Amazon Didn't Build a Delivery Network to Save Christmas. It Built Its Next AWS.

On Christmas morning 2013, Amazon did something a retailer almost never does: it blamed someone else, by name, in an email to its own customers. The someone was UPS, which along with FedEx had failed to deliver millions of packages in time for the holiday after air volumes blew past network capacity.² The story that hardened into legend is tidy — Amazon got burned, swore never again, and went off to build its own trucks and planes. It is a good story. It is also the wrong one, and the gap between the legend and the documents is exactly where the strategy lives.

The official story is that the 2013 Christmas debacle made Amazon build a delivery network. The truer story is that Amazon was already drawing one up that same year, and the debacle just handed it a reason it could say out loud.

Bloomberg later obtained an internal 2013 proposal, reported in 2016, that staff had nicknamed 'Operation Dragon Boat.' It recommended expanding Amazon's fulfillment operation into a global, end-to-end delivery network — first mile to last mile — and it had already reached senior management.¹ Read the dates side by side and the causation flips. The crisis didn't originate the plan. The plan was on the table, and the crisis was the proof-point that made the case unanswerable.

Why a carrier you don't own can never be fast enough

Here is the mechanism the Christmas story obscures. Amazon's entire promise — Prime, two days shrinking to one shrinking to same-day — depends on the speed and reliability of the last mile. But the last mile was owned by companies whose incentives weren't Amazon's. A third-party carrier optimizes its whole book of business: it smooths volume across thousands of shippers, it caps capacity to protect margin, and it treats peak season as a problem to ration rather than a promise to keep. Amazon's own 2014 10-K names the dependency plainly, flagging reliance on third-party carriers and holiday-season shipping margin as material risks to the business.³ When the thing your brand is built on sits inside another company's cost-benefit math, you do not control your own promise. You rent it — and at peak, the landlord rations it.

Amazon's science blog says the quiet part directly: before it built its own network, every package was completed by partners like USPS and UPS, and it was the rising customer demand for faster delivery that drove the creation of its Delivery Service Partner network.⁴ Shipping was also Amazon's single largest expense, so the carriers sat astride both the customer experience and the cost line at once.⁵ That is the textbook trigger for vertical integration: when an outside supplier controls a capability central to your strategy and your biggest cost, you stop buying the capability and start building it.

The asset that quietly became a product

This is where the story stops being about defense and starts rhyming with AWS. Amazon built server farms for itself, discovered the capacity was a sellable product, and turned a cost center into one of the most profitable businesses on earth. The logistics network is running the same play one beat behind. By 2023 the parcel count had tripled from roughly 2 billion in 2019 to 5.9 billion, enough to pass both UPS and FedEx by volume.⁶ A network at that scale isn't a delivery department anymore. It's infrastructure — and idle infrastructure begs to be rented.

In May 2026 the turn completed. Amazon launched Amazon Supply Chain Services, opening its freight, distribution, fulfillment, and parcel network to outside businesses of every kind — not just marketplace sellers — with early customers including Procter & Gamble, 3M, Lands' End, and American Eagle Outfitters. The market understood instantly what it meant: UPS shares fell nearly 10% and FedEx more than 9% on the news.⁷ The Christmas-revenge framing misses the whole arc. Amazon didn't build a fleet to never depend on a carrier again. It built a carrier, then opened the doors.

If the network is so good, why did FedEx come back?

The honest counter is that Amazon's own network clearly doesn't make third parties obsolete — Amazon keeps using them. It ended ground delivery with FedEx in 2019, the supposed declaration of independence, then in early 2025 quietly signed a new multi-year deal with FedEx to handle residential delivery of select large packages.⁸ If owning the last mile were a clean win, you wouldn't re-hire the company you publicly dumped. And there's the revenue gap that won't go away: despite moving more parcels than anyone, Amazon's logistics revenue was about $28.6 billion in 2023 against UPS's $68.9 billion and FedEx's $63.2 billion — less than half of either.⁶ By the metric that pays the bills, Amazon is still fourth.

Both objections are real, and both actually confirm the thesis rather than dent it. The FedEx rehire shows the network was never about replacement — it was about control of the promise. Amazon now uses partners where they're cheaper and its own trucks where speed is sacred, which is precisely the leverage you gain when you can build the capability instead of only buying it. And the revenue gap is the tell that this is infrastructure, not a shipping company: Amazon moves the most packages and earns the least from logistics because most of those packages are its own. The revenue line was always going to lag the asset. AWS looked like a cost center too — right up until it was the profit engine. We also can't confirm Dragon Boat was ever formally adopted; Bloomberg reviewed a proposal to senior management, not a board-stamped plan.¹ But the proposal hardly matters now. The network it described exists, and as of 2026 it sells to the public.

The legend says a bad Christmas taught Amazon a lesson. The documents say Amazon had already decided, and the bad Christmas just gave the decision a face. What it actually built wasn't revenge on UPS — it was the same trick it had already run on the data center: take the thing you depend on, the thing another company was rationing back to you, and turn it into something you own outright. The network was never the point. Owning the promise was — and once you own it at the scale of five billion packages, the promise is too valuable to keep to yourself.