What is Amazon's structural advantage in logistics, and can anyone compete
Why Amazon Is So Hard to Beat at Delivery — and What Would Have to Change
Based on analysis of a 91-node, 319-edge knowledge graph mapping Amazon’s logistics structure, competitive dynamics, and counter-pressures.
The Lemonade Stand That Sells Cloud Computing
Imagine a kid with a lemonade stand. The lemonade barely makes money, but the kid also runs a tutoring service out of the same house that makes a lot of money. The kid uses the tutoring profits to undercut every other lemonade stand in town — selling cups at cost, or even a little below — until no competitor can keep up.
That is the core of what this analysis found. Amazon’s delivery business is not really paying for itself. Amazon Web Services — the cloud computing business that companies, hospitals, and governments rent computing power from — generates enormous profits. Those profits flow into Amazon’s shipping, warehousing, and delivery operations, allowing Amazon to build and run a logistics system at a cost no shipping-only competitor can match.
The graph identified this as the single most important structural mechanism: the AWS profit engine funding logistics. It connects to at least eight major parts of the delivery system, including warehouse robots, drone delivery, self-driving trucks, and the seller fee structure. This is not a side detail — it is load-bearing. Remove it, and the delivery economics look very different.
Three Things That Hold Everything Together
The analysis found that the entire 91-node map essentially flows into or out of three central points.
The first is Amazon’s full ownership of the supply chain. Amazon does not just ship boxes — it warehouses inventory, forecasts what you will order before you order it, picks the item, packages it, loads it onto an Amazon plane, transfers it to an Amazon truck, and delivers it to your door. Every one of those steps used to be handled by separate companies (warehouses, freight brokers, FedEx, UPS). Amazon has replaced most of them with its own infrastructure. The graph calls this “Complete Vertical Stack Capture” and it is the node with the most connections in the entire map.
The second is seller lock-in. If you are a business selling products on Amazon, you are strongly incentivized to use Amazon’s own storage and fulfillment service (called FBA, Fulfillment by Amazon). The reason is simple: Amazon’s search result system — specifically, which product gets the “Buy Box,” the default button customers click — favors sellers who use FBA. If you ship your own products, you are less likely to be chosen automatically by Amazon’s system. This creates a loop: FBA gives you the Buy Box, and the Buy Box forces you to use FBA. Three separate Federal Trade Commission (FTC) investigations target this specific mechanism.
The third is the AWS profit subsidy. As described above, cloud profits fund logistics expansion. This node also receives profits from Amazon’s advertising business — when companies pay to have their products appear at the top of Amazon search results — and those advertising revenues feed back into the same subsidy pool.
Together, these three nodes account for more than a third of all connections in the graph.
How Physics Makes It Harder to Copy
One of the less obvious findings is that Amazon’s advantage is not just about money or technology. There is a physical dimension that capital alone cannot solve quickly.
Think about delivering to a neighborhood. If a delivery driver goes to a street with one package, the cost of that delivery is very high — almost all of it is the driver’s time driving there. If the same driver has 30 packages on that street, the cost per package drops dramatically. This is called delivery density, and it is the reason Amazon’s cost per delivery is lower than a small competitor’s, even if the small competitor builds the same vans and hires the same drivers.
Amazon’s volume — billions of packages per year — means its drivers almost always have dense, efficient routes. A new competitor starting from scratch would need to build that volume before the unit economics improve, but without good unit economics, it is hard to attract the volume. The graph identifies residential delivery density as the foundational physical constraint from which most of Amazon’s economic advantages derive. It is not a strategy — it is closer to a law of physics for the business.
The UPS and FedEx Retreat
The graph captures something that has been publicly announced but is worth seeing in structural terms: UPS and FedEx are both, in different ways, pulling back from competing with Amazon on standard parcel delivery.
UPS announced in January 2025 that it would significantly cut its Amazon business (which had become a low-margin drag) and refocus on healthcare and business-to-business shipping. FedEx has been restructuring toward the same direction. The graph shows these decisions not as isolated corporate choices but as part of a reinforcing cycle: as Amazon handles more of its own packages internally, the volume available to UPS and FedEx shrinks; as their volume shrinks, their fixed costs get spread over fewer packages, making them less efficient; as they become less efficient, they retreat to higher-margin niches — which frees up more standard parcel volume for Amazon to absorb.
The graph labels this a “reverse network spiral” and models it as self-reinforcing. The longer it continues, the harder it is to reverse.
The Regulatory Knot
The FTC has opened multiple investigations and cases targeting Amazon’s logistics practices — specifically the Buy Box mechanism, the seller lock-in, and what the graph calls “self-preferencing” (Amazon prioritizing its own products and logistics services over competitors’). These are real constraints with high-weight edges in the graph.
But the analysis surfaces a non-obvious complication: Amazon has been offering its fulfillment network to businesses that sell on other platforms — Shopify stores, TikTok sellers, and others — through a service called Multi-Channel Fulfillment (MCF). As more non-Amazon sellers depend on Amazon’s logistics infrastructure, the political and legal complexity of forcing Amazon to separate its delivery business from its marketplace increases. Enforced separation would now harm third parties who chose Amazon as their logistics provider. The graph models this as a deliberate or at least structurally useful side effect: by becoming the logistics backbone for competitors, Amazon has made it harder for regulators to surgically cut logistics away from the marketplace without collateral damage.
The USPS adds a second complication. The United States Postal Service has become financially dependent on Amazon delivery volume. This creates a political constraint: aggressive antitrust action against Amazon could damage USPS revenues, which is a politically sensitive outcome. The graph shows USPS dependency both limiting antitrust pressure and strengthening Amazon’s ability to negotiate low rates from USPS — the dependency works in Amazon’s favor on both fronts.
Things That Could Actually Change the Picture
The analysis identifies several real tensions — places where Amazon’s own actions create problems for itself, or where external forces could shift the structure.
Amazon’s robots might undercut Amazon’s own delivery model. Amazon built its DSP program (Delivery Service Partners — small, Amazon-contracted delivery companies) partly because DSP labor is cheaper than UPS or FedEx union labor. But Amazon is also investing heavily in warehouse robots, self-driving delivery vans, and autonomous trucks. As automation replaces workers, the labor-cost advantage that made DSPs attractive shrinks. The graph models this as Amazon’s automation investment partially undermining its own labor moat.
AI shopping agents could weaken the Buy Box. If consumers increasingly shop through AI assistants that search across multiple retailers and choose the best option automatically, Amazon’s Buy Box — which only matters on Amazon’s own platform — loses influence. The graph assigns this a high undermining weight against several of Amazon’s key advertising and discovery mechanisms. Simultaneously, Amazon’s cloud infrastructure (AWS) is positioned to run many of those same AI agents. Whether Amazon’s infrastructure advantage compensates for its platform disadvantage in an AI-shopping world is unresolved.
MercadoLibre shows it can be done, somewhere. The analysis includes one node that directly challenges the idea that Amazon’s logistics model is globally unreplicable: MercadoLibre, the dominant e-commerce and logistics operator in Latin America. MercadoLibre has built a full-stack logistics system — its own warehouses, last-mile delivery, financial services — that mirrors Amazon’s architecture. The graph models this as evidence that regional incumbents can pre-position before Amazon arrives at scale. Whether this scales to a global pattern or is specific to markets Amazon has not prioritized is an open question.
Bottom Line
The graph’s structural picture can be summarized in four sentences.
Amazon’s logistics advantage is not primarily a logistics advantage — it is a financial architecture advantage, where cloud and advertising profits subsidize delivery costs that no logistics-only competitor is designed to absorb. The physical reality of delivery density means that volume itself is a moat: you need scale to have low costs, and you need low costs to attract scale. The carrier retreat by UPS and FedEx is a reinforcing cycle, not a one-time event, and the graph models it as ongoing and self-amplifying. Regulatory tools exist and are actively applied, but their effectiveness is structurally complicated by Amazon’s expansion into serving competitors and by the financial dependency it has created in the USPS.
The most important single uncertainty is what happens if the AWS cross-subsidy is severed — either by FTC-ordered structural separation or by some other forcing event. The graph captures every mechanism that depends on it, but does not model what Amazon Logistics looks like if it has to cover its own costs. That is the question none of the current data resolves.