Amazon’s cloud division prints profits. Its retail machine prints shipping labels by the billion. If I had to choose which one will drive the next chapter of shareholder value, the answer looks obvious—until I examine what the two businesses actually do for each other.
Whenever I look at Amazon, I feel as though I am analyzing three companies wearing one trench coat.
There is the retailer that sells everything from televisions to emergency quantities of paper towels. There is the marketplace and logistics empire that lets third-party merchants borrow Amazon’s traffic, warehouses, delivery network, and increasingly large collection of fees. Then there is Amazon Web Services, the cloud-computing business that quietly became the financial adult in the room while the retail operation continued buying vans, building fulfillment centers, and promising to deliver dental floss before I remember ordering it.
The standard investor answer to “AWS or retail?” is AWS. Of course it is. AWS grows faster, earns far higher margins, and occupies the glamorous center of the artificial-intelligence buildout. Retail moves boxes. AWS rents out the digital nervous system of modern business. One operation worries about cardboard dimensions; the other helps companies train frontier models.
That comparison is directionally right, but it is also too simple.
Amazon retail is no longer merely an online store. It includes a marketplace, fulfillment services, Prime memberships, grocery, pharmacy, logistics, devices, streaming, and an advertising operation that would be a major company on its own. AWS may be the most powerful standalone value creator, but the retail ecosystem supplies Amazon with reach, cash flow, consumer habits, advertising inventory, and a laboratory for logistics and artificial intelligence.
So my conclusion is not that AWS is brilliant and retail is an expensive hobby with forklifts. My conclusion is more specific: AWS is likely to create more incremental shareholder value from here, while retail remains the strategic foundation that makes Amazon unusually difficult to compete with.
That may sound like I am generously awarding trophies to everyone. I am not. If I must pick one winner, I pick AWS. I simply refuse to insult the machine that helped create it.
The latest numbers are not subtle
Amazon’s second-quarter 2026 results make the basic financial argument almost comically clear.
AWS generated $42.2 billion in quarterly sales, up 37% from the prior year, and produced $16.6 billion of operating income. That works out to an operating margin of roughly 39%. North America and International, which together capture most of what investors casually call “retail,” generated about $158.4 billion in combined sales and $10.8 billion of operating income.
In other words, AWS produced substantially more operating profit on barely more than one-quarter of the combined sales of the two consumer-facing segments. Retail did the marathon. AWS arrived in a golf cart carrying the larger check.
The six-month comparison is just as revealing. AWS posted $79.8 billion in sales and $30.8 billion in operating income during the first half of 2026. North America and International combined for roughly $302.3 billion in sales and $20.5 billion in operating income. The precise comparison is imperfect because Amazon’s segments contain different mixes of businesses and shared corporate costs, but the message survives every reasonable adjustment: AWS converts revenue into profit at a rate the retail operation cannot approach.
This matters because shareholder value is not awarded for moving the most merchandise. Revenue is impressive, but revenue without attractive economics is mostly proof that a company is extremely busy. Grocery stores generate enormous sales on thin margins. Software companies can create extraordinary value from smaller revenue bases because each additional dollar carries more profit.
Amazon’s Q2 2026 earnings release showed total company sales rising 20% to $200.6 billion and operating income climbing 43% to $27.5 billion. AWS represented only 21% of sales but roughly 61% of segment operating income. That is not a supporting character. That is the actor carrying the picture while retail handles the explosions.
Why AWS has the cleaner path to value creation
The AWS case begins with growth, but it does not end there.
At 37% year-over-year growth in Q2 2026, AWS recorded its fastest expansion in 18 quarters and reached a $169 billion annualized revenue run rate. Businesses are migrating more workloads to the cloud, modernizing old systems, building data platforms, deploying generative AI, and experimenting with agents. All of this requires computing, storage, databases, networking, security, model access, and increasingly specialized chips.
AWS sells the picks, shovels, roads, power connections, rental property, and tollbooths of the digital gold rush. It is a wonderfully considerate arrangement—for Amazon.
The appeal is not merely that AI demand is large. The appeal is that AI broadens the number of services AWS can sell into existing customer relationships. A company may begin with basic computing and storage, then add databases, analytics, cybersecurity, model hosting, training infrastructure, inference, developer tools, and custom silicon. Each additional service deepens the relationship and raises switching costs.
Cloud infrastructure is not frictionless to replace. Moving significant workloads can involve rewriting applications, transferring data, retraining employees, changing security controls, and accepting operational risk. Customers can pursue multi-cloud strategies, and many do, but complexity itself often favors the established providers. Once a business has built deeply inside an ecosystem, leaving may resemble moving an apartment building one brick at a time.
AWS also benefits from scale. Amazon can spread the cost of data centers, networking, software development, and custom chips across a massive customer base. Its Graviton processors and Trainium AI chips give it another potential advantage: lowering its own infrastructure costs while offering customers alternatives to more expensive hardware.
Amazon said in July 2026 that both its AI business within AWS and its chips business had exceeded annual revenue run rates of $25 billion, with each growing at triple-digit rates. Those figures are management disclosures rather than separately reported accounting segments, so I would not treat them like carved stone tablets. Still, they show where demand and investment are concentrating.
If AWS can keep revenue growing at a healthy rate while sustaining margins anywhere near the mid-to-high 30% range, the earnings power becomes enormous. Even a slower growth rate on a revenue base approaching $150 billion on a trailing basis can add billions in annual operating income. That incremental profit can fund more infrastructure, acquisitions, buybacks, or expansion elsewhere in Amazon.
This is the compounding engine I would choose.
The catch: AWS now eats capital for breakfast
High margins can create the illusion that AWS is an easy business. It is not. Cloud and AI infrastructure require astonishing amounts of capital: data centers, servers, networking equipment, land, power agreements, cooling, and chips whose prices have apparently been negotiated by jewelers.
Amazon’s trailing-12-month purchases of property and equipment, net of proceeds and incentives, reached about $169 billion by June 2026, up 64% from a year earlier. The company said the increase primarily reflected AI investment. Operating cash flow rose to $161.4 billion, yet free cash flow fell to an outflow of $7.6 billion.
That is the less photogenic side of the AI boom. AWS may generate extraordinary operating income, but Amazon must spend heavily before new capacity produces revenue. The timing mismatch can crush near-term free cash flow even when demand is strong.
Management expected roughly $200 billion in companywide capital expenditures for 2026. Not revenue. Capital expenditures. Enough money to make almost any corporate project sound like someone forgot to carry a decimal.
The bull case says this spending reflects demand that Amazon can monetize for years. The bear case says hyperscalers are racing to build capacity before anyone knows what normalized AI economics will look like. Customers may become more efficient. Model costs may fall. Competition may pressure prices. Hardware can become obsolete quickly. Electricity availability can delay projects. A data center is not magically valuable because someone used the phrase “generative AI” during the groundbreaking ceremony.
I believe AWS will earn attractive returns on a meaningful portion of this investment, but shareholders must monitor the conversion from capital spending to revenue, operating profit, and eventually free cash flow. Growth purchased at any price is not value creation. It is enthusiasm with depreciation schedules.
Retail has worse margins—and more hidden weapons
The retail side looks inferior when I compare segment margins, but this is where the analysis can become unfair.
Amazon’s consumer segments do far more than buy products wholesale and resell them. Third-party sellers pay commissions, fulfillment fees, storage fees, and other charges to access Amazon’s marketplace and logistics network. Prime members pay subscriptions. Brands pay for advertising. Merchants use Amazon’s payment and shipping infrastructure. Consumers increasingly use pharmacy, grocery, streaming, and other services.
The low-margin sale may be the front door to several higher-margin revenue streams.
Advertising is the clearest example. A shopper searching Amazon for running shoes is not casually admiring the internet. That person is expressing purchase intent. Brands will pay heavily to appear at the exact moment a wallet begins clearing its throat.
Amazon reported that advertising grew 26% year over year in Q2 2026. The company does not report advertising as a separate operating segment, so investors cannot cleanly isolate its profitability. I consider it reasonable to infer that advertising carries margins far above first-party retail, but it would be foolish to pretend I know the exact number.
That advertising business exists because retail attracts shoppers. Prime increases shopping frequency. Fast delivery makes Amazon the default starting point for product searches. The marketplace expands selection without requiring Amazon to own every item. More selection attracts more customers, and more customer traffic attracts more sellers and advertisers. Each part strengthens the others.
This flywheel is not fresh investor poetry invented for an earnings call. It is the economic architecture of the business.
Retail also owns a physical network that took decades and vast sums to build. Fulfillment centers, sortation facilities, delivery stations, aircraft capacity, software, robotics, and last-mile operations create convenience that smaller competitors cannot easily replicate. Amazon delivered more than 40% more items same-day or overnight for Prime members in the first half of 2026, according to management.
Fast delivery is expensive, but speed changes customer behavior. If Amazon can reliably deliver household essentials the same day, it gains purchase frequency in categories that once belonged to grocery stores, pharmacies, and big-box retailers. Those purchases may carry thin direct margins, yet they reinforce memberships, advertising, seller services, and route density.
Retail’s income statement does not fully express the option value embedded in that network.
Retail’s margin opportunity is real but finite
Amazon has already shown that retail profitability can improve.
For full-year 2025, North America operating income reached $29.6 billion on $426.3 billion in sales, while International produced $4.7 billion on $161.9 billion in sales. In 2024, those figures were $25.0 billion and $3.8 billion, respectively. The international business, once treated as a dependable method for setting money on fire in several currencies, became meaningfully profitable.
Regional fulfillment, better inventory placement, faster delivery, increased automation, improved labor productivity, and a richer mix of advertising and third-party services can continue lifting margins. Amazon’s logistics system becomes more efficient as packages travel shorter distances and facilities handle greater local density. Robotics can reduce certain costs. AI can improve forecasting, customer service, product discovery, and seller tools.
But retail still faces structural limits.
Products cost money. Warehouses cost money. Delivery costs money. Returns cost money, particularly when consumers order six sizes and send back a small textile referendum. Grocery margins are thin. Wage inflation matters. Fuel and transportation costs matter. Tariffs, inventory mistakes, shrinkage, and competition matter.
Walmart, Costco, Target, grocery chains, specialized retailers, direct-to-consumer brands, and low-cost international platforms are not planning to hand Amazon the remaining commerce market as a gesture of respect. Physical retailers also have local stores that can function as fulfillment nodes, pickup points, and return centers.
I expect Amazon retail margins to improve over time, especially as services become a larger part of the mix. I do not expect the core physical-goods business to wake up one morning with AWS economics. A delivery van cannot serve unlimited customers at near-zero incremental cost. Physics remains stubbornly unwilling to become software.
Retail creates strategic value that the segment numbers miss
If Amazon separated AWS tomorrow, the cloud company would remain formidable. It has its own customers, technology, contracts, ecosystem, and brand. AWS does not need me to order laundry detergent to survive.
Yet the relationship between AWS and retail still matters.
Amazon’s retail operation gave the company an early internal customer for cloud infrastructure. The extreme demands of global commerce—traffic spikes, payments, recommendations, logistics, inventory, streaming, and marketplace operations—forced Amazon to build capabilities that became useful to outside customers. Retail served as both proving ground and reference case.
That continues with AI, chips, robotics, and logistics. Amazon can deploy technology internally at enormous scale, learn from the results, and turn some capabilities into external services. Its retail business offers data and operational complexity that few companies can reproduce in a laboratory.
The consumer ecosystem also creates strategic freedom. Amazon can accept lower margins in one area to strengthen the whole. Prime Video can improve membership retention. Fast delivery can increase shopping frequency. Advertising can subsidize better customer economics. Third-party fees can monetize infrastructure. Pharmacy can deepen household relationships. The pieces need not maximize profit independently if they increase the value of the network.
This is why a sum-of-the-parts valuation can mislead. Separating businesses on a spreadsheet is easy. Separating the advantages they exchange is harder.
Still, synergy should not become a magical word used to excuse every expense. Investors should demand evidence that newer retail initiatives improve engagement, revenue quality, or long-term returns. “It strengthens the ecosystem” is useful only when the ecosystem eventually strengthens per-share cash flow.
My scoreboard: AWS wins four categories, retail wins two
When I compare the businesses, I use six basic questions.
Growth: AWS wins. Q2 growth of 37% on a $42.2 billion quarterly base is exceptional. Retail’s 15% to 16% segment growth was strong, but AWS has the clearer exposure to cloud migration and AI infrastructure.
Margins: AWS wins decisively. A roughly 39% quarterly operating margin belongs in a different financial universe from North America’s roughly 8% and International’s roughly 4%.
Capital efficiency: This one is complicated, but I still give AWS a narrow win over a full cycle. Both businesses consume heavy capital. Retail requires inventory and logistics; AWS requires data centers and chips. Current AI spending makes AWS look ravenous, yet mature cloud capacity can support recurring, high-margin revenue for years. The verdict depends on utilization and technological obsolescence.
Competitive durability: AWS wins narrowly. Enterprise relationships, technical integration, scale, security, and a broad service portfolio create strong switching costs. Retail’s Prime ecosystem and logistics network are also durable, but consumer switching costs are lower. I can shop elsewhere in minutes. Migrating a bank’s core infrastructure may require more than clearing browser cookies.
Optionality: Retail wins. Its marketplace, ads, logistics, grocery, pharmacy, media, devices, and membership model provide numerous paths to monetize the same customer relationship.
Strategic breadth: Retail wins. It makes Amazon relevant to everyday households and millions of merchants while providing an operating laboratory for technology. AWS is a superior financial business; retail is a broader strategic platform.
The score is four to two for AWS. Unlike certain sports commentators, I do not need to pretend the contest is tied to keep viewers through the commercial break.
What could make my conclusion wrong?
The most obvious risk is that AI infrastructure becomes overbuilt.
If Amazon spends hundreds of billions on capacity that customers do not use at attractive prices, AWS returns could disappoint despite strong revenue growth. Competition from Microsoft Azure, Google Cloud, Oracle, specialized GPU clouds, customer-owned infrastructure, and new technical architectures could pressure margins. Custom chips may succeed, or customers may continue preferring other hardware. AI inference may become more efficient faster than demand expands.
Regulation is another risk. Cloud providers face scrutiny over market power, data sovereignty, security, energy use, and business practices. Governments may require more local infrastructure or impose restrictions that raise costs.
Retail could also surprise me on the upside. Advertising may become a much larger profit pool. Fulfillment automation could drive margins higher than I expect. Amazon’s logistics services could develop into a major third-party platform. Grocery or pharmacy could reach scale. AI shopping agents might strengthen Amazon’s control over product discovery rather than weaken it.
There is also a strange possibility: AI agents could change shopping so dramatically that consumers no longer browse marketplaces in the same way. If software chooses products on our behalf, the value of Amazon’s search advertising and interface could shift. Amazon is building its own shopping agents, but technological transitions rarely distribute prizes according to the current market-share table.
Finally, retail has sheer scale. A one-percentage-point improvement in the operating margin of a business generating hundreds of billions in annual sales can create several billion dollars of additional profit. AWS has better economics, but retail’s enormous revenue base means modest improvements matter.
My verdict: AWS creates more value, retail makes that value harder to attack
If I could own only one of the businesses as a standalone company, I would choose AWS.
I would be buying faster growth, higher margins, recurring enterprise demand, meaningful switching costs, and direct exposure to cloud computing and AI. I would also accept brutal capital requirements, formidable competitors, and the risk that today’s spending runs ahead of tomorrow’s economics.
AWS is the clearer incremental earnings engine. Its Q2 2026 operating income exceeded the combined operating income of North America and International by nearly $5.8 billion, despite generating less than one-third as much revenue. Its trailing-12-month operating income reached about $54.7 billion. If demand remains strong and new capacity earns sensible returns, AWS should contribute a disproportionate share of Amazon’s profit growth.
But I would not casually discard retail.
Retail gives Amazon hundreds of millions of customer relationships, a global merchant ecosystem, a valuable advertising surface, recurring Prime revenue, and physical infrastructure that would be nearly impossible to recreate quickly. It turns everyday transactions into multiple layers of monetization. It also gives Amazon something most cloud competitors lack: a giant internal proving ground spanning commerce, entertainment, advertising, logistics, devices, and consumer AI.
The best outcome for shareholders is not AWS defeating retail. It is AWS producing high-margin growth while retail becomes more efficient and shifts toward services. AWS can fund innovation across the company. Retail can deepen the moat, feed advertising, strengthen Prime, and create new platforms from logistics and commerce infrastructure.
I therefore see AWS as the engine and retail as the chassis. The engine creates the force. The chassis makes it useful, durable, and difficult to push off the road.
Amazon will continue spending enough money to make free-cash-flow analysis feel like weather forecasting during a hurricane. That means investors should watch capital spending, AWS growth, AWS margins, retail margin progression, advertising growth, and the eventual recovery of free cash flow. The company does not earn extra credit merely for making large investments. Those investments must produce per-share value.
From here, my money is on AWS to create more of that value.
Retail will remain the larger business by revenue, the more visible business to consumers, and the foundation of Amazon’s broader ecosystem. But AWS has the combination shareholders usually spend decades searching for: scale, growth, attractive margins, embedded customers, and expanding demand.
Amazon retail taught the company how to build infrastructure for itself. AWS taught Amazon how profitable it could be to rent infrastructure to everyone else.
That is the business I expect to carry the larger share of Amazon’s next wave of shareholder value—even if the boxes on my porch keep getting all the attention.
Sources
This article is commentary for informational purposes and is not individualized investment advice.
Comments
Post a Comment