AWS: How a Bookstore Became the Internet's Backbone
AWS：一家书店，如何成了互联网的骨干
Amazon Web Services is only about 15% of Amazon's revenue — and earns as much profit as the entire retail business, because cloud infrastructure runs on ~30% margins while selling books runs on ~1.5%. In its AWS deep-dive, Acquired debunks the famous "selling spare capacity" myth and tells the real story: AWS was the byproduct of fixing Amazon's own broken architecture, then realizing the infrastructure was a product. One of the great profit machines ever built. Figures as of the 2022 episode; not investment advice.
亚马逊云服务（AWS）只占亚马逊营收的 约15%——赚到的利润却抵得上整个零售业务，因为云基础设施跑的是约30%的利润率，而卖书跑的是约1.5%。在这期 AWS 深度解析里，Acquired 拆穿了那个著名的"卖闲置算力"神话，讲出了真实的故事：AWS 是亚马逊修自己那套坏掉的架构时的副产品，然后才意识到，这套基础设施本身就是一款产品。这是史上最伟大的印钞机之一。数据均为2022年那期口径；非投资建议。
The most profitable idea at Amazon
亚马逊最赚钱的那个点子
Start with the numbers, because they're the whole point. The thing most people think of as "Amazon's cloud side project" is actually where Amazon makes its money.
先从数字说起，因为它们就是全部的重点。大多数人当成"亚马逊那个云端副业"的东西，其实正是亚马逊赚钱的地方。
1.5% on lemonade, 30% on the cups
卖柠檬水赚1.5%，卖杯子赚30%
The hosts open with a parable. Run a lemonade stand and you sell the best lemonade for the lowest price — but it costs you 98.5 cents to make each dollar's worth, a punishing 1.5% margin. Then you notice you've gotten really good at the stuff behind the stand — the cups, the ice, the supply chain — and you can sell that to other businesses at a 30% margin. Those numbers aren't made up: ~1.5% is Amazon retail, ~30% is AWS.
主持人用一则寓言开场。你摆个柠檬水摊，以最低的价卖最好的柠檬水——可每一块钱的货，要花掉你98.5美分的成本，一道折磨人的1.5%利润率。然后你发现，自己对摊子背后那些东西——杯子、冰、供应链——已经很在行了，而这些，你可以以30%的利润率卖给别的生意。这两个数字不是瞎编的：约1.5%是亚马逊零售，约30%是 AWS。
15% of the revenue, ~all of the profit
15%的营收，约全部的利润
Here's the stat that reframes the company. AWS is only about 15% the size of Amazon's giant retail business by revenue — yet its operating income equals or exceeds the entire retail operation, essentially every year since Amazon started breaking out AWS financials in 2015. The cloud business everyone treats as a footnote to the store is, in profit terms, the main event. The internet runs on it.
下面这个数字，重新定义了这家公司。按营收算，AWS 只有亚马逊那庞大零售业务的约15%——可它的经营利润，却抵得上、甚至超过整个零售盘子，自2015年亚马逊开始单列 AWS 财务起，基本年年如此。这桩被所有人当成商城脚注的云业务，论利润，才是正戏。互联网就跑在它上面。
Why a worse business was the genius move
为什么一桩"更差"的生意，才是神来之笔
The cleverest part is counterintuitive. The incumbent vendors AWS disrupted — the companies selling servers, software and storage to enterprises — were earning ~70% margins. AWS came in happy to make just 30%. That's a worse business than the incumbents had, but a vastly better one than retail's 1.5% — and at 30% you can undercut a 70%-margin incumbent into oblivion while still printing money. Amazon disrupted a great business by being content with a merely good one.
最聪明的地方，反直觉。被 AWS 颠覆的那些老牌厂商——向企业卖服务器、软件、存储的公司——赚的是约70%的利润率。AWS 一来，乐意只赚30%。这是一桩比老牌厂商更差的生意，却是一桩远胜零售1.5%的生意——而在30%上，你能把一个70%利润率的在位者一刀杀到出局，自己还照样印钞。亚马逊靠满足于一桩"还不错"的生意，颠覆了一桩"很棒"的生意。
The myth: it wasn't spare capacity
那个神话：它不是闲置算力
Before the real story, the hosts kill the famous one — the origin tale almost everyone repeats, and which didn't actually happen.
在讲真实的故事之前，主持人先杀掉那个著名的——几乎人人都在转述、却根本没发生过的起源传说。
"They sold off leftover holiday servers"
"他们把节假日剩下的服务器卖了"
The popular origin myth goes like this: Amazon built huge server capacity for the holiday shopping rush, sat idle the rest of the year, and cleverly rented out the excess — and that became AWS. It's tidy, it's intuitive, and it's repeated in countless write-ups. The hosts found not one but four competing origin stories for AWS floating around — and this most famous one is the wrong one.
流行的起源神话是这样的：亚马逊为节假日购物高峰建起庞大的服务器产能，一年里其余时候闲着，于是聪明地把多出来的租了出去——这就成了 AWS。它干净、直觉、被无数文章一再转述。主持人发现，市面上流传的 AWS 起源故事不止一个，而是足足四个互相矛盾的版本——而这个最有名的，恰恰是错的那个。
Amazon's own CTO calls it a myth
亚马逊自己的CTO说，这是个神话
Amazon CTO Werner Vogels put it plainly in 2011: "The excess capacity story is a myth. It was never a matter of selling excess capacity." In reality Amazon doesn't carry much meaningful spare capacity outside the fourth quarter — a real cloud business would have outrun any leftover servers within months of launch. AWS wasn't a clever way to monetize idle hardware. Its origin is more interesting, and more instructive, than that.
亚马逊CTO维尔纳·沃格尔斯（Werner Vogels）在2011年说得明明白白："闲置算力的故事是个神话。它从来都不是卖多余算力这回事。"现实是，除了第四季度，亚马逊根本没有多少有意义的闲置产能——一桩真正的云生意，上线几个月内就会把任何剩余服务器消耗殆尽。AWS 不是把闲置硬件变现的小聪明。它的起源，比那有趣得多，也更有启发。
The myth hides the real lesson
神话藏住了真正的教训
The spare-capacity story is appealing because it makes AWS sound like a lucky accident — found money. The truth is the opposite: AWS came from years of hard internal engineering and a deliberate realization about what kind of company Amazon had become. Getting the origin right matters, because the real story is the actual playbook — and it's one other companies can learn from, where "we had extra servers" is not.
闲置算力的故事之所以诱人，是因为它把 AWS 说成一次幸运的意外——白捡的钱。真相恰恰相反：AWS 来自多年艰苦的内部工程，和一个关于"亚马逊已经变成一家什么样的公司"的有意识的醒悟。把起源弄对，是要紧的，因为真实的故事，才是真正能学的打法——它是别家公司能借鉴的，而"我们有多余的服务器"不是。
The real origin: breaking the monolith
真实的起源：劈开那块巨石
AWS wasn't dreamed up as a product. It grew out of a crisis in Amazon's own software — and a now-legendary memo.
AWS 并不是被当作一款产品构想出来的。它长自亚马逊自己软件里的一场危机——以及一份如今已成传奇的备忘录。
One giant codebase, grinding to a halt
一块巨大的代码库，慢得快停摆
By the early 2000s Amazon's software had become a single, sprawling monolith — one enormous codebase that every team had to work within. As the company grew, that monolith made launching anything new agonizingly slow: every change risked breaking everything else, and teams were tangled together. Amazon's biggest internal problem wasn't selling books; it was that its own engineering couldn't move fast enough to keep up with its ambitions.
到2000年代初，亚马逊的软件已经变成了一整块蔓生的巨石——一个庞大的代码库，每个团队都得在里头干活。随着公司变大，这块巨石让发布任何新东西都慢得折磨人：每改一处，都可能弄坏别处，团队彼此缠成一团。亚马逊最大的内部问题，不是卖书；而是它自己的工程，跟不上它的野心。
Bezos's memo that rewired the company
贝索斯那份重新布线整家公司的备忘录
The fix was architectural — and, famously, decreed. In a memo now part of tech lore, Jeff Bezos mandated that every team expose its data and functionality only through hard, well-defined service interfaces — APIs — with no exceptions and no back doors. The monolith was broken into independent services that could each move on their own. It was painful, but it forced Amazon to build its entire business as a set of clean, composable internal services.
解法是架构层面的——而且，出了名地，是用命令下达的。在一份如今已是科技界传奇的备忘录里，杰夫·贝索斯下令：每个团队都只能通过硬性的、定义清晰的服务接口——也就是 API——来暴露自己的数据和功能，没有例外，也没有后门。那块巨石被劈成一个个能各自前进的独立服务。这很痛苦，却逼着亚马逊把它整个业务，搭成一套干净、可拼装的内部服务。
We're actually an infrastructure company
我们其实是一家基础设施公司
Here's the hidden pattern. In solving its own scaling problem, Amazon discovered it had gotten genuinely excellent at building reliable, scalable infrastructure — storage, compute, databases — exposed as clean services. And if those services were good enough to run Amazon, they were good enough to sell. AWS wasn't conceived as a product and then built; it was infrastructure built for Amazon that turned out to be a product. The most important business in the company was hiding inside its own plumbing.
这里藏着那条暗线。在解决自己扩展难题的过程中，亚马逊发现，自己已经真正擅长搭建可靠、可扩展的基础设施了——存储、算力、数据库——而且都以干净的服务暴露出来。既然这些服务好到能跑亚马逊，那它们就好到能卖。AWS 不是先被构想成产品再造出来的；它是为亚马逊而造、却原来是一款产品的基础设施。这家公司里最重要的业务，一直藏在它自己的水管里。
Primitives
原语积木
Turning internal services into a product took a specific philosophy — and a specific person, who would one day run all of Amazon.
把内部服务变成产品，靠的是一套特定的哲学——和一个特定的人，他日后将执掌整个亚马逊。
Andy Jassy, Bezos's shadow
安迪·贾西，贝索斯的影子
The person who built AWS was Andy Jassy — a Harvard MBA and serial product launcher who had served as Jeff Bezos's "shadow," the technical assistant who sat in every meeting and absorbed how Bezos thought. He took that worldview into a new organization and turned a set of internal tools into a business. Two decades later, when Bezos stepped back, Jassy became CEO of all of Amazon — a measure of how central AWS turned out to be.
造出 AWS 的人，是安迪·贾西（Andy Jassy）——一位哈佛 MBA、连续做产品发布的人，曾当过杰夫·贝索斯的"影子"，那个坐在每一场会议里、吸收着贝索斯如何思考的技术助理。他把那套世界观带进一个新组织，把一堆内部工具变成了一桩生意。二十年后，当贝索斯退到幕后，贾西成了整个亚马逊的 CEO——这本身，就衡量出 AWS 后来有多核心。
Ship primitives, not finished solutions
交付原语，而不是成品方案
The core philosophy was "primitives." Instead of shipping a polished, all-in-one product, AWS would ship the smallest reusable building blocks — raw storage, raw compute — and let developers combine them however they wanted. It's the opposite of a packaged solution, and it's why AWS could expand endlessly: every new primitive multiplies what people can build, and the customers, not Amazon, figure out the use cases. The Lego strategy beats the toy-set strategy.
核心哲学，是"原语"（primitives）。AWS 不交付一个打磨光亮、一体化的产品，而是交付最小的、可复用的积木块——裸存储、裸算力——任开发者随心去拼。这是"打包方案"的反面，也是 AWS 能无止境扩张的原因：每多一块原语，就把人们能搭的东西翻上一番，而想出用途的，是客户，不是亚马逊。"乐高积木"的打法，胜过"成套玩具"的打法。
Kill the "undifferentiated heavy lifting"
干掉那些"无差别的苦力活"
The mission had a memorable name: remove the "undifferentiated heavy lifting" of IT — all the boring, identical work every company does to run servers and data centers, none of which makes any of them special. Wrap it in Amazon's working-backwards culture (write the press release and customer FAQ first, before building anything; no PowerPoint), and the pitch was irresistible: stop racking your own servers, and rent ours by the hour.
这项使命，有个让人记得住的名字：消除 IT 里那些"无差别的苦力活"——每家公司为运维服务器和数据中心而做的、无聊又雷同的活儿，没有一样能让谁变得与众不同。再裹上亚马逊"倒着干"的文化（动手造之前，先把新闻稿和客户 FAQ 写出来；不许用 PowerPoint），这套说辞就让人无法拒绝：别再自己上架服务器了，按小时租我们的。
S3 and EC2 (2006)
S3 与 EC2（2006）
In 2006 the abstract idea became two products that changed how software gets built — and quietly disproved the spare-capacity myth along the way.
2006年，那个抽象的点子，变成了两款改变软件建造方式的产品——而且，一路上还悄悄证伪了那个闲置算力的神话。
Infinite storage you rent by the gigabyte
按GB租用的无限存储
The first true AWS service was S3 — Simple Storage Service — launched in March 2006. It gave any developer effectively unlimited, reliable storage they could pay for by what they used, with no servers to buy or manage. Notably, S3 launched before EC2 existed, which is itself a quiet rebuttal to the spare-capacity myth: this was a deliberately built product line, shipped one primitive at a time, not a fire sale of leftover hardware.
第一款真正的 AWS 服务，是 S3——简单存储服务（Simple Storage Service）——于2006年3月上线。它给任何开发者实质上无限、可靠的存储，用多少付多少，不必买、也不必运维服务器。值得注意的是，S3 上线时，EC2 还不存在——这本身就是对闲置算力神话的一记悄声反驳：这是一条被有意打造、一块原语一块原语地交付出来的产品线，而不是一场剩余硬件的甩卖。
Rent a server by the hour
按小时租一台服务器
A few months later came EC2 — Elastic Compute Cloud — in beta in August 2006. If S3 was storage on tap, EC2 was raw computing on tap: spin up a virtual server in minutes, run your app, shut it down when you're done, and pay only for the hours you used. For a startup, it eliminated the single biggest upfront cost and risk of building anything — no more buying and racking machines before you had a single customer.
几个月后，EC2——弹性计算云（Elastic Compute Cloud）——于2006年8月以测试版登场。如果说 S3 是"打开水龙头就有"的存储，EC2 就是"打开水龙头就有"的裸算力：几分钟内开起一台虚拟服务器，跑你的应用，用完关掉，只为用过的小时数付费。对一家初创来说，它抹掉了建任何东西时最大的那笔前期成本和风险——再不用在还没有一个客户时，就先去买、去上架机器。
The model the whole industry now runs on
如今整个行业赖以运转的模式
Together, S3 and EC2 turned computing into a metered utility — like electricity. You stopped owning the power plant and started paying for what you drew from the wall. It is hard to overstate how much this unlocked: a generation of startups could be born for almost nothing, scale instantly when they took off, and never touch a screwdriver. Every cloud business since has been built on the template AWS shipped in 2006.
S3 和 EC2 合在一起，把计算变成了一种按表计费的公用事业——像电一样。你不再拥有发电厂，而是为从插座里取走的那点付费。这解锁的东西，怎么说都不为过：一整代初创，几乎零成本就能诞生，一旦起飞就能瞬间扩容，且永远不必碰一把螺丝刀。此后的每一桩云生意，都建在 AWS 于2006年交付的那套模板之上。
The profit machine
印钞机
What made AWS world-changing for developers also made it one of the great businesses in history — and it changed what Amazon itself could do.
让 AWS 对开发者而言改变世界的那一点，也让它成了史上最伟大的生意之一——而它，也改变了亚马逊自己能做什么。
Immediately, enormously cash-generative
即刻、海量地产生现金
Unlike most of Amazon's bets, AWS gushed cash almost immediately. A ~30%-margin business at the scale of the internet throws off staggering operating income — and it does so reliably, on long-lived customer relationships. The contrast with the incumbents it disrupted is stark: they clung to fat ~70% margins and were undercut; AWS took a smaller slice of a much bigger, faster-growing pie, and won.
和亚马逊大多数的赌注不同，AWS 几乎一上来就喷涌现金。一桩约30%利润率、却跑在互联网这个量级上的生意，吐出的经营利润大得惊人——而且靠着长久的客户关系，吐得很稳。它和被它颠覆的那些在位者，对比鲜明：后者死抱着约70%的肥利润率、被一刀切穿；而 AWS 在一块大得多、长得快得多的饼里，取了更薄的一片，赢了。
AWS pays the bills so retail can keep reinvesting
AWS 埋单，零售才能不断再投入
AWS's profits don't just sit there — they power Amazon's whole flywheel. The famously low-margin retail business can keep plowing money into prices, selection, logistics and new bets precisely because AWS generates the operating income that funds it. In effect, the boring infrastructure business subsidizes the consumer business that everyone associates with the Amazon name. Without AWS, the Amazon you know would have to look very different.
AWS 的利润不是搁着不动——它们驱动着亚马逊整个飞轮。那桩出了名低利润的零售业务，之所以能不停往价格、选品、物流和新赌注里砸钱，恰恰是因为 AWS 产出了为它供血的经营利润。实际上，是那桩无趣的基础设施生意，在补贴那桩人人一提亚马逊就想到的消费业务。没有 AWS，你所熟悉的那个亚马逊，将不得不长成很不一样的模样。
Microsoft and Google had to chase it
微软和谷歌只能追着它跑
Such a good business doesn't go uncontested. Microsoft eventually made a massive bet on Azure, and Google built Google Cloud — both racing to catch the leader. But AWS's years-long head start, and the fact that it had already become the default, meant rivals were chasing from well behind. As of this 2022 telling, AWS was still clearly the leader, with Azure the strongest challenger. (Standings as of the episode.)
这么好的生意，不会无人来争。微软最终在 Azure 上下了重注，谷歌做了谷歌云——两家都在追赶这个领头羊。但 AWS 领先数年的身位，加上它早已成了默认选项，意味着对手是从老远的后头追起。在2022年这期节目里，AWS 仍是明明白白的领头羊，Azure 是最强的挑战者。（排名为当期口径。）
The moat nobody can climb
无人能攀的护城河
Why has the lead held? The hosts run AWS through Hamilton Helmer's "7 Powers" — it's analysis, not gospel, but it names what keeps the business on top.
这份领先，为何守得住？主持人用汉密尔顿·赫尔默的"七种力量"把 AWS 过了一遍——这是分析，不是圣经，但它点出了，是什么把这桩生意稳稳压在顶上。
Incumbents couldn't bear to copy it
在位者下不了狠心去抄它
AWS's first power was counter-positioning. The incumbent IT vendors were addicted to ~70% margins; to compete with AWS they'd have had to blow up their own pricing and embrace a 30%-margin model their shareholders would revolt against. Amazon, which thinks in decades and was used to thin margins anyway, was perfectly happy to. The incumbents literally could not follow without destroying themselves — the cleanest kind of competitive advantage.
AWS 的第一种力量，是反向定位。那些老牌 IT 厂商，对约70%的利润率上了瘾；要和 AWS 竞争，它们就得炸掉自己的定价，去拥抱一套股东们会群起反对的30%利润率模式。而亚马逊，本就以十年为单位思考、本就习惯薄利，完全乐意。在位者要跟，就非得毁掉自己不可——这是最干净的那种竞争优势。
"The greatest scale-economies business of all time"
"史上最强的规模经济生意"
The deepest power is scale. The hosts call AWS maybe the greatest scale-economies business ever: it spreads enormous fixed costs — data centers, custom chips, networking, R&D — across the biggest customer base in the industry. That means AWS can run each unit cheaper than anyone else, charge less, and still earn more. Every new customer makes the cost structure better, which lets it cut prices, which wins more customers. The flywheel feeds itself.
最深的一种力量，是规模。主持人称 AWS 也许是史上最强的规模经济生意：它把庞大的固定成本——数据中心、定制芯片、网络、研发——摊到行业里最大的客户基底上。这意味着 AWS 能把每一单位跑得比谁都便宜、收得更少、却还赚得更多。每多一个客户，成本结构就更好一分，于是能降价，于是赢来更多客户。飞轮，自己喂自己。
Switching costs, brand, and the reframe
转换成本、品牌，与重新定义
Two more powers seal it. Switching costs: once a company's data and applications live on AWS, moving them is brutal, expensive and risky — so customers stay. And branding: AWS became the safe default — "nobody gets fired for choosing AWS." Stack them up and the reframe lands cleanly: the unglamorous infrastructure Amazon built to run its own bookstore became the backbone of the internet and the most important, most profitable business it has. (Analysis as of September 2022 — not investment advice.)
还有两种力量把它封死。转换成本：一旦一家公司的数据和应用住进了 AWS，搬出来既痛苦、又昂贵、又有风险——于是客户留下了。还有品牌：AWS 成了那个安全的默认选项——"选 AWS，没人会因此被炒"。把它们叠起来，重新定义就干净利落地落地了：亚马逊为运营自家书店而搭的那套不起眼的基础设施，成了互联网的骨干，也成了它手里最重要、最赚钱的生意。（分析为2022年9月口径——非投资建议。）