Amazon AWS Growth Accelerates to 37% as AI Demand Tests a $220 Billion Buildout

Amazon’s second-quarter 2026 results delivered the clearest evidence yet that its enormous artificial-intelligence infrastructure program is producing real revenue inside Amazon Web Services. AWS sales rose 37% year over year to $42.2 billion in the quarter ended June 30, the cloud division’s fastest growth in 18 quarters. Its operating income increased 64% to $16.6 billion, and its operating margin reached 39.4%.

Those figures explain why Amazon shares rallied after the release even though the company also raised its expected 2026 capital spending to approximately $220 billion and reported trailing-12-month free cash flow of negative $7.6 billion. Wall Street did not suddenly stop caring about cash. Rather, the quarter supplied something investors had been demanding from every large technology company: visible evidence that higher spending is being accompanied by faster commercial growth, stronger contracted demand and substantial operating profit.

The result does not settle the debate over whether the artificial-intelligence infrastructure boom will ultimately earn acceptable returns. Amazon is spending cash years before some data centers produce revenue. It is relying heavily on a relatively small group of frontier-model companies and other large customers. Its headline net income was inflated by a $53.4 billion non-operating pre-tax gain, primarily related to its investment in Anthropic. Long-term debt has risen sharply. Memory, power, construction and advanced-chip constraints remain material.

Yet the immediate answer to the dominant question around Amazon’s earnings is straightforward: AWS did not merely keep pace with the AI infrastructure cycle in the second quarter. It accelerated meaningfully, generated the majority of Amazon’s operating profit and strengthened the argument that Amazon’s capital program has a commercial foundation rather than being based only on distant promises.

Last updated: July 31, 2026, 3:30 a.m. EDT. Market information reflects the latest available regular-session and post-market data at the research cutoff.

Key Takeaways

• Main development: AWS revenue increased 37% year over year to $42.2 billion, accelerating from 28% growth in the first quarter and 24% in the fourth quarter of 2025.
• Profitability: AWS operating income rose to $16.6 billion and represented approximately 60.5% of Amazon’s consolidated operating income, despite AWS contributing about 21.1% of total revenue.
• Capital spending: Amazon raised its expected 2026 cash capital expenditures from approximately $200 billion to approximately $220 billion, citing infrastructure demand and higher memory costs.
• Cash flow: Trailing-12-month operating cash flow increased 33% to $161.4 billion, but free cash flow fell to an outflow of $7.6 billion because net purchases of property and equipment reached $169.0 billion.
• Earnings quality: Reported net income of $62.6 billion included $53.4 billion of non-operating pre-tax other income, primarily related to Anthropic. Operating income provides a cleaner view of Amazon’s underlying business performance.
• Market response: Amazon shares gained 3.9% in regular trading on July 30 and rose nearly 9% after the earnings release, according to Reuters. Post-market moves can change before the next regular session.
• What comes next: Investors will watch whether AWS can sustain growth above 30%, whether its backlog converts into revenue at attractive margins, and whether Amazon’s elevated debt and capital spending begin producing durable free cash flow.

What Amazon Reported and Why the Quarter Mattered

Amazon reported second-quarter net sales of $200.6 billion, compared with $167.7 billion a year earlier. The 20% reported increase was broad enough to matter: North American segment sales rose 16% to $116.2 billion, International segment sales increased 15% to $42.2 billion and AWS sales grew 37% to $42.2 billion. Advertising services revenue advanced 26% to $19.8 billion, while third-party seller services grew 16% to $46.8 billion.

The quarter’s importance, however, came from the interaction between AWS growth, margins and capital intensity. Amazon had already told investors in February that it expected to spend about $200 billion on capital expenditures during 2026. That announcement contributed to concern that Amazon was becoming far more capital intensive just as AI competition was forcing cloud providers to purchase expensive accelerators, memory, networking equipment, land and power capacity. By July, the question was no longer whether spending would be high. The question was whether operating performance could justify it.

AWS provided the strongest answer Amazon could reasonably have offered in one quarter. Revenue growth accelerated by nine percentage points from the first quarter. Quarterly AWS sales increased 12.4% sequentially, from $37.6 billion to $42.2 billion. Operating income increased faster than revenue, and the segment’s operating margin rose from 32.9% in the year-earlier period to 39.4%.

This combination is more informative than a single revenue beat. Rapid growth can be purchased through discounting or inefficient capacity deployment. High margins can be produced temporarily by underinvesting. In Amazon’s second quarter, AWS reported both faster growth and higher segment profitability while the company continued to invest aggressively. That does not prove every new data center or AI server will earn attractive lifetime returns, but it does show that the existing AWS platform is monetizing demand at scale.

The market reaction also reflected the change in expectations. Before the release, investors had watched Google Cloud and Microsoft Azure report sharp growth, raising the performance threshold for AWS. Reuters cited LSEG data showing that analysts expected AWS growth of approximately 31.2%. The reported 37% rate exceeded that benchmark by nearly six percentage points and reduced concern that Amazon was permanently losing momentum to Microsoft and Google.

Amazon AWS Earnings: The Numbers Behind the Acceleration

AWS’s growth trajectory changed considerably over the preceding five quarters. The division grew 17% year over year in the second quarter of 2025, 20% in the third quarter, 24% in the fourth quarter, 28% in the first quarter of 2026 and 37% in the second quarter. The corresponding quarterly revenue figures were $30.9 billion, $33.0 billion, $35.6 billion, $37.6 billion and $42.2 billion.

Source: Amazon quarterly earnings releases. Figures are reported in U.S. dollars and may not sum because of rounding.

This sequence matters because cloud businesses benefit from operating leverage when capacity is well utilized. A data center requires substantial upfront construction and equipment spending. Once that capacity is available, incremental usage can contribute attractive profit if pricing, utilization, energy costs and depreciation remain favorable. AWS’s margin expansion suggests that the company was not simply adding revenue by carrying uneconomic workloads.

There are still reasons to avoid extrapolating 39.4% indefinitely. Quarterly cloud margins can be affected by server useful-life assumptions, energy prices, workload mix, customer credits, depreciation timing and the pace at which newly built capacity enters service. A period in which demand runs ahead of available supply can also create unusually favorable utilization. As Amazon adds hundreds of billions of dollars of assets, future depreciation will rise and could place pressure on margins even if revenue remains strong.

Nevertheless, the second-quarter numbers demonstrate scale that is difficult to dismiss. An annualized run rate based on one quarter is not a forecast, but multiplying $42.2 billion by four produces approximately $169 billion, the figure Amazon highlighted. AWS is already larger on that simple run-rate basis than many major global technology companies, and its quarterly operating income alone exceeded the total quarterly operating profit of numerous large public companies.

Why Gil Luria Called the AWS Result Remarkable

Gil Luria, D.A. Davidson’s head of technology research, described the AWS result as “nothing short of remarkable” in a Bloomberg Television discussion after the earnings release. His assessment centered on three connected observations: the acceleration occurred at enormous scale, the business appeared to be monetizing AI demand rather than merely absorbing costs, and Amazon was selling more than raw computing capacity.

The scale argument is important. Percentage growth rates are easier to produce from a smaller base. Google Cloud grew faster in the same reporting season, and Microsoft reported strong Azure growth, but AWS added approximately $11.4 billion of quarterly revenue from the year-earlier period. Annualizing that increase produces roughly $45 billion of additional revenue, although actual future quarters will vary. The absolute expansion was large enough to change consolidated Amazon results.

Luria also focused on Amazon’s claim that its AI and chip businesses had each exceeded $25 billion annual revenue run rates. Those figures are company-reported run rates rather than audited standalone segment results. Amazon does not provide separate income statements for the AI and chip businesses, and investors cannot independently determine how much overlap exists between those categories and broader AWS revenue. Still, the disclosures indicate that the company’s AI commercialization is no longer limited to experimental projects.

The third point is the software layer. Enterprise customers generally need more than access to a frontier model or a bank of accelerators. They need identity management, data governance, security, model selection, orchestration, monitoring, application integration, cost controls and deployment tools. AWS can capture value at several points in that stack through services such as Amazon Bedrock, Bedrock AgentCore, databases, storage, networking and security products.

That distinction helps explain why the cloud providers may earn durable revenue even if the identity of the leading model changes. An enterprise can switch from one foundation model to another while continuing to use the same data lake, security architecture, developer tools and cloud infrastructure. Amazon’s strategic position is therefore not limited to predicting which model laboratory wins. It is attempting to become the operating environment through which customers use many models.

Luria’s positive reading should not be treated as proof that Amazon’s capital returns are guaranteed. Analysts can change views, and the economics of long-lived infrastructure depend on assumptions about future demand, pricing and asset obsolescence. The value of his argument lies in identifying what the quarter actually improved: Amazon produced more evidence that demand is converting into high-margin revenue now, rather than remaining solely in backlog or management forecasts.

The $169 Billion AWS Run Rate Is a Scale Indicator, Not Guidance

Amazon emphasized that AWS had reached a $169 billion annualized revenue run rate. The calculation is simple: four times quarterly AWS revenue of $42.2 billion. It helps readers understand the size of the business, but it should not be confused with a formal forecast for the next 12 months.

Cloud usage can be seasonal. Large customer contracts can ramp in stages. Capacity constraints can limit recognized revenue. Currency movements can change reported growth. Customers can optimize workloads or migrate among providers. A run rate therefore describes the latest quarter’s pace rather than guaranteeing that AWS will report exactly $169 billion over the next four quarters.

Even with that limitation, the figure is useful. AWS generated trailing-12-month revenue of $148.4 billion through June 2026, up 28% from the comparable trailing period. The difference between the $148.4 billion trailing figure and the $169 billion annualized pace illustrates how rapidly the latest quarter accelerated.

The run rate also provides context for Amazon’s capital-spending decision. A $220 billion companywide cash-capital plan sounds extreme in isolation. Compared with an AWS business approaching a $170 billion annual pace, plus a retail, logistics, advertising, subscription, entertainment, satellite and robotics portfolio, the spending is still aggressive but less detached from existing revenue. The relevant question becomes how much of that spending supports future revenue and how much represents replacement, maintenance or speculative capacity.

Amazon does not disclose enough detail to calculate project-level returns. Investors do not know the exact division of the $220 billion among AWS data centers, custom chips, retail logistics, robotics, satellite infrastructure and other projects. They also do not know the weighted average contract duration, customer concentration or pricing for reserved 2027 and 2028 capacity. Those omissions matter because a dollar committed to a long-term, take-or-pay cloud contract has a different risk profile from a dollar committed to uncontracted capacity.

The appropriate conclusion is therefore measured. The run rate shows that AWS has reached a size where even moderate percentage growth creates enormous absolute revenue. It does not prove that every dollar of incremental capital spending will earn the same margin as the current business.

AWS Generated Most of Amazon’s Operating Profit

AWS accounted for approximately 21.1% of Amazon’s second-quarter revenue but about 60.5% of its operating income. That profit concentration is central to understanding Amazon as a public company. The group’s retail and marketplace businesses create scale, customer relationships, advertising inventory and logistics advantages, but AWS remains the largest source of segment operating profit.

North America generated $9.1 billion of operating income on $116.2 billion of sales, an operating margin of approximately 7.9%. The International segment produced $1.7 billion on $42.2 billion of sales, a margin of approximately 4.1%. AWS generated $16.6 billion on $42.2 billion, a margin of approximately 39.4%.

The result does not mean the retail businesses are unimportant. North American operating income increased 21%, and International operating income rose 15%. Faster delivery, third-party seller services, advertising and subscription revenue can improve retail economics. Amazon’s physical network also supports businesses that competitors cannot easily copy, including fulfillment services and rapid grocery or pharmacy delivery.

Still, AWS’s profit contribution gives the cloud segment disproportionate influence over Amazon’s valuation. A one-percentage-point change in AWS margin on $42.2 billion of quarterly revenue is worth roughly $422 million of quarterly operating income, before considering taxes or other items. A sustained shift in cloud growth or profitability can therefore outweigh meaningful changes elsewhere in the company.

This leverage works in both directions. Strong demand, disciplined pricing and high utilization can expand consolidated margins quickly. A cloud price war, underused data centers, faster depreciation or customer concentration problems could reduce earnings just as quickly. Investors who view Amazon primarily as an online retailer risk missing the financial importance of AWS; investors who view it only as a cloud company risk overlooking the capital needs and competitive pressures across the rest of the organization.

The AI Revenue Claim: What Is Confirmed and What Is Not

Amazon said its AWS AI business exceeded a $25 billion annual revenue run rate and was growing at triple-digit percentages year over year. It also said its chips business surpassed a $25 billion annual run rate and was growing at triple-digit rates. These are significant disclosures, but they require careful interpretation.

First, a run rate is not the same as recognized annual revenue. It annualizes a recent pace. Second, Amazon has not published standalone financial statements for either category. The AI figure may include infrastructure, managed services, model access, databases, storage and software used in AI workloads. The chips figure may reflect revenue commitments or consumption associated with Graviton, Trainium and related instances. The categories may overlap because custom chips power AI services sold through AWS.

Third, the company has not disclosed segment-level costs or margins for these businesses. Luria’s discussion estimated attractive economics, but outside readers cannot verify a specific margin from the public release. The broader AWS margin is known; the margin of the narrower AI or chip categories is not.

What is confirmed is that Amazon chose to disclose these figures at a time when investors were questioning the revenue generated by AI investment. The company also reported that customers spent more on Bedrock in the second quarter than in all prior quarters combined and that hundreds of thousands of customers use the service. Those claims are first-party metrics rather than independently audited customer-usage data, but they indicate a sharp increase in commercial activity.

The most useful interpretation is that AWS’s AI business has become large enough to influence segment growth. It is no longer credible to describe Amazon’s AI effort as a purely defensive cost center. It is equally premature to assume the current growth rate or margins will persist once capacity expands and competition intensifies.

Anthropic Is Both a Strategic Asset and a Concentration Risk

Anthropic sits near the center of Amazon’s AI strategy. The relationship includes equity investment, cloud consumption, custom-chip collaboration and distribution of Claude models through Amazon Bedrock. Anthropic has identified AWS as its primary cloud and training partner, while Amazon has used the partnership to create large-scale demand for Trainium infrastructure.

In April 2026, Anthropic announced an expanded agreement securing up to five gigawatts of new Amazon compute capacity for training and deploying Claude. The company said significant Trainium2 capacity would come online during 2026 and that the relationship would extend across future Trainium generations. Amazon separately announced an additional $5 billion investment, with the possibility of investing more under the expanded arrangement.

The commercial logic is clear. Amazon provides capital and infrastructure. Anthropic commits to use AWS capacity and collaborates on chip optimization. AWS gains a large anchor customer, while Bedrock gains access to a prominent model family. Anthropic receives funding and the computing resources required to train and serve increasingly expensive systems.

The arrangement also creates analytical complications. Amazon’s investment gains can increase reported net income without producing operating cash. Anthropic’s purchases contribute to AWS revenue, while Amazon’s capital supports Anthropic’s growth. This does not make the revenue unreal or the relationship improper. Large strategic partnerships routinely combine investment and commercial agreements. It does mean investors should separate operating performance, investment revaluation and cash movement.

Customer concentration is another concern. Reuters reported that Anthropic’s ramp was a major contributor to AWS acceleration and that the cloud provider had entered large infrastructure agreements with other frontier-model companies and technology firms. If a small number of customers account for a meaningful share of new AI demand, changes in their funding, model strategy or hardware preferences could affect utilization.

Anthropic also uses multiple hardware platforms. A primary partnership does not necessarily imply exclusivity. Frontier-model developers seek redundancy, negotiating leverage and access to the best available accelerators. Amazon must therefore keep improving Trainium performance, software compatibility and economics rather than relying only on contractual alignment.

The strongest reading is that Anthropic gives AWS a powerful source of contracted demand and product feedback. The strongest skeptical reading is that Amazon is taking investment, counterparty and infrastructure risk around a private company whose long-term economics remain uncertain. Both interpretations can be true at the same time.

Why the Anthropic Gain Distorted Amazon’s Headline Net Income

Amazon reported net income of $62.6 billion, or $5.75 per diluted share, compared with $18.2 billion, or $1.68 per diluted share, in the second quarter of 2025. The increase appears extraordinary until the non-operating items are examined.

The quarter included $53.4 billion of non-operating pre-tax other income, primarily from Amazon’s Anthropic investments. The consolidated statement of operations shows total non-operating income of $53.4 billion, compared with $1.7 billion a year earlier. Income before taxes was $80.9 billion, while operating income was $27.5 billion.

The investment gain matters economically because an increase in the value of Amazon’s ownership can strengthen its balance sheet and create future strategic or financial value. It should not be treated as equivalent to revenue earned from selling cloud services, merchandise or advertising. The gain is non-operating, and much of it is non-cash at the point of recognition.

Amazon’s cash-flow statement illustrates the distinction. The company began with net income and then subtracted $53.4 billion of non-operating income in reconciling earnings to operating cash flow. Deferred taxes and other working-capital adjustments also affected the calculation. Operating cash flow for the quarter was $45.4 billion, not $62.6 billion.

This is why operating income is the more useful starting point for evaluating the quarter. Consolidated operating income increased 43% to $27.5 billion, and operating margin rose to approximately 13.7% from 11.4% a year earlier. Those changes reflect the performance of Amazon’s operating businesses rather than a private-company valuation adjustment.

Investors should not ignore the Anthropic stake, but they should value it separately. Private-company valuations can change sharply. Observable financing rounds may trigger accounting adjustments. A higher carrying value does not mean Amazon could sell the entire position immediately at the recorded amount, and a later decline could create losses.

Amazon’s $220 Billion Capital Plan

Amazon increased its expected 2026 cash capital expenditures to approximately $220 billion from the approximately $200 billion level disclosed earlier in the year. Chief Executive Andy Jassy attributed part of the increase to higher memory costs and said the company still expected demand to exceed available capacity.

The amount places Amazon at the center of the largest infrastructure expansion in the technology sector’s history. The spending supports data centers, servers, chips, networking, power connections and related construction, but it also includes investments elsewhere in Amazon, including fulfillment, robotics and other long-term projects. The company does not provide a complete project-by-project allocation.

During the second quarter, Amazon purchased $54.2 billion of property and equipment. After $1.1 billion of proceeds from property sales and incentives, net purchases were approximately $53.1 billion. Quarterly operating cash flow was $45.4 billion, producing a simplified quarterly free-cash-flow outflow of approximately $7.7 billion under Amazon’s stated definition.

For the trailing 12 months, operating cash flow increased to $161.4 billion. Net purchases of property and equipment reached $169.0 billion, resulting in negative free cash flow of $7.6 billion. A year earlier, Amazon generated $18.2 billion of trailing free cash flow.

That deterioration is not hidden. It is the direct financial consequence of the buildout. Amazon is effectively exchanging present cash for long-lived assets and future capacity. The investment can create substantial value if customers use the capacity at favorable prices for many years. It can destroy value if demand slows, equipment becomes obsolete, electricity is constrained or pricing falls faster than costs.

Jassy argued that the timing creates an optical mismatch. Data-center spending begins well before a facility opens, while revenue arrives later. He also said AI servers can repay their cost within a few years and continue generating profit afterward. Those are management claims about economics and asset life, not guarantees. They depend on utilization, pricing, maintenance, power and the pace of technological change.

Free Cash Flow Is Negative, but the Underlying Picture Is More Nuanced

Negative free cash flow normally deserves caution. It means the company’s operating cash generation was insufficient to cover the property and equipment purchases included in its definition. For a mature company with a market value measured in trillions of dollars, that is a meaningful change.

Amazon’s situation is different from a business burning cash because its core operations are unprofitable. Operating cash flow increased 33% over the trailing year to $161.4 billion. The negative free-cash-flow figure arose because net capital purchases increased 64% to $169.0 billion. In other words, the operating engine strengthened while investment grew even faster.

That distinction does not make the cash outflow harmless. Capital expenditures are real cash uses. New assets will create depreciation expense. Data centers require ongoing energy, maintenance and network spending. Technology assets may become obsolete before buildings do. A company can have strong operating cash flow and still overinvest.

The more relevant analytical questions are whether capital is committed against credible demand, whether expected contract revenue compensates for financing and operating costs, and whether Amazon has enough balance-sheet capacity to complete projects without weakening financial flexibility.

The evidence is encouraging but incomplete. AWS backlog rose sharply, and Amazon said much of 2027 capacity was already reserved. Revenue growth and margins accelerated. Those facts support the investment case. The missing details include customer concentration, contract cancellation terms, pricing, expected return on invested capital and the proportion of capex devoted to maintenance rather than expansion.

Investors should therefore avoid two extremes. It is inaccurate to describe the entire $220 billion program as reckless cash burn when AWS is growing 37% and producing a 39.4% margin. It is equally inaccurate to treat contracted demand as proof that the projects cannot disappoint. Long-duration infrastructure returns remain exposed to technology, financing and customer risk.

Debt Is Now Part of the AI Investment Story

Amazon’s long-term debt increased from $65.6 billion at the end of 2025 to $128.9 billion at June 30, 2026, an increase of approximately $63.2 billion. The cash-flow statement shows $67.0 billion of proceeds from long-term debt during the first six months of 2026, compared with less than $1 billion in the year-earlier period.

The company remained highly liquid. It reported $78.2 billion of cash and cash equivalents and $44.8 billion of marketable securities at June 30, totaling approximately $123.0 billion before considering restricted cash or other assets. Operating cash flow was substantial. Amazon is not facing a near-term liquidity crisis.

Still, debt changes the capital-allocation equation. Interest expense was $1.3 billion in the second quarter, up from $516 million a year earlier. Higher leverage can be rational when contracted infrastructure demand offers attractive returns, but it reduces flexibility if economic conditions, customer demand or financing costs deteriorate.

The balance-sheet expansion is also visible in property and equipment, which increased from $357.0 billion at year-end to $446.0 billion at June 30. Total assets increased to nearly $1.1 trillion, partly reflecting property investment and the higher value of financial investments. Amazon is becoming a much larger owner and financier of physical and digital infrastructure.

This shift resembles utilities, telecommunications companies and industrial networks more than the capital-light internet model many investors once associated with large technology companies. The distinction matters for valuation. Capital-intensive companies require sustained returns above their cost of capital, careful asset-life assumptions and disciplined financing. Revenue growth alone is not sufficient.

Amazon’s advantage is that AWS demand appears strong and its consolidated operations generate significant cash. Its risk is that the scale of investment leaves little room for forecasting error. A modest shortfall applied to a $220 billion annual program can produce a large absolute amount of underutilized capital.

Backlog Provides Visibility, but It Is Not the Same as Revenue

Reuters reported that AWS contract backlog reached $496 billion at the end of the quarter, up from $364 billion three months earlier. Jassy said the majority of 2027 compute capacity had already been reserved and that customers had also reserved meaningful capacity for 2028.

Backlog is valuable because it shows customers have entered agreements for future services. It can reduce the risk that Amazon builds entirely speculative capacity. It also supports management’s argument that current spending is constrained by demand rather than driven only by competitive fear.

Backlog should not be treated as immediately recognized revenue. Cloud contracts can extend over several years. Revenue is recorded as services are delivered. Commitments may include minimum usage, variable consumption, renewals, capacity reservations or other terms. Public disclosures do not reveal the exact timing, margins or cancellation provisions for every contract.

The increase from $364 billion to $496 billion is unusually large and supports the view that AI laboratories and enterprises are securing capacity well in advance. It also increases execution pressure. Amazon must deliver data centers, chips and power on schedule. Delays can postpone revenue and damage customer relationships. Cost inflation can reduce project margins even when demand remains intact.

Backlog quality matters as much as size. Contracts with well-capitalized, diversified enterprises may carry different risk from commitments made by private AI companies dependent on future fundraising. A concentrated backlog can be vulnerable to a change in one customer’s strategy. Amazon has not disclosed enough detail to quantify that risk.

The responsible conclusion is that backlog strengthens the case for investment while leaving important questions unanswered. It is evidence of demand, not a substitute for cash flow.

Bedrock and the Software Layer May Matter as Much as Compute

Much of the public AI debate focuses on chips and frontier models, but the enterprise profit opportunity may be spread across a broader software and infrastructure stack. Amazon Bedrock gives customers access to multiple foundation models while integrating them with AWS security, data, monitoring and application services.

Amazon said hundreds of thousands of customers use Bedrock and that second-quarter customer spending exceeded all previous quarters combined. It also said the service added models from several providers and expanded capabilities for deploying AI agents at scale.

The strategic value is choice. A customer can use Anthropic, OpenAI, Amazon or other models without rebuilding its entire cloud architecture. That reduces the importance of Amazon owning the single best model. AWS can earn revenue from compute, storage, databases, networking, governance and orchestration while model developers compete above it.

The model-agnostic position is not complete insulation. If one model provider develops a deeply integrated cloud platform, customers may prefer that provider’s native environment. Microsoft benefits from its relationship with OpenAI, while Google controls both leading models and custom chips. Amazon must ensure that Bedrock offers competitive performance, pricing and access.

Enterprise AI adoption also depends on measurable return on investment. Early experiments can generate cloud consumption without becoming durable production workloads. Chief information officers will eventually demand lower error rates, stronger security and clear productivity gains. Amazon’s growth can remain high only if deployments move from prototypes to recurring business processes.

Luria’s point that companies need the control plane and orchestration layer captures this opportunity. Model access alone is increasingly commoditized. The cloud provider that helps customers connect models to proprietary data, permissions and workflows can capture a larger share of spending and make switching more difficult.

Trainium and Graviton Are Amazon’s Attempt to Control Its Economics

Amazon’s custom-chip strategy is designed to reduce dependence on third-party processors, improve price-performance and differentiate AWS. Graviton serves general-purpose computing workloads, while Trainium targets AI training and inference. Inferentia also supports inference workloads, although Amazon’s recent disclosures emphasize Trainium.

The company said its chip business exceeded a $25 billion annual revenue run rate. It reported that Graviton is used by 98% of its top 1,000 EC2 customers and that the latest generation provides improved performance. Amazon has also described large, multi-year Trainium commitments from Anthropic, OpenAI and other customers.

Custom silicon can improve margins in several ways. Amazon can avoid part of the supplier markup embedded in merchant chips, optimize processors for its own data centers, design networking and software around the hardware, and offer customers a lower-cost alternative. It can also use chips as a strategic tool when scarce accelerators limit cloud growth.

The challenge is software. Nvidia’s strength is not only hardware performance; it includes a mature development ecosystem. Customers must be able to move workloads, train engineers and maintain model quality without excessive friction. A cheaper chip that requires substantial rewriting can be more expensive in total.

Anthropic’s collaboration provides a demanding test environment. Frontier-model workloads can expose hardware and software weaknesses quickly. If Trainium performs well for Claude training and inference, Amazon gains both a large customer and a reference case. If performance or reliability falls behind, Anthropic and other customers can diversify toward Nvidia GPUs, Google TPUs or competing accelerators.

Custom silicon also introduces manufacturing and supply-chain risk. Amazon designs chips but relies on external foundries and memory suppliers. Advanced packaging, high-bandwidth memory and power delivery can remain bottlenecks even when processor design is successful.

The second-quarter results indicate the chip strategy is contributing commercially. They do not yet reveal whether Trainium will produce durable economics across a broad customer base or remain concentrated among a few strategic partners.

How AWS Compares with Microsoft Azure and Google Cloud

The cloud competition intensified before Amazon reported. Microsoft said Azure revenue increased 43% in its fiscal fourth quarter ended June 30, while Alphabet reported an 82% increase in Google Cloud revenue to $24.8 billion. AWS’s 37% growth was slower in percentage terms than both figures, but the comparisons are imperfect.

AWS reports segment revenue directly. Alphabet’s Google Cloud segment includes Google Cloud Platform and Google Workspace. Microsoft reports Azure growth but does not provide a directly comparable quarterly Azure revenue figure, and its broader Intelligent Cloud and Microsoft Cloud categories include other businesses. Currency treatment also differs.

Scale changes the interpretation. AWS produced $42.2 billion of quarterly segment revenue. Google Cloud produced $24.8 billion. Microsoft’s Azure business is large, but the exact quarterly revenue cannot be derived cleanly from the reported growth percentage alone. A slower rate on a larger base can represent more absolute revenue.

Amazon’s acceleration also matters relative to its own recent history. In early 2026, D.A. Davidson had expressed concern that AWS was losing its lead as rivals grew faster. The move from 24% growth in the fourth quarter to 28% in the first and 37% in the second weakens that specific momentum argument, even though market-share competition remains unsettled.

Google’s growth demonstrates that the market is expanding rather than simply shifting among providers. Enterprises are increasing spending on AI infrastructure, data platforms and cloud services. The leading providers can all grow rapidly if total demand expands fast enough.

The competitive risk is not disappearing. Microsoft combines Azure with enterprise software and OpenAI distribution. Google combines models, TPUs, search data and cloud services. Oracle and specialized “neocloud” providers compete for high-performance workloads. Customers increasingly use multiple clouds to manage risk and access specialized products.

AWS’s second quarter shows it remains a formidable incumbent. It does not prove that market share will remain stable or that pricing will stay favorable as new capacity arrives.

Sources: company earnings releases for quarters ended June 30, 2026. Reported growth rates are not fully comparable because segment definitions differ.

Cloud Market Share Is Useful but Often Misunderstood

Third-party market-share estimates generally place AWS first in global cloud infrastructure services, followed by Microsoft and Google. The exact percentages vary by methodology, reporting period and definition. Some estimates include infrastructure and platform services while excluding software applications; others classify certain hosted products differently.

Market share can also decline while revenue rises rapidly if the overall market expands faster than one provider. That distinction is important during the AI buildout. A provider can lose a percentage point of share and still add tens of billions of dollars of annual revenue.

The more relevant questions for Amazon are whether AWS can retain enough scale to support attractive unit economics, whether it remains a default choice for enterprise workloads and whether its products capture spending beyond commodity compute. The second-quarter performance improved the answers to all three, but did not end the competition.

Market share measured by revenue also reflects pricing. A provider charging more for a specialized service may gain revenue share without serving more workloads. A provider using lower-cost custom chips may deliver more compute while reporting less revenue. Raw share figures therefore do not reveal customer satisfaction, workload volume or profitability.

Amazon’s greatest defense is the breadth of its platform and installed base. Large customers have built complex systems around AWS. Migration can be expensive and risky. Its greatest vulnerability is that AI creates a new workload category in which customers may make fresh infrastructure decisions rather than simply extending older arrangements.

The acceleration to 37% suggests AWS is winning a meaningful share of that new spending. Sustaining the result will require product execution, not only incumbent advantage.

Retail Growth Made the Quarter Broader Than an AWS Story

AWS dominated the market discussion, but Amazon’s retail and marketplace operations also strengthened. North America sales increased 16% to $116.2 billion, compared with 12% growth in the first quarter. International sales increased 15% to $42.2 billion, compared with 11% growth in the first quarter. Both segments remained profitable.

Several factors supported the result. Amazon held Prime Day during the second quarter, shifting sales that would otherwise have appeared later in the year. The company continued to shorten delivery times, expand same-day and overnight service, and increase the proportion of everyday essentials in its sales mix. Third-party seller services revenue rose 16%, indicating higher activity across the marketplace and the fees Amazon earns for fulfillment, commissions and related services.

Faster delivery can improve customer frequency, but it also carries cost. Local inventory placement, more delivery stations and smaller fulfillment nodes can raise fixed expenses if order density is insufficient. Amazon’s ability to increase North American operating income alongside faster sales suggests the network handled the higher volume efficiently during the quarter.

International profitability is another important development. The segment generated $1.7 billion of operating income, compared with $1.5 billion a year earlier. International retail has historically been less profitable because Amazon operates in markets with different labor costs, logistics systems, consumer habits and competitive structures. A 4.1% operating margin remains far below AWS, but positive earnings reduce the degree to which overseas expansion consumes cloud profits.

The retail business also contributes to the AI story. Amazon can apply machine learning to inventory positioning, demand forecasting, warehouse robotics, shopping recommendations, advertising and customer service. Some of those benefits may appear as lower costs or higher conversion rather than separately reported AI revenue. The company’s AI economics therefore extend beyond AWS, even though AWS provides the most visible revenue line.

The principal caution is that the second quarter benefited from event timing. Amazon’s third-quarter guidance explicitly noted that Prime Day occurred in different quarters in 2025 and 2026. Comparisons must therefore adjust for the shift rather than treating the second-quarter retail acceleration as a new steady-state rate.

Advertising Is Becoming a Second High-Margin Growth Engine

Advertising services revenue increased 26% to $19.8 billion. The business is now approaching the scale of major global media companies and is strategically connected to Amazon’s commerce data, marketplace and streaming properties.

Amazon can sell ads near the point of purchase, where customer intent is often clearer than on general social or entertainment platforms. Sponsored product placements, display ads, video inventory and third-party advertising technology allow the company to monetize traffic without owning all the merchandise sold.

Advertising can also improve retail profitability because the incremental cost of showing an ad is generally lower than the cost of storing and delivering a physical product. Amazon does not disclose a separate advertising operating margin, but digital advertising businesses are commonly more profitable than first-party retail.

The relationship between advertising and marketplace competition deserves scrutiny. Sellers may feel pressure to purchase ads to maintain visibility, effectively adding a marketing cost on top of commissions and fulfillment fees. Regulators and merchants may question whether Amazon gives its own products or preferred placements an advantage. These issues can affect the durability of ad growth.

AI can make the advertising business more efficient by automating campaign creation, targeting and measurement. Amazon said advertisers using its Ads Agent achieved lower cost per impression and lower customer-acquisition costs, based on company data. Those metrics are promising but require independent validation and longer-term observation.

From an investor’s perspective, advertising diversifies Amazon’s profit sources. AWS remains dominant, but a fast-growing advertising business can offset pressure in retail and help finance infrastructure investment. It also makes Amazon’s ecosystem more integrated: sellers pay for marketplace access, fulfillment, payment processing and visibility, while consumers receive faster delivery and broader selection.

Third-Quarter Guidance Was Stronger Than the Headline Comparison Appears

Amazon projected third-quarter net sales of $197 billion to $202 billion, representing reported growth of 9% to 12% from the third quarter of 2025. It forecast operating income of $22.5 billion to $26.5 billion, compared with $17.4 billion a year earlier.

The revenue growth range appears much slower than the second quarter’s 20%, but calendar timing explains part of the difference. Prime Day occurred in the third quarter of 2025 and the second quarter of 2026. Amazon said that excluding the effect of Prime Day in both periods, third-quarter year-over-year growth would be nearly four percentage points higher.

The company also expected foreign exchange to reduce growth by approximately 0.8 percentage point. These adjustments do not eliminate all deceleration, but they show why a simple comparison between 20% and the 9% to 12% guidance range would be misleading.

The operating-income range remains wide. At the midpoint of $24.5 billion, operating income would increase approximately 41% year over year. The low end would still represent meaningful growth, while the high end would continue the pattern of margin expansion. The range reflects uncertainty around spending, seasonal demand, energy costs and other variables.

Guidance is a management forecast, not an achieved result. Amazon’s actual performance can differ because of currency, consumer demand, cloud utilization, project timing, litigation, restructuring or macroeconomic changes. The company also stated that the guidance assumed no additional material acquisitions, restructurings or legal settlements.

For AWS, the most important question is whether the division can sustain growth above 30% as it laps stronger comparisons. The quarter demonstrated acceleration; the next several quarters will reveal whether that rate reflects a lasting step-up in demand or the timing of large capacity additions and customer ramps.

Why Investors Accepted Higher Spending This Time

Technology stocks have reacted very differently to large AI spending plans. Investors have rewarded companies that show clear revenue growth and punished those whose monetization plans appear distant or uncertain. Amazon’s second-quarter reaction fits that pattern.

The company reported four forms of evidence supporting its capital program. First, AWS revenue growth exceeded expectations. Second, AWS operating margin expanded. Third, backlog increased sharply. Fourth, management said much of future capacity was already reserved. Together, those points made the higher capital forecast easier to accept.

The contrast with other companies is instructive. Microsoft paired strong cloud demand with substantial cash generation and a mature enterprise software business. Google reported rapid cloud growth and benefits across search and advertising. Meta’s infrastructure spending supports internal products but lacks an external cloud business of comparable scale, making direct monetization less visible.

Amazon sits between a traditional cloud vendor and a diversified industrial platform. It can sell compute externally through AWS, use AI internally to improve retail and advertising, and design custom chips to influence input costs. That combination creates more potential revenue channels than a company building infrastructure solely for internal use.

The market’s acceptance is conditional. If AWS growth slows while capex remains near $220 billion, free cash flow and leverage will receive more attention. If margins compress because new assets begin depreciating before utilization rises, the current optimism could reverse. Investors have not given Amazon a permanent exemption from capital discipline; they responded to one quarter in which the evidence improved.

The Stock Reaction and What It Actually Signaled

Amazon shares rose 3.9% during regular trading on July 30 and nearly 9% after the release, Reuters reported. The regular-session price was $235.50, according to market data available after the close. Post-market prices can be volatile and may not match the next day’s opening or closing price.

The reaction suggests that AWS growth and margins outweighed concern about higher capex, negative free cash flow and cautious revenue guidance. It does not prove that investors collectively reached one explanation. Stock prices reflect short covering, options positioning, index flows, revised earnings models and changes in perceived risk.

A large after-hours move can also reflect the difference between actual results and expectations rather than the absolute quality of a quarter. Amazon’s spending was already anticipated to be very high. The surprise was that cloud growth accelerated more than expected and that the company presented stronger evidence of contracted demand.

The reported net-income beat was less informative because of the Anthropic gain. Professional investors generally separate such non-operating items, focusing on revenue, operating income, cash flow and guidance. The share response therefore should not be read as approval of the $62.6 billion profit figure itself.

Valuation remains sensitive to assumptions about AWS growth, margins and long-term capital intensity. A business that requires $220 billion of annual capex deserves a different cash-flow analysis from one producing similar earnings with minimal reinvestment. The value created depends on what the assets earn over their useful lives.

How to Think About the Return on Amazon’s AI Capital

Calculating the return on Amazon’s AI investment from public data is not possible with precision. The company does not disclose the full allocation of capital expenditures, project-level revenue, contract prices, maintenance spending or depreciation schedules for each asset class. Still, investors can organize the analysis around several measurable concepts.

Utilization

Data centers earn attractive returns only when customers use the capacity. A nearly full facility can spread fixed costs across more revenue; an underused facility leaves power, land and equipment idle. Amazon’s backlog and capacity reservations support future utilization, but actual usage and contract enforcement remain critical.

Pricing

Strong demand can support pricing, while industry overcapacity can trigger discounts. Cloud providers compete on headline prices, committed-use agreements and specialized services. Custom chips may allow Amazon to lower prices while protecting margins, but price competition could still reduce returns.

Asset life

Buildings and power infrastructure can last decades. AI servers and accelerators have much shorter economic lives. A chip that remains technically functional may become uneconomic when a newer generation offers better performance per watt. Return calculations must therefore separate long-lived facilities from rapidly obsolescing equipment.

Energy and memory costs

Jassy identified higher memory prices as one reason for the increased spending forecast. AI systems require large quantities of high-bandwidth memory, networking and electricity. Cost inflation can lower project returns unless Amazon passes it to customers or offsets it through efficiency.

Software attachment

Compute may attract customers, but databases, security, orchestration and data services can improve margins and retention. The higher the software and managed-service revenue attached to each unit of infrastructure, the stronger the potential return.

Financing

Debt-funded projects must earn more than their financing cost after considering risk. Amazon’s borrowing increased substantially in the first half. The company’s scale and cash generation provide capacity, but higher interest expense raises the hurdle rate.

The second-quarter AWS margin offers encouraging evidence, but it reflects the existing asset base and current utilization. The $220 billion program will be evaluated over many years, not one earnings cycle.

The Accounting Timing Problem: Cash Leaves Before Revenue Arrives

Amazon’s defense of the buildout rests partly on timing. Data-center projects require land, power agreements, construction, servers and networking before customers can consume services. Cash expenditures can precede revenue by many quarters.

When the facility becomes operational, the company capitalizes eligible costs and recognizes depreciation over estimated useful lives. The cash has already left, but the accounting expense appears gradually. Free cash flow can therefore look weakest during the build phase, while operating income may remain strong until more new assets enter service and depreciation rises.

This timing makes neither measure sufficient on its own. Free cash flow captures immediate capital intensity but can understate the future earning capacity of newly built assets. Operating income captures current revenue and depreciation but may not yet reflect the full expense associated with projects still under construction.

A disciplined analysis tracks both. Investors should watch whether operating cash flow continues growing, whether capital spending eventually moderates relative to revenue, and whether depreciation rises faster or slower than AWS gross profit.

Amazon has previously benefited from extending estimated server useful lives, which lowers annual depreciation expense. Such changes can be economically justified when equipment remains productive longer, but they affect margins and comparisons. Future useful-life assumptions will matter more as the asset base expands.

The timing problem also explains why backlog is so important. If Amazon spends two years before a facility opens, customer commitments reduce the risk of building without demand. They do not remove construction, cost or counterparty risk.

Capacity Constraints Are Positive Until They Are Not

Management said Amazon still lacked enough capacity to satisfy all 2026 demand and expected constraints to persist into 2027. In the near term, constrained supply supports utilization, contract visibility and pricing. It can also cause revenue to be deferred because AWS cannot serve every workload immediately.

Persistent shortages may push customers toward competitors. Enterprises with urgent needs cannot always wait for Amazon to complete a facility. Microsoft, Google, Oracle and specialized providers can use Amazon’s constraints as a sales opportunity.

Capacity shortages also create incentives to overbuild. When every large provider sees demand exceeding supply, each may assume current conditions will persist. Projects started simultaneously can arrive after demand growth slows, turning scarcity into excess capacity.

The AI industry is particularly difficult to forecast because efficiency improves rapidly. New models may require more compute because usage expands, or less compute per task because algorithms and hardware improve. Both can occur at once. The resulting demand depends on how lower costs affect adoption.

Amazon’s customer reservations into 2028 reduce some uncertainty. The company still must match the location, chip type, network and power configuration customers require. A general shortage does not guarantee that every individual asset will be well utilized.

Memory, Power and Construction Are Strategic Constraints

The AI buildout is not limited by capital. High-bandwidth memory, advanced packaging, power-generation capacity, transformers, grid connections, cooling equipment and skilled construction labor can all delay projects. Amazon’s decision to raise capex partly because of memory pricing shows that supply constraints affect both timing and cost.

Power may be the most persistent bottleneck. Large AI data centers require electricity at a scale that can exceed local grid capacity. Projects can face multi-year interconnection queues, regulatory approval and community opposition. Companies increasingly sign long-term energy contracts or explore dedicated generation.

Higher electricity use also creates environmental and political risk. Data centers compete with households and industry for power and water. Regulators may impose efficiency, disclosure or siting requirements. Local communities may question tax incentives or infrastructure costs.

Amazon has advantages in procurement scale, engineering and global site selection. It can negotiate long-term supply agreements and distribute workloads across regions. Scale does not eliminate exposure; it makes the company one of the largest buyers in constrained markets.

Cost inflation can be passed through only if customers accept higher prices. If competing providers absorb costs to gain share, margins may fall. Custom chips can reduce processor costs, but memory and power remain external inputs.

The Strongest Case for Amazon’s Strategy

The bullish interpretation begins with demand. AWS grew 37% on a $42.2 billion quarterly base, backlog reached $496 billion, and future capacity was heavily reserved. These are not merely product announcements; they are indicators of commercial scale.

Second, profitability remained exceptional. A 39.4% AWS operating margin and 64% operating-income growth suggest Amazon can monetize the demand rather than competing solely through discounts. AWS generated more than 60% of consolidated operating income.

Third, Amazon controls several layers of the stack. It designs chips, operates data centers, sells infrastructure, distributes models through Bedrock and provides databases, security and developer tools. That vertical integration can improve economics and customer retention.

Fourth, Amazon has multiple ways to use AI. AWS sells it externally; retail applies it to forecasting and logistics; advertising uses it for targeting and campaign creation; Alexa and shopping assistants can increase engagement; robotics can improve fulfillment productivity.

Fifth, the company has the balance sheet and operating cash flow to invest through the cycle. Negative free cash flow does not arise from a collapsing core business. Operating cash flow reached $161.4 billion over 12 months.

Under this interpretation, the current cash outflow represents a temporary investment phase. As data centers open and contracted capacity begins producing revenue, AWS could sustain high growth while capital spending eventually grows more slowly. Free cash flow would recover, and custom chips could protect margins.

The Strongest Skeptical Case

The skeptical case begins with capital intensity. A $220 billion annual plan is so large that even a modest forecasting error can destroy tens of billions of dollars of value. The company’s trailing free cash flow is already negative, and long-term debt has roughly doubled in six months.

Second, current demand may be concentrated among frontier-model developers whose economics are not proven. Some AI laboratories depend on repeated fundraising and strategic investment. Contracts can reduce risk but do not make every counterparty equivalent to a mature enterprise customer.

Third, part of the commercial ecosystem is circular. Amazon invests in Anthropic, Anthropic commits to purchase AWS capacity, and Amazon records investment gains as Anthropic’s valuation rises. The cloud revenue can be genuine while the overall system remains dependent on external capital and rapidly rising private valuations.

Fourth, hardware obsolescence may shorten asset lives. AI accelerators improve quickly, and software optimization can reduce compute required per task. Capacity that appears scarce today may become less valuable if new chips deliver much better efficiency or customers change architectures.

Fifth, competition is intense. Microsoft, Google and specialized providers are also expanding. If all major companies build against optimistic forecasts, pricing could weaken after capacity enters service.

Sixth, the headline profit obscures the operating picture. The $53.4 billion Anthropic-related gain made net income look far stronger than cash generation. Investors who focus on earnings per share without separating non-operating items may misjudge the quality of the quarter.

Under the skeptical interpretation, the second quarter represents peak scarcity economics. Growth and margins are strong because demand exceeds supply, but returns decline as new capacity, depreciation and price competition arrive.

What Evidence Would Resolve the Debate

Several future disclosures would help distinguish between the supporting and skeptical interpretations.

• Sustained AWS growth: Growth above 30% for several quarters would indicate the acceleration is not limited to one customer ramp or capacity release.
• Stable margins: AWS operating margin near the upper 30% range after new depreciation enters the income statement would support strong project economics.
• Free-cash-flow recovery: A narrowing outflow while revenue and backlog grow would show that investment is moving toward monetization.
• Capex moderation: Capital spending growing more slowly than operating cash flow would improve financial flexibility.
• Customer diversification: Broader enterprise adoption would reduce reliance on a few AI laboratories.
• Trainium adoption: Use by customers without strategic investment ties would provide stronger evidence of independent chip competitiveness.
• Backlog conversion: Consistent revenue recognition and limited cancellations would validate contract quality.
• Debt stabilization: Slower borrowing would reduce concern that infrastructure growth requires continuously expanding leverage.

No single quarter can answer all of these. Amazon’s second quarter improved the evidence on growth and margins while worsening the reported picture on free cash flow and debt.

Material Risks Investors Should Not Ignore

Demand concentration

Large AI customers can accelerate AWS revenue but create exposure to a small number of counterparties. A strategy change, funding problem or shift to another hardware platform could affect utilization.

Technology obsolescence

Accelerators and networking equipment may lose economic value faster than expected. Useful-life estimates affect both project returns and reported depreciation.

Price competition

Microsoft, Google, Oracle and specialized providers can discount capacity or bundle services. Lower prices may expand usage while compressing margins.

Execution

Data centers require power, permits, construction and complex supply chains. Delays can move revenue into later periods and increase costs.

Financing

Higher debt and interest expense reduce flexibility. A rise in borrowing costs or decline in operating cash flow would make the buildout more difficult.

Regulation

Cloud concentration, data sovereignty, AI safety, competition, privacy and energy use can attract regulatory action. Rules may differ across jurisdictions.

Cybersecurity and reliability

AWS is critical infrastructure for many customers. Outages, security incidents or misconfigurations can create financial and reputational damage.

Anthropic valuation

Amazon’s reported investment value can rise or fall with private financing events and accounting estimates. Non-operating gains should not be assumed to recur.

Retail cyclicality

Consumer spending, tariffs, labor costs and delivery expenses affect the non-cloud businesses that still produce most of Amazon’s revenue.

Environmental constraints

Power and water requirements can delay projects, raise costs and create community opposition.

What the Quarter Says About the Wider AI Investment Cycle

Amazon’s results strengthen the argument that the AI infrastructure cycle is producing substantial revenue for cloud providers. Microsoft and Google also reported rapid cloud growth. Enterprises and model developers are purchasing large quantities of compute, and contracted demand extends beyond the current year.

The quarter does not prove that the entire industry’s spending will be profitable. Cloud providers can earn attractive returns while some customers lose money. Infrastructure demand can remain strong during a period of speculative financing. Revenue earned today does not guarantee that all capacity delivered in 2028 will be fully utilized.

The distinction between provider economics and application economics is important. AWS can be paid for computing services regardless of whether every AI startup eventually becomes profitable, as long as customers can meet their obligations. However, if application revenue fails to support continued spending, demand can slow when contracts expire.

Amazon’s advantage is that AI demand is attached to a broad existing cloud platform. Customers use AWS for non-AI workloads, data storage, databases and security. This creates cross-selling opportunities and makes the infrastructure less dependent on one application category.

The wider cycle will be judged by productivity. Businesses must eventually show that AI increases revenue, reduces costs or improves services enough to justify cloud bills. Infrastructure providers are currently the clearest financial beneficiaries. The durability of that benefit depends on their customers finding economic value.

Amazon’s Transformation into an Infrastructure Company

Amazon began as an online retailer, but its financial profile increasingly resembles a global infrastructure operator. It owns fulfillment centers, aircraft capacity, delivery stations, data centers, custom-chip designs, media rights and a satellite network. The company’s competitive advantage comes partly from assets that require enormous upfront investment.

This transformation changes how results should be evaluated. Growth must be measured against the capital required to produce it. Operating margins must be considered alongside depreciation and maintenance. Cash flow must distinguish between expansion and replacement spending.

Amazon’s history supports long investment horizons. The company built fulfillment and cloud networks before their economics were obvious. AWS itself emerged from internal infrastructure and became the group’s principal profit engine. That record gives management credibility, but it does not make new projects automatically successful.

The scale is now different. A failed project measured in hundreds of millions is manageable. A broad misallocation within a $220 billion annual program can affect the entire company. Governance and capital discipline become more important as investment rises.

The second-quarter result demonstrates why Amazon is willing to accept that risk. AWS growth accelerated, and margins expanded. The company sees an opportunity to build infrastructure that customers may use for decades. Whether the economic life of AI equipment supports that vision remains one of the central questions in global markets.

What Happens Next

The next stage of the story will unfold through operating data rather than product announcements. Investors should watch AWS revenue growth, segment margin, backlog, operating cash flow, capital expenditures, debt and depreciation.

Amazon’s third-quarter results will also require careful comparison because of Prime Day timing. Retail growth may slow mechanically even if underlying demand remains stable. AWS will be less affected by that calendar shift and will provide a cleaner view of AI momentum.

Capacity additions scheduled for 2027 and 2028 will test management’s claim that projects are substantially reserved. If revenue follows capital spending with a predictable lag, free cash flow should eventually improve. If capital needs continue rising faster than operating cash flow, the debate will intensify.

Trainium adoption beyond Anthropic and other strategic partners will be particularly informative. Independent customer demand would validate Amazon’s claim that its custom chips offer competitive price-performance and reduce dependence on merchant accelerators.

Investors should also monitor financing. The first-half increase in long-term debt was substantial. Amazon can support more borrowing than most companies, but continued leverage growth would increase interest expense and lower tolerance for execution problems.

A Timeline of the AWS Reacceleration

The second-quarter result is easier to understand as the latest stage in a multi-quarter transition rather than an isolated surprise. AWS entered 2025 growing in the mid-teens after customers spent an extended period optimizing cloud bills. That optimization cycle included deleting unused resources, renegotiating commitments and shifting workloads to lower-cost configurations. It reduced near-term revenue growth but also encouraged customers to build more efficient architectures.

By the second quarter of 2025, AWS growth was 17%. Growth increased to 20% in the third quarter and 24% in the fourth. In February 2026, Amazon announced an approximately $200 billion capital-spending plan for the year, arguing that demand across AI, chips, robotics and satellites justified the investment. The market’s initial response was cautious because the spending increase arrived before investors had seen enough acceleration in free cash flow or cloud revenue.

First-quarter 2026 results improved the picture. AWS growth reached 28%, its fastest pace in years at that point, and the segment’s operating margin rose to 37.7%. Amazon also reported large commitments for Trainium capacity and increasing Bedrock adoption. Even so, Google Cloud was growing more quickly, and analysts continued to debate whether AWS had fallen behind in the AI era.

During April, Amazon and Anthropic expanded their compute relationship, providing a clearer commercial path for a portion of AWS’s future capacity. Microsoft and Google subsequently reported strong June-quarter cloud results, increasing expectations for Amazon. When AWS delivered 37% growth, the result was interpreted not only as a beat but as evidence that Amazon’s earlier capacity investments were moving into production.

The timeline shows why the market reacted positively to higher spending in July after reacting more cautiously in February. The spending commitment did not become smaller. The evidence connecting it to revenue became stronger.

What Amazon’s Earlier Cloud Cycles Teach

AWS has experienced several periods in which growth slowed as customers optimized spending, followed by periods of reacceleration when new workloads and capacity came online. The pattern is inherent in usage-based cloud computing. Customers can reduce consumption quickly, but they can also expand rapidly when new applications reach production.

The 2022–2023 optimization cycle demonstrated that cloud revenue is not immune to economic pressure. Companies facing slower growth examined technology budgets and eliminated waste. AWS remained profitable, but growth decelerated. Amazon responded by emphasizing cost controls, reducing headcount in some areas and extending asset utilization.

The current AI cycle differs because customers are reserving very large blocks of capacity before facilities open. Traditional enterprise migrations often occurred application by application. Frontier-model training and inference can require clusters measured in thousands or millions of accelerators and power commitments measured in gigawatts. The size and duration of contracts are therefore potentially greater, but so are the risks.

Another difference is hardware specialization. Earlier cloud growth relied heavily on general-purpose processors and storage. AI growth depends on accelerators, high-bandwidth memory and specialized networking. These assets may have different useful lives and resale values. A building can support several generations of equipment; a particular accelerator can become less competitive quickly.

Amazon’s history supports the idea that temporary cloud slowdowns do not necessarily indicate structural decline. It also warns against assuming every new cycle resembles the last. The economics of AI infrastructure must be evaluated on their own terms.

Scenario Analysis for the $220 Billion Program

Because project-level data is unavailable, a scenario framework is more responsible than a single forecast.

High-demand scenario

In the favorable scenario, enterprise AI applications move into production, frontier-model companies continue expanding and AWS converts its backlog into high-utilization revenue. Custom chips lower cost per unit of compute, Bedrock attaches profitable software services and capital spending begins growing more slowly after the current construction wave. AWS maintains growth above 30% and margins in the high 30% range. Operating cash flow rises enough to restore positive free cash flow even before capex returns to historical levels.

Balanced-growth scenario

In a middle scenario, demand remains healthy but growth gradually slows as comparisons become harder. AWS revenue expands in the 20% range, new depreciation offsets some operating leverage and free cash flow improves only gradually. Amazon earns acceptable returns, but the investment does not produce the extraordinary economics implied by the strongest bullish models. Debt stabilizes rather than falling quickly.

Overcapacity scenario

In the adverse scenario, customers improve model efficiency, funding for AI laboratories tightens and competing capacity enters the market simultaneously. Cloud pricing weakens, utilization falls below plan and Amazon carries depreciation on assets that are not fully used. AWS remains a large profitable business, but margins decline and free cash flow stays negative for longer. Amazon may delay projects, renegotiate supplier commitments or reduce spending elsewhere.

The second-quarter results increase the probability of the first two scenarios because they show current demand and strong profitability. They do not eliminate the third because much of the $220 billion program will be monetized in future periods.

Governance, Competition and Regulatory Questions

Amazon’s scale creates regulatory exposure beyond ordinary earnings risk. Governments increasingly view cloud providers as critical infrastructure. Public agencies, banks, healthcare systems and large businesses depend on a small number of providers. Regulators may examine concentration, resilience, switching costs and the terms that make it difficult to move data or applications.

AI adds further questions. Customers need clarity about data use, model training, security, intellectual property and liability. Bedrock’s governance tools can become a competitive advantage if they help companies satisfy regulatory requirements. A major security failure or misuse of customer data would have the opposite effect.

Strategic investments in model developers may also receive competition scrutiny. Amazon’s relationship with Anthropic combines ownership, cloud supply and distribution. Similar arrangements exist across the industry. Regulators may ask whether such partnerships reduce competition, lock customers into one infrastructure provider or give cloud companies influence over model markets.

Energy and environmental policy can affect project economics. Data-center demand can raise local electricity requirements, strain water supplies and require new transmission. Permitting delays may become a practical limit on growth. Disclosure standards for energy use and emissions could increase compliance costs.

Governance inside Amazon is equally important. Management must balance AWS expansion with retail investment, employee costs, shareholder returns and financial resilience. The board’s oversight of capital allocation becomes more consequential when annual spending reaches a level larger than the economies of many countries.

Why Operating Margin Alone Cannot Measure Success

AWS’s 39.4% operating margin is a powerful signal, but it is not a complete return metric. Operating margin measures profit relative to revenue during a period. It does not show how much capital was required to create the assets generating that revenue.

Two businesses can report the same margin while producing very different returns on capital. A software company requiring little equipment may generate high cash returns. A data-center operator with the same margin may need constant reinvestment. Amazon increasingly combines both models: high-value software services running on enormous physical infrastructure.

Return on invested capital would be more informative, but public disclosures do not allocate enough assets and liabilities to AWS for a precise calculation. Investors can approximate trends by comparing segment operating income with companywide property investment, but such estimates mix retail, cloud and other projects.

Free cash flow adds another perspective, yet the current construction wave depresses it before revenue arrives. The most useful approach is to monitor a group of measures: AWS growth, operating margin, operating cash flow, capital expenditures, depreciation, debt and backlog conversion.

Success would look like sustained revenue growth, stable margins and eventual free-cash-flow recovery without continuously accelerating leverage. Failure could appear as strong accounting profit accompanied by persistent cash outflows, rising depreciation and underused assets.

The Quality of AI Demand Matters More Than the Label

Not every dollar described as AI revenue has the same economic quality. A recurring inference workload embedded in a customer-facing application may generate steady consumption for years. A one-time training run can create a large burst of revenue and then stop. An experimental pilot can disappear when a budget review arrives. A long-term capacity reservation may provide greater visibility, but its value depends on contract terms and counterparty strength.

Amazon’s disclosures combine several forms of demand. Frontier-model developers require large training clusters and inference capacity. Enterprises use Bedrock and related services to build internal tools and customer applications. Existing AWS customers add vector databases, storage, security and data-processing services around AI workloads. Each category has different growth, margin and retention characteristics.

Training is extremely compute intensive but can be episodic. Inference may begin smaller but become more durable as millions of users interact with deployed models. Enterprise applications can create high switching costs when they are integrated with proprietary data and workflow systems. The best long-term outcome for AWS would therefore be a shift from a few massive training customers toward a broad base of recurring inference and enterprise software workloads.

The available evidence suggests that both large and distributed demand contributed to the quarter. Anthropic was important, while Amazon also reported hundreds of thousands of Bedrock customers and growth in services used to deploy agents. Public reporting does not provide enough detail to calculate the revenue mix.

Future disclosures about customer count, consumption, remaining performance obligations and service mix would help investors assess demand quality. A high growth rate supported by diversified recurring usage deserves a different valuation from the same rate driven by a small number of capital-dependent customers.

Enterprise Customers Still Need to Prove Their Own Returns

Cloud providers can show strong AI revenue before their customers demonstrate equally strong financial returns. An enterprise may spend heavily on model access and compute while testing use cases. AWS recognizes revenue when it provides the service, even if the customer later decides the application did not justify the cost.

The next phase of adoption will depend on measurable outcomes. Customer-service systems must reduce handling costs or improve satisfaction. Coding tools must increase developer output without introducing unacceptable security or quality problems. Marketing systems must improve conversion enough to exceed model and data expenses. Industrial applications must reduce downtime or improve yield.

Many companies can fund experimentation for several quarters. Sustained production spending requires budget owners to compare AI projects with other investments. Cloud bills are visible and can rise quickly as usage expands. Cost optimization, which previously slowed AWS growth, can return if customers conclude that model size or frequency exceeds business value.

Amazon can reduce this risk by making workloads cheaper and easier to govern. Trainium is intended to lower compute costs. Bedrock offers model choice and enterprise controls. Databases, monitoring and agent tools can reduce the engineering work required to deploy applications. The more AWS improves customers’ total economics, the more durable its own revenue becomes.

The company’s second-quarter growth shows that customers are willing to spend. It does not yet show the average payback period of those customers’ projects. That distinction will become more important as AI moves from strategic experimentation to ordinary budgeting.

Trainium’s Real Test Is Adoption Without Financial Ties

Strategic partners can help a new hardware platform reach scale, but broad independent demand provides stronger proof of competitiveness. Anthropic’s use of Trainium is meaningful because its workloads are technically demanding. The relationship is also supported by Amazon investment and long-term commercial agreements.

For Trainium to become a durable alternative, customers with no material financial relationship to Amazon must choose it because performance, availability and total cost are attractive. Those customers will compare hardware speed, networking, memory, software tools, developer availability and migration expense.

Amazon has reported adoption by startups and large companies, and it says Trainium commitments are expanding. The company also benefits from Bedrock, where it can route suitable inference workloads to its own chips behind a managed service. Customers using a model API may care less about the underlying processor as long as price, latency and reliability are competitive.

This managed-service approach can accelerate chip utilization. It also makes independent evaluation harder because customers may not select the chip directly. Over time, benchmarks, customer case studies and broader third-party software support will indicate whether Trainium is competitive beyond Amazon-controlled services.

Nvidia’s ecosystem remains a formidable standard. Developers are familiar with its tools, and many models are optimized for its hardware. Amazon does not need to displace Nvidia entirely. Capturing a meaningful share of its own cloud workloads can reduce procurement dependence and improve negotiating leverage. Even partial success can have substantial economic value at AWS scale.

Capital Allocation Across Amazon Is Becoming Harder to Read

The $220 billion forecast covers Amazon as a whole, not only AWS. That breadth protects competitive information but makes investor analysis more difficult. A data center supported by a multiyear customer commitment has a different expected return from a speculative satellite project, a fulfillment center or a robotics program.

Amazon has several capital-intensive initiatives. It is expanding ultra-fast delivery, building technology infrastructure, developing custom chips, deploying warehouse robots and creating a low-Earth-orbit satellite network. Each project competes for cash, engineering talent and management attention.

Some investments reinforce one another. Better logistics can increase retail frequency and advertising inventory. Custom chips can lower AWS costs. Satellite connectivity can create new distribution or cloud opportunities. Other projects may remain separate and carry distinct risks.

Investors would benefit from greater disclosure of growth versus maintenance capex and the broad allocation among AWS, fulfillment and other projects. Management may resist because detailed figures could help competitors or create misleading precision. The absence of disclosure means outside analysts must avoid attributing the entire capital program to AI.

Capital allocation should also include acquisitions, strategic investments and debt. Amazon spent heavily on non-marketable investments during the first half and raised substantial long-term financing. The combined commitment extends beyond property purchases.

The relevant governance question is whether Amazon evaluates each project against a consistent return threshold or gives strategic initiatives more flexibility. Long-term thinking can create value when markets underappreciate future demand. It can also become a justification for weak accountability if project economics remain opaque.

Valuation Implications of a More Capital-Intensive Amazon

Traditional price-to-earnings analysis is particularly unreliable for this quarter. Net income includes the large Anthropic-related gain, while current depreciation may not yet reflect all assets under construction. A low earnings multiple calculated from reported profit would therefore overstate recurring earnings power.

Free-cash-flow valuation also requires judgment. Current free cash flow is negative because investment is unusually high. Treating negative $7.6 billion as a permanent run rate would ignore the revenue expected from projects under construction. Assuming free cash flow quickly returns to historical levels would be equally speculative.

A segment-based framework is more informative. Analysts can estimate the value of AWS using revenue growth, operating margin and normalized capital needs; value advertising based on growth and likely margins; and evaluate retail and logistics separately. The Anthropic stake can be considered as a financial asset with a discount for illiquidity and valuation uncertainty. Debt must then be subtracted.

The greatest uncertainty is normalized capex. If Amazon must spend near $220 billion every year simply to maintain competitive position, future cash conversion will be lower than in a scenario where spending falls after the present buildout. Small changes in that assumption create very large valuation differences.

AWS’s 37% growth supports a higher revenue outlook. The 39.4% margin supports strong operating value. Negative free cash flow and rising debt require a higher degree of caution. The stock’s post-market rally reflects improved expectations, not a final answer to normalized value.

Why Circularity Is a Risk to Analyze, Not a Reason to Dismiss Revenue

Amazon’s relationship with Anthropic has been described as circular because Amazon invests in the company and Anthropic commits to buy AWS services. Similar structures exist across the AI industry, where cloud providers supply both capital and computing infrastructure to model developers.

The word can imply that revenue is artificial, but the analysis should be more precise. AWS provides real computing services, incurs real costs and receives contractual payment. If Anthropic uses the capacity to train and serve models, the economic activity is genuine.

The risk arises from funding dependence. If a customer’s ability to pay depends partly on capital supplied by the vendor or by investors expecting continued valuation increases, demand may be less independent than ordinary enterprise consumption. A downturn in private funding could affect both Amazon’s investment value and cloud revenue.

The structure can still create strategic value. Amazon gains a large customer, improves Trainium through collaboration and distributes Claude through Bedrock. Anthropic receives capital and infrastructure. The arrangement may accelerate a platform that later becomes self-sustaining.

Investors should therefore track separate channels: cash invested in Anthropic, the carrying value of the stake, cash received for AWS services, contractual commitments and the customer’s broader financing. Combining them into one narrative can either exaggerate risk or hide it.

Operational Efficiency Outside AWS Can Finance the Buildout

Amazon’s ability to fund infrastructure depends partly on businesses outside AWS. North American retail operating income of $9.1 billion and International operating income of $1.7 billion provided meaningful cash-generating capacity. Advertising and third-party services improve the economics of the marketplace.

Delivery speed is one lever. Placing inventory closer to customers can lower transportation distance and improve order frequency, although it requires accurate forecasting. Robotics can reduce repetitive work and improve throughput. Higher-margin advertising and seller fees can offset the thin economics of first-party merchandise.

General and administrative expense declined in the quarter despite higher revenue, while sales and marketing grew modestly. These trends suggest operating discipline, although one quarter does not establish a permanent cost structure.

Efficiency matters because Amazon is funding several long-term projects simultaneously. Every percentage point of retail margin on more than $150 billion of quarterly combined North American and International sales has substantial value. The non-cloud businesses do not need AWS-level margins to contribute materially.

The risk is that management cuts too deeply in customer service, safety, innovation or employee capability to finance capital spending. Cost reductions that improve near-term margins can create long-term operational problems. The best outcome combines productivity gains with service quality, rather than shifting costs to workers, sellers or customers.

Signals That Would Challenge the Current Optimism

Several developments would weaken the positive interpretation of the quarter. A rapid return of AWS growth below 25% would suggest the acceleration was concentrated or temporary. A decline in AWS margin while capex remains elevated would indicate that new capacity is less profitable or depreciation is catching up.

A backlog decline or reduced commentary about reservations would raise questions about demand visibility. Delays in Trainium deployment or customer migration toward competing chips would weaken the custom-silicon thesis. Additional large debt issuance without corresponding operating-cash-flow growth would increase financing concern.

Amazon could also face pressure if Anthropic’s valuation falls or if the company reduces spending. The effect would extend beyond non-operating income because Anthropic is a major infrastructure customer. Regulatory restrictions on strategic cloud partnerships could complicate the arrangement.

Outside AWS, weaker consumer demand or rising delivery costs could reduce the cash available to finance investment. Advertising growth could slow if sellers resist higher marketing costs or regulators change marketplace rules.

These signals are not predictions. They are observable indicators that can test whether the second-quarter result marks a durable improvement.

A Detailed Cash-Flow Bridge for the Second Quarter

The gap between Amazon’s reported net income and cash generation is unusually large and deserves a line-by-line explanation. The company started with $62.6 billion of second-quarter net income. It then added non-cash expenses such as $20.0 billion of depreciation and amortization and $6.0 billion of stock-based compensation. Those additions normally make operating cash flow higher than net income.

The Anthropic-related accounting gain moved the calculation in the opposite direction. Amazon subtracted $53.4 billion of non-operating income when reconciling net income to cash from operations because the gain did not represent operating cash received during the quarter. Deferred tax expense added $17.7 billion, while changes in receivables, other assets, inventory and liabilities produced additional movements.

After those adjustments, operating cash flow was $45.4 billion. Amazon then spent $54.2 billion on property and equipment and received $1.1 billion from property sales and incentives. The difference between operating cash flow and net property purchases was an outflow of approximately $7.7 billion for the quarter.

The investing section included another $24.4 billion outflow for acquisitions, non-marketable investments and other activity. That category is separate from Amazon’s free-cash-flow definition. It helps explain why total investing cash outflow reached $79.2 billion and why the company relied on financing.

Financing activities provided $10.1 billion during the quarter, including $13.6 billion of long-term debt proceeds, partially offset by repayments and other items. Across the first six months, financing activities provided $62.9 billion, reflecting the much larger debt program.

This bridge reveals three distinct stories. Core operations produced substantial cash. Property investment consumed more than that operating cash. Strategic investments and other investing activity required additional financing. Describing the entire movement as either healthy growth or dangerous cash burn would omit important parts of the picture.

What a 39.4% AWS Margin Does and Does Not Prove

AWS’s operating margin increased approximately 6.5 percentage points from the year-earlier quarter. That is a large improvement for a business of this size. It indicates strong utilization, favorable service mix, cost control or some combination of those factors.

The margin also gives Amazon room to absorb future cost pressure. Even if depreciation, energy and labor expenses rise, the segment begins from a profitable base. This is one reason the market treated Amazon’s capex differently from spending by a company without a mature cloud profit pool.

The figure does not isolate the profitability of new AI infrastructure. The margin combines older and newer AWS assets, traditional cloud workloads, AI services and software products. Existing facilities may be highly depreciated or fully utilized, while newly opened projects may have different economics.

Nor does one quarter establish a normalized margin. Customer credits, capacity timing, server useful lives and workload mix can move results. A better test will be whether margins remain strong as the 2026–2028 construction program enters service.

Amazon’s Long-Term Architecture: From Infrastructure to Applications

Amazon’s strategic ambition extends from physical infrastructure to end-user applications. At the bottom of the stack are data centers, power, networking and chips. Above them are EC2, storage, databases and machine-learning services. Bedrock provides model access and orchestration. Amazon’s retail, advertising, Alexa and workplace tools then use AI directly.

This architecture allows value to move through the organization. A custom chip can lower AWS costs. AWS can sell the same capability to external customers. Internal teams can use the infrastructure without paying a third-party cloud margin. Consumer products can create more engagement and advertising inventory.

Vertical integration can also create conflicts. External customers may question whether Amazon competes with them using insights or infrastructure advantages. Model developers may resist dependence on one provider. Regulators may examine bundling and switching costs.

The economic opportunity is nevertheless substantial. Amazon does not need every layer to lead independently. A competitive infrastructure platform can support a broad software ecosystem, while successful applications improve utilization of the underlying assets.

The second-quarter results show that the infrastructure layer is monetizing rapidly. The next question is whether higher-level software and applications increase customer value enough to sustain that growth after the current capacity shortage eases.

Why This Earnings Report Matters Beyond Amazon

Amazon’s quarter influences decisions far beyond one stock. Semiconductor manufacturers use cloud-provider capital plans to forecast demand for accelerators, memory and networking equipment. Utilities and energy developers use data-center pipelines to plan generation and transmission. Construction companies, real-estate developers and local governments evaluate land, water and grid requirements. Software companies decide which cloud ecosystems to support.

The result also affects corporate technology budgets. When AWS, Azure and Google Cloud all report rapid AI-related growth, chief information officers face pressure to develop their own strategies. That can accelerate adoption, but it can also encourage spending before organizations have identified clear returns.

Financial markets use the cloud providers as a test of the broader AI cycle. Strong AWS revenue suggests demand is real at the infrastructure layer. Negative free cash flow and rising debt show that the cycle is not costless. Both signals are relevant to chip companies, data-center operators and businesses supplying power equipment.

For economic policymakers, the buildout raises questions about productivity and resource allocation. If AI infrastructure improves business efficiency across industries, the investment can support long-term growth. If too much capital is concentrated in overlapping facilities and speculative applications, the adjustment could affect technology employment, credit markets and regional power systems.

Amazon’s report therefore provides neither a simple confirmation nor rejection of an AI bubble. It shows an industry with enormous current demand, strong provider profits and equally enormous capital requirements. The balance between those forces will shape technology investment for the remainder of the decade.

The research cutoff is also important. Amazon had furnished its earnings exhibit and released financial tables, but a full second-quarter Form 10-Q was not yet available in the sources reviewed before publication. The eventual filing may add detail on investments, debt, commitments, accounting policies and risk factors. Readers should treat the earnings release and conference-call reporting as the most current company information at this cutoff, while recognizing that later filings can refine the picture.

That distinction does not weaken the central findings, which are supported by the filed earnings exhibit. It does mean that leverage, contractual obligations and investment accounting should be revisited when the complete quarterly report becomes available. Financial analysis is strongest when conclusions remain open to better disclosure rather than treating an early earnings release as the final record.

Frequently Asked Questions

How much did AWS revenue grow in Amazon’s second quarter of 2026?

AWS revenue increased 37% year over year to $42.2 billion in the quarter ended June 30, 2026. It was the division’s fastest growth in 18 quarters.

Why was the AWS result considered remarkable?

The growth occurred on a very large revenue base, accelerated sharply from prior quarters and was accompanied by a 39.4% operating margin. AWS also exceeded analyst growth expectations cited by Reuters.

How profitable was AWS?

AWS generated $16.6 billion of operating income on $42.2 billion of revenue, an operating margin of approximately 39.4%. It contributed about 60.5% of Amazon’s consolidated operating income.

What does the $169 billion AWS run rate mean?

It is four times the latest quarterly revenue of approximately $42.2 billion. It illustrates the current scale of AWS but is not formal guidance for the next 12 months.

Why did Amazon’s free cash flow turn negative?

Trailing operating cash flow rose to $161.4 billion, but net purchases of property and equipment increased to $169.0 billion. Under Amazon’s definition, that produced a trailing free-cash-flow outflow of $7.6 billion.

How much does Amazon expect to spend on capital expenditures in 2026?

Management raised its expected 2026 cash capital expenditures to approximately $220 billion from approximately $200 billion. The spending supports AWS and other company projects.

Did Amazon really earn $62.6 billion in the quarter?

Amazon reported net income of $62.6 billion, but the figure included $53.4 billion of non-operating pre-tax other income, primarily related to Anthropic. Operating income of $27.5 billion is more representative of the underlying businesses.

What role does Anthropic play in Amazon’s strategy?

Anthropic is an investment, a major AWS customer, a Trainium development partner and a model provider on Bedrock. The relationship supports demand but also creates concentration and accounting complexity.

Is AWS growing faster than Microsoft Azure and Google Cloud?

No. Microsoft reported 43% Azure growth and Alphabet reported 82% Google Cloud growth for comparable quarter-end dates. The metrics are not directly comparable because the companies use different segment definitions and disclosures.

Why did Amazon stock rise after earnings despite negative free cash flow?

The market appeared to focus on stronger-than-expected AWS growth, higher cloud margins, expanding backlog and evidence that capital spending was tied to customer demand. Reuters reported a nearly 9% post-market increase.

What is the biggest risk to Amazon’s AI spending plan?

The largest broad risk is that Amazon builds expensive capacity based on demand that later slows or shifts, leaving underused assets and high depreciation. Customer concentration, technology obsolescence and financing costs add to that risk.

What should investors watch in the next earnings report?

The most useful indicators will be AWS growth and margin, capital expenditures, operating cash flow, debt, backlog conversion and management’s comments about 2027 capacity.

Final Assessment

Amazon’s second-quarter 2026 results materially strengthened the case for its AI infrastructure strategy. AWS grew 37% to $42.2 billion, produced a 39.4% operating margin and generated more than 60% of consolidated operating income. The acceleration was broad enough to counter the immediate concern that AWS had lost momentum to Microsoft and Google.

The quarter also made the financial cost impossible to ignore. Trailing free cash flow was negative $7.6 billion, property investment rose sharply, long-term debt increased to $128.9 billion and the expected 2026 capital program reached approximately $220 billion. Reported net income was heavily influenced by a non-operating Anthropic gain.

The strongest evidence in Amazon’s favor is not management enthusiasm. It is the combination of current cloud revenue, expanding margin, contracted demand and backlog. The strongest concern is not that AI produces no revenue. It is that the industry may be committing capital faster than long-term demand, pricing and equipment life can justify.

Amazon has shown that its AI spending can produce substantial near-term business. It has not yet shown what normalized free cash flow will look like after the buildout, how concentrated future demand is or whether current AWS margins can survive the arrival of enormous new capacity. Those questions will determine whether the $220 billion plan becomes another example of Amazon investing ahead of a durable market or an unusually expensive period of overbuilding.

Sources

• Amazon.com Q2 2026 earnings release and financial tables
• Amazon Investor Relations: second-quarter 2026 results
• Amazon SEC-filed Q2 2026 earnings exhibit
• Reuters: Amazon lifts investment plans after strong cloud sales
• Amazon Investor Relations: first-quarter 2026 results
• Amazon Investor Relations: fourth-quarter 2025 results and 2026 investment outlook
• Amazon 2025 annual report filed with the SEC
• Anthropic: expanded Amazon compute partnership
• Amazon: expanded Anthropic investment and Trainium collaboration
• Amazon: Andy Jassy on the company’s chip business
• Microsoft fiscal 2026 fourth-quarter earnings release
• Alphabet second-quarter 2026 earnings release
• Meta second-quarter 2026 earnings release
• Synergy Research Group cloud infrastructure market analysis

This article is provided for general informational purposes and does not constitute financial, investment, tax, or legal advice.

Affiliate disclosure: Businessfinance.news may earn compensation from qualifying actions completed through selected links on this website, at no additional cost to the reader. Affiliate relationships do not influence our editorial reporting, analysis, or conclusions.