Amazon Web Services

• 1. Introduction

  ›

  Amazon Web Services began not as a product, but as a side effect of Amazon trying to fix itself. By the early 2000s, Amazon's own engineering teams were spending 70% of their time on what insiders called "undifferentiated heavy-lifting" - the grinding infrastructure work that kept the lights on but produced nothing customers could see or feel. The question the company's leaders kept asking was simple: what would happen if engineers could spend that time building things people actually wanted? The answer they stumbled toward would become the dominant force in cloud computing, eventually holding 31% of the global cloud infrastructure market as of early 2023. This is the story of how a frustration inside one retailer's IT department turned into a platform that spy agencies, rocket programs, and streaming giants now depend on - and what happens when that platform goes down, gets breached, or ends up in the middle of geopolitical conflict.

• 2. The Internal Frustration That Started It All

  ›

  Matt Round, an engineering leader at Amazon in the early 2000s, put together a set of recommendations that would quietly reshape the entire company. His proposals included maximizing autonomy for engineering teams, adopting REST architecture, standardizing infrastructure, and removing bureaucratic gate-keepers from decision-making. He also called for engineers to spend more of their time actually building software rather than fighting internal systems. Amazon acted on these ideas, creating a shared IT platform so its own divisions could stop sinking time into plumbing and start focusing on customers.

  Tom Killalea, Amazon's first CISO, had already been running the company's data centers in what he described as a "fast, reliable, cheap" way, migrating services to commodity Linux hardware and relying on open-source software to handle peak traffic, especially during the holiday season. When Amazon built Merchant.com, an e-commerce platform letting third-party retailers stand up their own web stores, it pushed the company toward a service-oriented architecture. That architectural shift, led by then-CTO Allan Vermeulen, set the stage for everything that followed.

  In July 2002, Amazon.com Web Services launched its first public web services, managed by Colin Bryar, opening the Amazon.com platform to all developers. More than one hundred applications were built on top of it by 2004. That level of unexpected developer interest caught Amazon by surprise and convinced the company that developers were, in the words of those involved, "hungry for more".

• 3. The Founding Vision

  ›

  Andy Jassy took over Colin Bryar's portfolio at the behest of Rick Dalzell during the summer of 2003, after Allan Vermeulen, Jeff Bezos's first choice for the role, declined the offer. Jassy then mapped out a vision for what he called an "Internet OS" - a set of foundational infrastructure primitives that would remove the key obstacles to shipping software faster. By fall of 2003, databases, storage, and compute had been identified as the first building blocks Amazon should bring to market.

  Jeff Barr, an early AWS employee, credits himself, Vermeulen, Jassy, Bezos, and a handful of others with originating the ideas that would become EC2, S3, and RDS. Jassy himself remembers the concept emerging from roughly a week of brainstorming with what he described as "ten of the best technology minds and ten of the best product management minds," working through about ten different internet applications to find the most primitive building blocks each one required. Werner Vogels pointed to Amazon's drive to make the cycle of "invent, launch, reinvent, relaunch, start over, rinse, repeat" as rapid as possible. That drive pushed Amazon to break up its organization into what Vogels called "two-pizza teams" and to decompose its applications into distributed systems - changes that ultimately opened the way for AWS and its stated mission "to expose all of the atomic-level pieces of the Amazon.com platform."

  Brewster Kahle, co-founder of Alexa Internet, which Amazon acquired in 1999, has said that his startup's compute infrastructure helped Amazon work through its big data problems and later informed the innovations at the core of AWS. Jassy assembled a founding team of 57 employees drawn from both engineering and business backgrounds, with a majority hired from outside Amazon. That group included Jeff Lawson, later CEO of Twilio; Adam Selipsky, later CEO of Tableau; and Mikhail Seregine, a co-founder at Outschool.

• 4. Building the First Services

  ›

  In late 2003, Chris Pinkham and Benjamin Black presented an internal paper describing a vision for Amazon's retail computing infrastructure that was completely standardized and fully automated, drawing on web services for functions like storage. Near the end of that paper, they raised the possibility of selling access to virtual servers as a product. Pinkham then worked with Willem van Biljon and lead developer Christopher Brown to build the Amazon EC2 service from a team based in Cape Town, South Africa.

  November 2004 saw AWS launch its first public infrastructure service, Simple Queue Service, or SQS. Then on the 14th of March 2006, AWS launched Amazon S3 cloud storage, followed by EC2 in August 2006. Pi Corporation, a startup co-founded by Paul Maritz, was the first beta user of EC2 outside Amazon; Microsoft was among EC2's first enterprise customers. Before that year was out, SmugMug, one of the earliest AWS adopters, credited S3 with saving the company around $400,000 in storage costs. Werner Vogels later noted that S3 launched in 2006 with just 8 microservices and had grown to over 300 microservices by 2022.

  In September 2007, AWS announced its first annual Start-up Challenge, a contest with prizes worth $100,000 aimed at entrepreneurs and software developers in the United States. The first edition drew participation from Justin.tv, which Amazon would later acquire in 2014. Ooyala, an online media company, took the prize. Additional services rolled out across this period included SimpleDB, Mechanical Turk, Elastic Beanstalk, Relational Database Service, DynamoDB, CloudWatch, Simple Workflow, CloudFront, and Availability Zones.

• 5. Growth Into a Global Business

  ›

  By November 2010, all of Amazon.com's own retail sites had migrated to AWS. Before 2012, AWS was treated as part of Amazon.com rather than a separate unit, so its financials did not appear on their own in Amazon's reports. In 2012, industry watchers estimated AWS revenue for the first time, putting it above $1.5 billion.

  On the 27th of November 2012, AWS hosted its first re:Invent conference in Las Vegas, chosen for its relatively cheaper connectivity to locations across the United States and around the world. The three-day event drew around 6,000 attendees from over 190 countries, with early registration priced at $1,099 per head. Andy Jassy and Werner Vogels delivered keynotes, and Jeff Bezos joined Vogels for a fireside chat. Reed Hastings, CEO of Netflix, took the stage with Jassy to announce plans to migrate 100% of Netflix's infrastructure to AWS.

  In January 2015, Amazon acquired Annapurna Labs, an Israel-based microelectronics company, for a reported $350-370 million. Three months later, in April 2015, Amazon reported that AWS was profitable, with $1.57 billion in first-quarter sales and $265 million in operating income. Bezos described it as a fast-growing $5 billion business. By October 2015, AWS's operating income had reached $521 million in a single quarter, with operating margins at 25%. Fourth-quarter 2015 revenue for AWS grew 69.5% year over year to $2.4 billion, producing a 28.5% operating margin and a $9.6 billion annualized run rate. Gartner estimated that year that AWS customers were deploying ten times more infrastructure on AWS than on the next fourteen providers combined.

  In early 2016, AWS became more profitable than Amazon's own North American retail business for the first time, with $2.57 billion in revenue and $604 million in net income in the first quarter alone - a 64% increase over the same period the year before. Jassy was promoted to CEO of the division. Annual revenue reached $17.46 billion in 2017 and climbed to $46 billion by the end of 2020.

• 6. Outages and Vulnerabilities

  ›

  On the 20th of April 2011, parts of the Elastic Block Store service became stuck and could not process read or write requests, triggering a major outage that took at least two days to fully resolve. The following year, on the 29th of June 2012, a severe storm hit Northern Virginia, where AWS hosts its largest cluster of data centers, knocking several sites offline.

  the 28th of February 2017 brought what became one of the most widely felt failures: a massive outage of S3 services in the Northern Virginia region. A majority of websites depending on S3 either stalled or hung entirely. The cause was human error during a debugging session - more server capacity was removed than intended, triggering a cascade of failures across connected systems. Amazon reported that AWS was fully online again within five hours, and no data was reported lost.

  In June 2022, it came to light that in 2019 Capital One had failed to secure its AWS resources properly. A former AWS employee exploited that gap, downloading personal information belonging to more than 100 million Capital One customers and using computing power to mine cryptocurrency. The employee was convicted of hacking into the company's cloud servers.

  In July 2025, a security researcher disclosed a supply-chain vulnerability in the Amazon Q Developer Extension for Visual Studio Code, specifically version 1.84. A malicious pull request had introduced a prompt instructing the AI assistant to delete local files and AWS resources. AWS acknowledged the issue in Security Bulletin AWS-2025-015 and released version 1.85 to remove the injected prompt. Critics, including Corey Quinn writing in Last Week in AWS, questioned the transparency of the response, noting the absence of a changelog entry, a CVE assignment, or a public statement beyond the bulletin itself.

• 7. Controversy and Geopolitics

  ›

  In October 2013, AWS was awarded a $600 million contract with the CIA. By 2019, more than 80% of Germany's listed DAX companies were reported to be AWS customers. UK spy agencies including GCHQ, MI5, and MI6, along with the Ministry of Defence, contracted AWS to host classified materials. The U.S. Navy announced in August 2019 that it had moved 72,000 users from six commands to an AWS cloud system as a first step toward migrating all its data and analytics to the cloud.

  Project Nimbus, a contract AWS entered into alongside Google to provide cloud services to the Israeli government, drew protests and condemnation from employees and shareholders at both companies. Critics raised concerns that the technology would enable surveillance of Palestinians, unlawful data collection, and the expansion of Israeli settlements. A government procurement document listing obligatory customers of Nimbus - including Israel Aerospace Industries and Rafael Advanced Defense Systems, two state-owned weapons manufacturers - was published in 2021 and updated periodically through October 2023.

  The Department of Homeland Security's use of software called ATLAS, running on Amazon Cloud, generated its own controversy. ATLAS scanned more than 16.5 million records belonging to naturalized Americans and flagged approximately 124,000 for manual review by USCIS officers regarding possible denaturalization. The algorithm and its selection criteria were kept secret. Amazon faced protests from both employees and outside activists over the program.

  In March and April of 2026, during the 2026 Iran War, Iranian drone and missile strikes damaged AWS data centers in the United Arab Emirates and Bahrain, described as the first deliberate military targeting of a commercial data center. Iran claimed the facilities hosted US military workloads. AWS advised customers to migrate workloads to other regions, and reports indicated that full recovery could take several months.

• 8. Scale, Environment, and What Comes Next

  ›

  AWS operates from 38 geographical regions with 120 Availability Zones, including nine regions in North America, nine in Europe, four in the Middle East, thirteen in the Asia-Pacific, and three in Australia and New Zealand. The global network includes over 700 points of presence worldwide. AWS has announced plans for 10 more Availability Zones and 3 more regions in Saudi Arabia, Chile, and a European sovereign cloud.

  In 2024, AWS paid $650 million for a data center co-located at the 2.5 GW Susquehanna nuclear power plant, drawing 300 MW of power. In 2025, AWS signed an $18 billion power purchase agreement with the same station, targeting 840-1,200 MW in 2029 and 1,680-1,920 MW in 2032, continuing through to 2042. Amazon as a whole became the largest corporate purchaser of renewable energy in the world in 2020 and has held that position since; by 2022-90% of Amazon's global operations, including data centers, ran on renewables.

  In September 2023, AWS announced it would serve as the primary cloud provider for AI startup Anthropic, with Amazon committing up to $4 billion in investment and taking a minority ownership position. The same announcement introduced Amazon Bedrock, a managed service providing access to foundation models from multiple AI companies through a single API. At the end of 2024, AWS announced Project Rainier, described as a massive one-of-a-kind machine "designed to usher in the next generation of AI." In July 2024, AWS won a $1.3 billion contract with the Australian government to build a top-secret cloud service in partnership with the Australian Signals Directorate, whose director general, Rachel Noble, said it would "provide a state-of-the-art collaborative space for our intelligence and defense community to store and access top-secret data."