The question Wall Street spent six months debating was simple: can a company spending $220 billion a year on infrastructure actually make money doing it? Amazon answered that last night, and the answer came not from the revenue line, but from the margin line inside AWS.
Let’s be precise about what happened. AWS grew 37% year over year in Q2, its fastest growth in 18 quarters, reaching roughly a $169 billion annualized revenue run rate. Analysts had penciled in about 31% growth. The beat was not a rounding error. It was roughly six percentage points of upside on a business doing about $42 billion in a single quarter.
But the number that should command more attention is operating margin. AWS operating margin came in at about 39% for the quarter, a step up from Q1. The business is expanding margins while simultaneously absorbing the largest infrastructure buildout in Amazon’s history. That combination is not what a spending cycle looks like when returns are deteriorating.
The $220 Billion Argument
Amazon entered 2026 guiding to about $200 billion in capital expenditures. Last night, CEO Andy Jassy raised the company’s 2026 capex outlook to about $220 billion. The company tied the increase largely to higher memory costs. The bears will read that headline and stop there. That would be a mistake.
Jassy has addressed the cash flow squeeze directly, framing it as a matter of timing. Data centers require capital well ahead of demand, he has said, and the data-center assets can be monetized over decades. Amazon has run this playbook before. It did it building out fulfillment centers. It did it in the original AWS buildout. Both cycles looked financially ugly at the peak of the spend. Both rewarded investors who held through the depreciation overhang.
The free cash flow reversal is real and worth understanding clearly. Capital spending on a trailing 12-month basis reached $169 billion, up 64% year over year, driven by investments in AI infrastructure. Free cash flow on a trailing 12-month basis swung to an outflow of $7.6 billion from an inflow of $18.2 billion in the prior year period. That swing is the cost of positioning for a decade-long opportunity. The question is whether the opportunity justifies it.
Why the Backlog Settles the Debate
The word that changes everything in the Amazon AI story is backlog. AWS growth accelerated for the fifth consecutive quarter, backlog reached $496 billion, and margins expanded despite the unprecedented buildout. A $496 billion contractual backlog means the revenue is already sold. The debate over whether AI demand is real or speculative ends there.
The supply picture reinforces it. Despite the higher spending, Jassy warned that Amazon will still not have enough capacity to meet AI demand in 2026 or 2027. When a company spending about $220 billion tells you it still cannot satisfy existing demand, the constraint is supply, not customers. That is not a business with a demand problem. It is a business in a controlled sprint to catch up to customers already committed to spending.
The signed compute commitments are real: OpenAI committed to about 2 GW of Trainium capacity through AWS beginning in 2027, and Anthropic signed an agreement that secures up to 5 GW of capacity for training and deploying Claude, including Trainium capacity. Those are not letters of intent. They are contracted load that will land in AWS revenue before 2028.
The Hidden Business Nobody Is Pricing
Inside the AWS story is a smaller story that is growing faster and drawing almost no coverage. Amazon’s in-house chip business, spanning Trainium and Graviton, is becoming a standalone competitive force. Andy Jassy said Amazon’s AI business and its chips business each eclipsed run rates of more than $25 billion. A $25 billion annual run rate in custom silicon sits inside AWS as a cost advantage. Every Trainium chip deployed internally is a GPU Amazon does not have to buy from Nvidia.
Amazon has increasingly looked to highlight its in-house chips division, which includes the Trainium and Graviton brands, as a newer growth pillar. Its AI products like the Bedrock model marketplace have primarily been targeted for enterprises. The distinction matters: Trainium lowers the cost of training models, Bedrock provides enterprise access to those models, and together they create a flywheel that deepens customer lock-in on every new AI workload signed.
Jassy also told analysts that the company has a “clear line of sight to strong financial returns” on its AI bet and said AWS could eventually become a trillion-dollar annual revenue business over time given what he described as striking demand for the technology. He also said demand for 2028 is striking and that the company expects AWS to at least double its prior projection of becoming a $200 billion revenue business.
The Competitive Context
Amazon did not report in a vacuum. Alphabet last week reported Google Cloud growth of 82%, while Microsoft’s Azure and other cloud services revenue rose 43% during the fiscal fourth quarter. The hyperscaler capex race is accelerating across all three. The difference at Amazon is that AWS is doing this with the largest absolute revenue base, making a 37% growth rate proportionally more impressive than comparable percentage moves from smaller starting points. A 37% gain on a roughly $169 billion annualized business adds more incremental revenue than Azure’s 43% gain does on its smaller base.
Operating income rose 43% to $27.5 billion, compared with $19.2 billion in the same period last year. The operating leverage is intact even as the building cycle peaks. That is the signal that matters to a multi-quarter investor.
Risks
The bear case deserves a clear reading. Amazon plans to spend about $220 billion in 2026, which may pressure free cash flow in the near term. The company raised its full-year capex estimate mid-cycle largely because memory costs rose, not because demand pulled forward. That is manageable, but it introduces a variable cost dynamic that can keep capex elevated longer than the original schedule implied.
The tolerance for cash burn has limits. If debt continues to rise to fund the buildout, at some point free cash flow recovery becomes a condition of the investment thesis, not just a consequence of timing. For the third quarter, Amazon guided for net sales between $197 billion and $202 billion and operating income between $22.5 billion and $26.5 billion. Management said the quarter-to-quarter slowdown from Q2 to Q3 reflects the shift in Prime Day timing and a foreign exchange headwind. Q3 will look softer optically, which could give the market a reason to revisit capex concerns before Q4 data arrives.
What to Watch
Three signals will determine whether this thesis converts over the next two to three quarters. First, AWS growth durability: a fifth and sixth consecutive quarter of acceleration would confirm demand is structural rather than cyclical front-loading. Second, the capex trajectory: the company has said that if AWS growth keeps accelerating, they will continue to invest more, meaning capex estimates for 2027 are likely to go up given how much growth the company is generating, despite still being in a supply-constrained environment. Watch whether Jassy pre-guides to a 2027 number at Q3 earnings. Third, AWS margins at the 40% threshold: CFO Brian Olsavsky noted that AWS margins were up 650 basis points year over year, or 520 basis points excluding a derivative accounting gain. Sustaining that improvement into a heavier depreciation cycle would end the margin debate entirely.
Final Verdict
The conventional framing of Amazon’s capex story is wrong. This is not a case of a company spending recklessly and hoping returns follow. The $496 billion backlog, the roughly 39% AWS operating margin, the supply-constrained demand environment, and the Trainium compute commitments from OpenAI and Anthropic are not speculative. They are the receipts.
The free cash flow reversal is real, and Q3 guidance will look soft compared to Q2. None of that is hidden. All of it is already known to the market, which is why Amazon reported surging cloud growth during the second quarter, pointing to strong artificial intelligence demand, and the stock jumped in after-hours trading. The market has decided the returns are credible. The data from this quarter supports that read. The question for a new investor today is not whether the AI demand is real. The question is whether 37% AWS growth at about a 39% operating margin, with a $496 billion contractual backlog and Jassy explicitly saying demand still exceeds capacity, is adequately reflected in the price. That is a harder argument to make against than the free cash flow headline suggests.
