아마존의 3번째 반복: 물류망 오픈이 말해주는 진짜 의미

The Third Time Amazon Did This: What “Amazon Supply Chain Services” Tells Us About What Amazon Is Gad Allon May 18, 2026 3 1 Share In May 2026, Amazon launched Amazon Supply Chain Services , opening its multimodal freight, automated warehousing, and last-mile parcel delivery network to any company that wants to use it. The first publicly named enterprise customers are Procter & Gamble, 3M, Lands’ End, and American Eagle Outfitters . Amazon spent $83 billion in capex in 2024 , more than seven times the combined capex of UPS, FedEx, and the U.S. Postal Service. I know. Not all of that is logistics-related. Read it next to the other parcel-economics story of 2026. On UPS’s January 27 earnings call , CFO Brian Dykes told investors UPS would eliminate up to 30,000 positions this year. In February, the company identified 22 union-staffed package facilities for closure, the next tranche of a plan to shut up to 200 facilities by 2030 . Across that same window, UPS continued to draw down its Amazon volumes toward a 50% cut by mid-year. The conventional read of Amazon Supply Chain Services (ASCS), “Amazon expands again,” misses what is happening. The story is not about logistics. It is about what Amazon does to its own cost centers, and what that has historically meant for everyone in the cost center next door. Amazon has built three businesses by accident. ASCS is the third. Subscribe The Pattern Amazon Has Now Run Three Times Start with Amazon Marketplace. In 1999, Amazon decided to allow third-party sellers to list products alongside its own retail catalog. The internal logic was simple: Amazon had built the front-end to be the best in the world at converting intent into transactions and was not using that capability to its full potential. By the end of 2024, 62% of paid units on Amazon.com were sold by third-party sellers. The marketplace that began as a way to fill the catalog became the catalog. Then AWS. In 2006, Amazon launched what is now AWS, exposing the internal compute, storage, and database services its retail group had built. The internal pitch was identical to Marketplace seven years earlier. The marginal cost of letting a startup, then a bank, then a federal agency run workloads on the same fabric was a tiny fraction of what those institutions would pay to do it themselves. AWS in Q1 2026 reported $37.6 billion in quarterly revenue , roughly $150 billion annualized, with $14.2 billion in segment operating income and an operating margin near 38%. Now ASCS. The internal logistics network Amazon spent 15 years and more than $200 billion in cumulative capex building, including more than 1,300 facilities, an in-house air cargo operation, and last-mile delivery in 99% of U.S. zip codes, is being turned into an external service. The first enterprise customers are running per-unit cost comparisons against the alternative and discovering that the math does not add up. The pattern is quite simple. Build infrastructure to serve internal operations at a scale no rational external buyer would justify. Optimize it to a level that drives marginal cost below the buyer’s internal alternative. When the surplus capacity is too large to write off, open the API and sell it . Each iteration, at launch, looked like a bad idea: Marketplace would cannibalize retail margin. AWS would expose infrastructure to competitors. ASCS puts Amazon’s logistics network in the hands of brands that compete with Amazon. Each time, Amazon decided the surplus capacity was worth more than the cannibalization risk. Share What Cost-Center Conversion Actually Requires Three things have to be true. Scale that an external buyer cannot match : AWS spends $15-20 billion per quarter on cloud capex and runs higher utilization than any single tenant can; ASCS operates 1,300+ facilities, an air fleet, and same-day reach in almost every zip code. Surplus capacity the parent cannot consume : AWS scaled externally because retail could not absorb the data-center commitments; ASCS launches now because retail growth has normalized, and the network was built for a holiday peak that has softened. Cultural willingness to sell to competitors : P&G makes Pampers; Amazon Basics makes diapers. Most companies will not put their freight on the rails of a direct retail competitor. Amazon will, because by 2026, it has done it twice already, and the strategic math has worked both times. The deeper question for any operating company is whether it has anything in its cost center that, run at an unreasonable scale and with the willingness to sell to a competitor, could attract external buyers. For most companies, the answer is no, because the cost center never reached the necessary scale. That is the real moat: not the surplus, but the decades-earlier willingness to build the cost center at the scale that eventually produced the surplus. The Square Root Law of Pooling If there is one piece of operations theory that explains why Amazon’s cost-curve gap is structurally durable, it is the square root law of inventory pooling, formalized by Eppen , now standard in any operations textbook (the cleanest derivation can be found in my colleagues Cachon & Terwiesch ’s book, Chapter 14 if you are really curious). The result is simple, and students who took Operations Strategy with me know that I have real joy teaching it. Take N facilities, each serving independent demand. The safety stock required scales linearly with N: it is proportional to N × σ. Pool them into one centralized network serving the same demand, and the safety stock drops to √N × σ. The savings ratio is 1 − 1/√N. Quadruple the network and save half the buffer. Multiply N by sixteen, save three-quarters. At N = 25, the pooled network needs roughly 80% less buffer to serve the same demand at the same service level. The same √N scaling shows up in last-mile routing. The optimal tour through n stops in a fixed service area scales as √n, so the cost per stop falls as 1/√n. The Amazon-UPS cost gap is not an operational gap. It is the gap between two points on the same curve. Let’s walk through the math (you can skip the next part, but if you do skip, are you even a real reader? Not judging. Just saying). Take the 2024 Pitney Bowes parcel volumes: Amazon: 6.3 billion; UPS: 4.7 billion; USPS: 6.9 billion. Index each carrier to USPS = 1.00: N_Amazon = 6.3 / 6.9 = 0.913, N_UPS = 4.7 / 6.9 = 0.681. Cost per parcel scales as 1/√N, so cost_Amazon = 1 / √0.913 = 1.046 and cost_UPS = 1 / √0.681 = 1.212 on the same index. The ratio is √(N_Amazon / N_UPS) = √(6.3 / 4.7) = √1.340 ≈ 1.158. UPS pays about 15.8% more per parcel than Amazon for the simple reason that Amazon’s network moves 1.34× the volume across roughly the same geographic footprint. That 15.8% is the structural floor. It does not depend on Amazon being better-run than UPS, only on Amazon being bigger. If UPS spent the next decade matching Amazon on every operational dimension, the floor would persist as long as Amazon’s volume grew faster, which it has every year since 2018. The same law generalizes across every Amazon platform business. AWS pools demand variability across millions of customer workloads. Marketplace pools SKU discovery across millions of sellers. Amazon Ads pools impressions across the entire retail surface and Prime Video. In each case, the marginal cost of serving an additional external customer falls along the √N curve while the marginal revenue is roughly linear. The gap is the operating margin. ASCS productizes these effects in physical logistics. Adding P&G’s freight to a network already moving 6.3 billion Amazon parcels does not require a new network; it only adds throughput to a cost curve that improves as √N. P&G pays Amazon for a margin uplift, largely already generated by existing at that scale. The Per-Package Math The cost gap is not subtle. Industry-wide, the Pitney Bowes 2024 Parcel Shipping Index puts average revenue per parcel at $9.09. The fully loaded cost figures below are M