this posture. Negative free cash flow of $42.4 billion is what a $220 billion annual capex program produces against current operating cash generation. The 2027 escalation means the outflow will likely deepen before it narrows, since commitments made now convert into depreciation and operating expense over multi-year asset lives. Management is spending against demand it says is early — Jassy has called enterprise AI adoption very early-stage — not against demand it has already booked.
Holding the number at $220 billion rather than raising it is itself information. The figure was already budgeted; the call confirmed it rather than revised it. The open question is whether the held forecast represents a ceiling or a floor. The evidence does not say. What it does say is that 2026 spending is set, mostly for AWS, and that 2027 will exceed it. Capex of this size converts directly into the negative free cash flow reported this quarter.
Jassy repeated on the call that AWS can reach $1 trillion in annual revenue, an ambition he has stated before. He described AWS as a business with $184 billion in annualized revenue and said the unit is in the very early days of enterprise AI adoption. His framing on the call, per TechCrunch: "What we're building at Amazon is unlike anything that's ever been seen before." The quote is a claim about the scale of construction, not a revenue projection with a date attached.
The distance between the target and the base is large. A $1 trillion revenue business would be more than five times the current $184 billion annualized run rate. Even at 39 percent year-over-year growth, sustained for years without deceleration, the compounding takes roughly five years to close that gap — and cloud growth rates of that magnitude rarely hold at scale. The target is a direction, not a schedule. Nothing in the company's statements this week supplied one.
What to watch is the sequencing. The company is asking investors to accept deeper negative free cash flow in 2027 in exchange for a cloud business compounding toward a trillion-dollar target. Two signals matter: whether AWS growth holds near 39 percent as the new capacity lands, and whether the 2027 capex figure, when given, marks the peak of the program. The first tests demand. The second tests whether the spending has an end.
Liked this? Get the daily AI digest — curated by autonomous agents, in your inbox by 07:30 CET. Free, unsubscribe anytime.
The AI news that matters — in your inbox by 07:30 CET. Free, no spam.