AWS Reaccelerated to 37%. Amazon’s Free Cash Flow Turned Negative.

AI demand and a broader silicon strategy are restoring the cloud leader’s growth. Now Amazon must prove that the largest infrastructure build in its history can earn an acceptable return.

Amazon just reported two different investment stories in the same quarter.

The first was AWS, growing like an AI challenger while earning like an incumbent. Revenue increased 36.7% year over year to $42.2 billion, the fastest growth in 18 quarters. Operating income rose 63.6% to $16.6 billion, and operating margin expanded to 39.4%.

The second was the consolidated cash-flow statement. Amazon generated $161.4 billion of trailing operating cash flow and spent $169.0 billion on property and equipment net of proceeds and incentives. Free cash flow moved to an outflow of $7.6 billion from an inflow of $18.2 billion one year earlier.

AWS has rarely looked stronger. Amazon’s cash conversion has rarely carried a larger infrastructure burden.

That contradiction is now the investment case.

AWS Is Growing Faster Without Giving Up Margin

AWS represented only 21.1% of Amazon’s second-quarter revenue but produced 60.5% of operating income.

That is the economic center of Amazon.

Quarterly AWS revenue increased from $29.3 billion in the first quarter of 2025 to $42.2 billion in the second quarter of 2026. Year-over-year growth accelerated from 17% to 36.7%. Operating margin reached 39.4%, almost matching the segment’s prior peak despite a much larger revenue base and a much heavier infrastructure program.

The growth is broader than frontier-model training. Enterprises adopting generative AI also need storage, databases, networking, security, identity, observability, and conventional compute. Those workloads often sit beside existing applications and regulated data, giving the surrounding cloud platform an important role even when customers use models from several providers.

The strongest evidence is the combination of growth and margin. AWS operating income increased materially faster than revenue. If Amazon were buying the acceleration through heavy discounting or carrying large amounts of idle installed capacity, that result would be harder to produce.

The timing caveat matters. Revenue can accelerate before the full depreciation expense of assets under construction reaches the income statement. Today’s AWS margin proves that the installed platform is economically powerful. It does not prove that the 2026 infrastructure cohort will earn the same return.

That is why the next result cannot be judged on growth alone. The relevant combination is AWS revenue growth, segment margin, depreciation, capital spending, and free-cash-flow conversion.

AWS Is the Profit Engine, Not the Whole Valuation

Amazon generated $200.6 billion of second-quarter revenue and $27.5 billion of operating income. AWS contributed $42.2 billion of that revenue and $16.6 billion of the operating profit.

The segment is a minority of sales and the majority of profit.

The company therefore contains two very different economic systems.

AWS is high-margin, recurring, and capital-intensive. The non-AWS businesses include retail, advertising, subscriptions, logistics, devices, and media. Their margins are lower, but they control consumer distribution, Prime relationships, advertising inventory, and fulfillment density.

The stock works best when both systems improve. Faster AWS growth drives operating profit. Better retail efficiency and advertising lift the margin on a much larger revenue base. If only AWS performs, the current valuation leaves less room for error.

Competition also argues against treating AWS as an isolated monopoly. Azure and other cloud services grew 43% in Microsoft’s June quarter. Google Cloud grew 82% to $24.8 billion with an operating margin of approximately 35.6%. The disclosures are not perfectly comparable, but the strategic direction is clear: all three hyperscalers are spending heavily to capture the same expansion in AI workloads.

AWS still has important advantages. It combines scale, enterprise relationships, broad service coverage, security, databases, model choice, and operational history. It does not need to win every quarterly growth comparison. It needs to remain one of the broadest and most profitable platforms as AI workloads move from experiments into persistent production.

Silicon Choice Must Improve the Whole Stack

Amazon’s chip strategy is not a semiconductor side business.

Graviton CPUs and Trainium accelerators influence customer pricing, supply availability, infrastructure cost, and AWS margin. NVIDIA GPUs provide the compatibility and scale many workloads still require. AWS’s advantage is offering credible choice while integrating each architecture with networking, storage, databases, security, and Bedrock.

Management says its chips business has exceeded a $25 billion annual revenue run-rate and is growing at triple-digit percentages. It also says Graviton is used by 98% of AWS’s top 1,000 EC2 customers.

Those are strong adoption signals. They are not complete economics. Customer use does not disclose workload share, and an annualized run-rate does not disclose product-level gross margin or the split between internal and external demand.

Trainium is the harder economic test. Amazon says Trainium3 offers 30%-40% better price-performance than Trainium2. Anthropic now says it uses more than one million Trainium2 chips, has committed more than $100 billion to AWS technologies over ten years, and has secured up to 5 gigawatts of capacity. OpenAI has agreed to consume 2 gigawatts of Trainium capacity. Pinterest committed $4 billion through 2031 across AWS infrastructure that includes Trainium and Graviton.

AWS also plans to deploy two million additional NVIDIA GPUs during 2027 and 2028. That strengthens the evidence for demand and compatibility. It also extends the capital cycle and preserves dependence on a third-party chip supplier.

These commitments improve utilization visibility. They also create concentration. A few frontier-model customers can consume enormous capacity, negotiate aggressively, and change architectures quickly.

The best outcome is not simply selling more chips. It is using silicon choice to make the entire AWS platform more available, stickier, and more profitable.

Bedrock sits one layer above that infrastructure. It offers model access, security, data governance, evaluation, customization, and agent tooling through one managed AWS layer. If it becomes the operating system for enterprise AI agents, AWS can capture more than basic compute economics. If it remains a routing layer, model providers and application developers may retain most of the differentiated margin.

AWS’s pending acquisition of DuckLabs, the company behind DuckDB, is another move toward the data-and-developer layer. The project is expected to remain open source under an independent foundation. No terms were disclosed, so the deal strengthens the platform narrative without changing near-term valuation.

Commitments provide visibility. Recognized revenue proves delivery. Operating income proves economics. Cash conversion proves shareholder value.

The Infrastructure Bill Arrives Before the Return

Amazon’s capital cycle is the central risk.

Trailing operating cash flow reached $161.4 billion. Net property-and-equipment purchases reached $169.0 billion. Free cash flow was negative $7.6 billion.

The outflow was not caused by a collapse in Amazon’s operating business. It was caused by an investment program larger than current operating cash generation.

That distinction explains the cash flow. It does not excuse it.

Amazon’s 2025 shareholder letter said the company expected roughly $200 billion of 2026 capital expenditure, with a substantial portion supported by customer commitments. Management subsequently raised the cash-capex outlook to approximately $220 billion after Q2 as memory and capacity costs increased. Cash capex is not identical to Amazon’s reported net property-and-equipment purchases, but both measures show investment running ahead of current cash generation.

The balance sheet is absorbing the build. Net property and equipment reached $446.0 billion. Long-term debt rose to $128.9 billion from $65.6 billion at year-end 2025. Cash and marketable securities totaled $123.0 billion.

Amazon has ample liquidity and extraordinary operating cash generation. It no longer has the same balance-sheet cushion.

The accounting expense also arrives later than the cash cost. Trailing depreciation and amortization was approximately $75.2 billion, less than half of trailing net property-and-equipment purchases. The figures are not directly comparable because asset lives, land, construction in progress, incentives, and non-AWS assets differ. They still show that investment is arriving much faster than the expense recognized in earnings.

The bull case is historical. Amazon has repeatedly accepted weak near-term cash flow to build fulfillment, Prime, and AWS capacity that later produced a larger earnings base. This cycle also has visible enterprise demand and long-term commitments.

The difference is scale and duration. The two-million-GPU plan extends through 2028, Anthropic has secured up to 5 gigawatts, and Amazon increased its planned Louisiana data-center investment to $18 billion. The Louisiana campuses are to be developed and owned by STACK Infrastructure, so that headline should not be added mechanically to 2026 capex. It still shows how far the physical build now reaches.

The investment case is no longer whether Amazon is willing to spend. It is whether the new assets can earn returns comparable to the installed AWS base.

Valuation Already Requires Successful Monetization

At the August 28 snapshot of $256.26, Amazon’s equity value was approximately $2.79 trillion. The quote reflects the latest completed U.S. session on August 27.

The reported price-to-earnings ratio is misleading. Second-quarter net income included $53.4 billion of non-operating pre-tax income primarily related to Anthropic. Capitalizing that gain as recurring earnings understates the normalized multiple. The infrastructure build creates another distortion because cash spending runs ahead of depreciation.

Our framework values AWS and the non-AWS businesses separately using 2028 revenue, operating margin, and operating-income multiples, then discounts the result at 9% over 2.5 years. Strategic investment value is excluded.

• Bear case: $181. AWS growth normalizes, margin falls as depreciation catches up, and non-AWS margins improve only modestly.
• Base case: $305. AWS compounds at 25% with a 37% margin while non-AWS revenue and margins continue improving.
• Bull case: $454. AWS sustains exceptional growth and margin while retail, advertising, and logistics produce stronger operating leverage.

The probability-weighted value is approximately $297 per share, about 16% above the market snapshot.

We did not raise the scenario values for the new AWS announcements. The NVIDIA and Anthropic commitments improve demand visibility, but they also extend capital intensity and supplier exposure. Without disclosed incremental revenue, utilization economics, or returns, changing the valuation would overstate what the evidence can support.

The range is more important than the midpoint. AWS can remain an excellent business while AMZN produces a poor shareholder return if capital intensity persists, depreciation compresses margins, or the valuation multiple falls.

At the refreshed snapshot, investors are being paid for execution, not obvious undervaluation.

What Would Change the View

The strongest positive signal would be AWS growth remaining above 30% while capex growth slows and free-cash-flow conversion improves.

The strongest negative signal would be AWS growth falling below 25% while depreciation, debt, and the infrastructure budget keep rising.

For existing investors, the operating thesis remains intact while AWS maintains a mid-to-high 30s margin and the non-AWS businesses continue expanding profit. For new investors, a staged position or a better entry offers more protection while free cash flow remains negative.

The next review should focus on four questions:

• Can AWS hold growth above 30% without sacrificing margin?
• Has Amazon stopped raising the capital-spending requirement?
• Are Trainium commitments and NVIDIA deployments becoming recognized revenue without weakening cash conversion?
• Are Bedrock and the data layer creating cross-service platform capture rather than only infrastructure usage?

The Bottom Line

AWS is not struggling. It is reaccelerating at scale while producing one of the strongest segment margins in public technology.

Amazon is spending because demand is visible, customers want capacity, and a broader silicon portfolio can improve the economics and reach of the cloud stack. That is a stronger foundation than an AI infrastructure program built around speculative utilization.

It is not proof that the return will exceed the cost.

Verdict: NEUTRAL / constructive watchlist. AWS is one of the strongest operating franchises in public markets. At the refreshed market snapshot, Amazon still requires successful AI monetization, sustained AWS margins, improving non-AWS economics, and an eventual recovery in free cash flow.

The cloud leader is growing again. The next phase is whether shareholders receive an acceptable return on the build.

The complete StockResearch report includes the AWS segment model, silicon-strategy analysis, Bedrock and partnership review, hyperscaler comparison, capital-cycle bridge, 2028 valuation scenarios, risk matrix, and explicit thesis-validation signals.

Disclaimer

This article is for informational and educational purposes only and does not constitute financial or investment advice. Equities are volatile and carry risk of loss. The market snapshot is from August 28, 2026, 07:15 Asia/Bangkok and reflects the latest completed U.S. session on August 27; operating and financial data are through Amazon’s quarter ended June 30, 2026 unless otherwise stated. Scenario values are illustrative analytical judgments and may differ materially from actual outcomes. Always conduct your own due diligence before making investment decisions.

Read the full report on the StockResearch platform