Amazon Web Services’ data center empire is far larger and far more complex than the public has long assumed. Newly surfaced internal documents indicate the company now operates more than 900 data centers across over 50 countries, a footprint that dwarfs earlier industry estimates and highlights how deeply the AI era is reshaping global cloud infrastructure.
For years, AWS’ presence has been synonymous with its sprawling, purpose-built server farms in Virginia, Oregon and Ohio. But these flagship complexes, massive as they are, represent only part of the story. Amazon quietly extends its reach through an intricate network of colocation facilities, renting space from more than 180 partners worldwide. These so-called colos range from a single dedicated room to entire buildings housing AWS systems alongside other tenants.
According to the documents, published on the investigative site SourceMaterial, colocation sites accounted for roughly one-fifth of AWS’ computing power last year. The company also relied on more than 220 edge locations, typically embedded within telecom hubs, to help customers route data into Amazon’s core infrastructure with minimal latency. In effect, it’s a distributed global grid that blends owned facilities, long-term leases and a vast constellation of third-party sites.
Explosive Demand for AI
AWS is Amazon’s financial engine, delivering $33 billion in quarterly revenue, but its future hinges on meeting the explosive, unpredictable demand for AI training and inference. That growth has prompted Amazon to accelerate data center development at a pace rarely seen in the tech sector.
The company is also pushing aggressively into AI-specific infrastructure, especially in the United States. One bellwether is in Indiana, where Amazon is investing $15 billion in a rapid-build complex the company calls Project Rainier. The facility is intended to house over a million Tranium 2 chips and support workloads developed with AI partner Anthropic. Amazon says early buildings will come online within a year, a timeline that reflects the urgency of serving customers racing to deploy next-generation machine learning systems.
Environmental Concerns and Clean Energy
Environmental advocates warn that the accelerating construction of hyperscale and colo facilities threatens to strain regional power grids and extend reliance on fossil fuels. Internal estimates suggest AWS’ colocation sites consumed 7.8 million megawatt-hours of electricity in 2023 alone. In markets like Germany and Japan, some rented facilities are among the company’s most energy-intensive.
Amazon points to its multibillion-dollar renewable energy investments, noting it has been the world’s largest corporate purchaser of clean power for five consecutive years. Still, the tension is clear: AI computing is ravenous, and even the most efficient chips require vast amounts of electricity to deliver the performance customers now expect.
What emerges from the new data is a portrait of AWS’ operations that is both sprawling and remarkably fluid. The company may prefer to own its facilities (it controls about 24 million square feet and leases a similar amount) but its global footprint depends on a layered mix of property ownership, rental agreements, regional partnerships and regulatory accommodations, particularly in China.
A Size that Enables Dominance
The picture also helps explain why AWS continues to dominate cloud market share even as Microsoft and Google surge ahead with their own AI-focused investments. The network Amazon has built is not just immense, it is distributed in ways that allow the company to offer compute capacity almost anywhere a customer demands it.
In the AI era, the measure of a cloud provider is no longer just how much capacity it controls, but how quickly it can deploy more. By that metric, Amazon’s 900-plus data centers suggest a company determined to stay ahead of the curve, even as that curve grows steeper by the month.
