Amazon launched its Simple Storage Service, known as S3, in March 2006 and its Elastic Compute Cloud, known as EC2, in public beta that August, letting any individual developer rent object storage and virtual servers by the hour from a company still best known at the time for selling books, diapers, and general merchandise online.

Startups could suddenly launch new products without purchasing physical servers outright or negotiating lengthy colocation contracts with traditional data-center operators. Product-experimentation cycles shortened dramatically as a direct result, and outright failure became genuinely cheap to absorb financially, provided a developer remembered to actually shut an idle instance down rather than leave it quietly accumulating hourly charges overnight.

The mechanism was operational abstraction: engineers increasingly thought purely in terms of managed services, geographic regions, and simple API calls, while Amazon's own internal infrastructure teams, overseen in AWS's early years by executive Andy Jassy, quietly handled the underlying physical server racks, electrical power redundancy, and automatic failover that used to require an expensive dedicated in-house operations staff at every company.

Amazon gained a business division that would eventually generate substantially more operating profit than its original core retail division; by the early 2020s, AWS had become the single largest contributor to Amazon's overall corporate profitability. Traditional enterprise hardware vendors and in-house corporate data-center teams lost significant ground steadily as computing workloads shifted en masse to Amazon, and later to competing platforms including Microsoft Azure, launched in 2010, and Google Cloud.

Retail-industry analysts in 2006 openly and publicly wondered why an online bookstore was investing so heavily in building server infrastructure at all. Contemporary coverage substantially underweighted how completely the coming waves of mobile apps, software-as-a-service businesses, and eventually large-scale AI model training would come to depend entirely on a small handful of hyperscale cloud providers and their specific regional data-center footprints.

The widespread October 2021 outage that simultaneously took down Facebook, Instagram, and WhatsApp for hours, and periodic AWS-specific regional outages — including a major December 2021 disruption originating in AWS's heavily used US-EAST-1 region that disrupted numerous unrelated consumer services — together demonstrate just how concentrated critical internet infrastructure has become around a genuinely small handful of dominant providers.

Amazon Web Services launched core services such as S3 (2006) and EC2, productizing the internal infrastructure Amazon had built to run its retail site. Startups could rent compute and storage by the hour instead of buying servers — collapsing capital expenditure into operating expense.

The mechanism was programmatic infrastructure: APIs, identity/access controls, and regional redundancy that made 'the cloud' a default architecture. Microsoft Azure and Google Cloud followed, but AWS's early lead set pricing expectations, talent markets, and the shared-responsibility security model enterprises still live inside.

Outages at a single hyperscaler began to register as public incidents affecting thousands of dependent companies — proof that cloud was shared fate, not just rented hardware. Resilience engineering and multi-cloud talk grew from that dependency, even when most firms remained practically single-cloud.

Enterprise IT departments that once measured success by data-center uptime began measuring it by cloud-bill predictability and IAM hygiene. The cultural shift — from owning machines to governing permissions — is as important as the technical one, and it still determines who gets hired into infrastructure roles.

S3's simple object API and EC2's rental VMs taught a generation of CTOs to treat infrastructure as code. Startups that would have needed a million dollars in servers launched on credit cards. Amazon's retail-scale operations knowledge became a product line that eventually rivaled retail in operating importance.

Shared-responsibility security models shifted breach narratives: configure your IAM wrong and the failure is yours even when the building is Amazon's. Multi-cloud strategies arose as political hedges more than clean technical architectures — dependency had become too obvious to ignore after high-profile regional outages.

Governments that once feared foreign clouds eventually bought the same APIs under compliance overlays. The strategic dependency is bipartisan in practice even when rhetoric differs. Cloud regions are now discussed like ports and pipelines: critical infrastructure with a monthly bill.

National policy debates over data-residency requirements, corporate multi-cloud redundancy strategy, and antitrust scrutiny of cloud-market concentration among regulators worldwide are all downstream consequences of a strategic decision, made with remarkably little public notice back in 2006, that computing itself should be sold and billed as a metered utility service rather than owned physical hardware.

Century Signals note: AWS launch materials for S3/EC2; contemporaneous developer and enterprise IT coverage; later analyses of hyperscaler market structure and outages. Editorial judgment about what still structures the present — not a comprehensive history.