The Kapital · Global Deep Dive · United States · 20 August 2026
Amazon just delivered the kind of cloud quarter that makes enormous AI spending look rational. AWS revenue grew 37% to $42.2 billion, contract backlog jumped to $496 billion and management raised expected 2026 capital spending to roughly $220 billion. The problem is that free cash flow has moved in the opposite direction. For me, this is now the central Amazon question: is the company sacrificing cash temporarily to build a scarce, pre-sold AI infrastructure asset—or has the cloud business entered a permanently more capital-intensive era?
I have always liked Amazon most when the market focuses on the wrong expense line. Historically, the company created value by spending ahead of visible demand: fulfillment centers before Prime scaled, AWS infrastructure before cloud computing became obvious, and logistics capacity before same-day delivery became normal. That history creates a temptation to treat every new investment wave as automatically intelligent.
I do not think investors should grant that assumption for free. The current AI buildout is different in scale. Data centers, accelerators, networking and power infrastructure absorb extraordinary capital, and the useful life of some hardware may be shorter than the buildings that house it. The correct question is not whether AWS demand is strong. It clearly is. The correct question is whether the return on each incremental dollar of infrastructure remains attractive after depreciation, power and rapid hardware obsolescence.
The quarter was genuinely strong
Amazon reported second-quarter net sales of $200.6 billion, up 20% year over year. Operating income rose to $27.5 billion from $19.2 billion. AWS was the standout: revenue climbed 37% to $42.2 billion, its fastest growth rate in more than four years according to Reuters.
That acceleration matters because hyperscaler capex is easiest to defend when the associated revenue engine is accelerating too. A company spending 50% more while cloud growth falls from 30% to 15% deserves scrutiny. Amazon is currently showing the opposite pattern: capacity investment is rising while AWS growth reaccelerates.
The backlog is the strongest evidence that this is not speculative capacity
The most important number in the quarter may be AWS backlog. Reuters reported that committed contract value reached $496 billion, up from $364 billion one quarter earlier. That is a huge increase in contracted demand and helps explain why management is willing to raise annual investment plans by about 10%.
Backlog is not the same as immediate revenue and it can contain long-dated commitments, but it tells me customers are reserving capacity rather than merely expressing interest. Management also said much of AWS capacity for 2027 is already reserved and some 2028 capacity has been committed.
That is the best argument for the bull case. Amazon is not building empty data centers and hoping someone eventually arrives. It appears to be building into a shortage. In a shortage, returns on capital can be excellent because the scarce input—compute—has pricing power.
But the free-cash-flow reversal is real
The uncomfortable side of the story is free cash flow. Reuters reported trailing-twelve-month free cash flow of negative $7.6 billion compared with positive $18.2 billion a year earlier. That reversal is the cost of the investment wave.
For an investor who grew accustomed to Amazon finally becoming a large free-cash-flow compounder, the number looks ugly. I think the right response is neither panic nor dismissal. Free cash flow is telling us that the economic burden of AI infrastructure is arriving before all of the associated revenue.
The critical distinction is timing. If free cash flow rebounds as installed capacity starts generating revenue, then the current period is a classic investment trough. If free cash flow stays structurally weak because every dollar of AWS growth requires another large wave of accelerators and power equipment, then Amazon deserves a lower multiple than an asset-light software company.
Why AWS economics can still be exceptional
Cloud infrastructure is capital intensive, but AWS has several advantages that can preserve attractive returns. First, scale lowers procurement and operating costs. Second, customers often use higher-level services—databases, analytics, security and AI tools—on top of raw compute. Third, switching costs can be meaningful once applications and data are deeply integrated.
AI may deepen those economics. Amazon said its AI and custom-chip businesses each exceed $25 billion annualized run rates. Chips such as Trainium allow AWS to compete on cost and reduce dependence on a single accelerator supplier. If customers consume both Amazon-designed chips and higher-margin managed services, the return on the full stack can be better than the return on hardware alone.
This is why I am more comfortable with Amazon’s capex than I would be with a smaller cloud company making the same relative commitment. Amazon owns the customer relationship, infrastructure, software services, advertising platform and commerce ecosystem. It has multiple ways to monetize the same AI capability.
The retail business is helping fund the wager
North America sales rose 16% to about $116.2 billion and International sales grew 15% to about $42.2 billion. Advertising sales rose 26% to $19.8 billion. These businesses matter because Amazon is not an AWS pure play. Advertising in particular is a high-margin asset that can support consolidated operating income while cloud capex is elevated.
The diversification cuts both ways. Retail can dilute margins, but it also provides data, demand and distribution for AI. Better recommendation systems, advertising tools, fulfillment planning and seller services allow Amazon to monetize AI internally rather than relying solely on external AWS customers.
Advertising is included as a service line and overlaps with segment revenue presentation; the chart is for scale comparison, not additive consolidation.
What would make $220 billion of capex dangerous
I see four failure modes. The first is overbuilding: capacity comes online after demand growth has already slowed. The second is price compression: competitors build enough capacity that compute becomes a commodity. The third is obsolescence: accelerators age faster than expected, forcing Amazon to refresh hardware before the original investment earns an adequate return. The fourth is power economics: data-center electricity and grid constraints push operating costs higher.
Any one of these can reduce return on invested capital even while AWS revenue keeps growing. That is why I will not judge the capex plan by revenue growth alone. I want evidence that operating income and eventually free cash flow scale with the installed asset base.
My valuation framework
Amazon recently traded around the mid-$260s. I value the company as a combination of AWS, advertising, retail and logistics rather than applying one multiple to consolidated earnings. Because the AI investment cycle creates unusually wide outcomes, I prefer scenarios.
Bear case: $200 to $220
In my bear case AWS growth normalizes quickly into the low 20s, capex remains above $200 billion for longer than expected and free cash flow recovers slowly. Investors lower the multiple because Amazon looks more infrastructure-heavy. I use roughly $210 as the center.
Base case: $265 to $285
My base case assumes AWS remains above 25% growth through the near term, backlog converts cleanly, capex growth slows after the current capacity wave and consolidated free cash flow rebounds. Advertising continues to compound at a healthy rate. Around $275 is my central value.
Bull case: $325 to $345
The bull case requires AWS to maintain exceptional AI demand, custom silicon to improve infrastructure economics and operating cash flow to grow faster than capex by 2027. In this world Amazon becomes the clearest hyperscaler beneficiary of the AI buildout while retail margins continue improving.
Illustrative scenarios, not price targets. Recent price reference is from mid-August 2026 trading.
What I will watch next
The first metric is AWS backlog growth. If backlog flattens while capex stays near current levels, my confidence falls quickly. The second is AWS operating margin: strong revenue growth is less valuable if depreciation and infrastructure costs consume the incremental profit. Third is trailing free cash flow. I do not need it to recover immediately, but I need a credible path back toward strongly positive cash generation.
Fourth is custom silicon adoption. Trainium and other Amazon chips can materially improve bargaining power and cost. Fifth is capacity commentary. Today, scarcity is Amazon’s friend. The moment management shifts from “we cannot build fast enough” to “we have ample capacity,” the valuation discussion changes.
Why depreciation will matter almost as much as capex
One accounting line I expect to become increasingly important is depreciation. Cash leaves when infrastructure is built, while the income statement recognizes much of that cost over the useful life of the assets. During a rapid buildout, operating income can therefore look stronger than free cash flow because depreciation lags the cash investment. Later, as the installed base grows, depreciation catches up.
This does not make Amazon’s earnings misleading. It simply means I want to compare AWS operating income with the capital base required to produce it. A cloud business that earns $1 of incremental operating profit for relatively little additional capital deserves a software-like multiple. A cloud business that requires several dollars of constantly refreshed accelerators for the same profit deserves a more infrastructure-like multiple. The next few years should reveal where AI-era AWS sits on that spectrum.
My conclusion
Amazon’s $220 billion capex plan would terrify me if AWS were slowing. It is not. AWS grew 37%, backlog reached $496 billion and large portions of future capacity are already reserved. Those facts make the spending economically plausible.
But investors should not confuse plausible with proven. The free-cash-flow reversal shows that the AI buildout has a real cost, and the long-term return depends on how much revenue and operating profit each new data-center dollar produces.
At roughly $266, the stock sits close to my $275 base-case value. I like the business more than I like the margin of safety. A pullback would interest me, but the more important trigger would be evidence that operating cash flow begins to outrun capex again. That is when today’s enormous infrastructure bill becomes tomorrow’s moat rather than tomorrow’s depreciation problem.
Sources and data status
Data status: 20 August 2026. Capital expenditure guidance and backlog can change as Amazon updates investment plans and customer commitments.
- Reuters: AWS growth and $220bn capex plan
- Amazon Investor Relations: quarterly results
- Reuters: hyperscaler AI investment and returns
This article reflects my own analysis and is not investment advice. Valuation scenarios are estimates and can change materially with growth, margins, rates and capital intensity.
