Amazon Web Services added more data center capacity in 2025 than any other company on the planet, including 1.2GW in the last quarter alone. The company disclosed the figures in its annual sustainability report, and they dwarf anything the industry has seen in a single year.
From October 2024 to October 2025, AWS brought 3.8GW of new data center power online. CEO Andy Jassy said the company has doubled its power capacity since 2022 and plans to double again by 2027. That pace is unmatched.
The numbers matter because they show how far ahead AWS has pulled in the cloud infrastructure race. While competitors scramble to build out their own AI-ready facilities, Amazon is already operating at a scale that would take others years to match. The gap is widening, not closing. It's a story that extends well beyond one company's earnings report, with implications for anyone tracking cloud and edge computing capacity.
The Capacity Gap
AWS operated 56.8 million square feet of data center and office space in 2025, a 16 percent increase over 2024, according to its 10-K filing. About half of that space is owned, the rest leased. For context, the entire United Kingdom had roughly 1.6GW of data center capacity in 2024.
Microsoft, AWS's closest rival, added 2GW of capacity in its 2025 fiscal year and another 1GW in the quarter that overlapped with AWS's Q4. Google doesn't disclose comparable figures. Synergy Research puts AWS's global cloud market share at 28 percent, with Microsoft at 21 percent and Google at 14 percent.
The capital spending gap is just as stark. Amazon has committed tens of billions of dollars to data center construction, and the 3.8GW added in a single year suggests the company is deploying that capital faster than anyone expected. The Data Center Dynamics analysis notes that Amazon's Q4 alone accounted for more capacity than most countries have in total.
Virginia, the world's largest data center market, has about 8GW of operational capacity with another 4GW under construction, according to JLL Research. AWS's 3.8GW addition in one year is approaching half of that entire pipeline. No other company is adding capacity at that rate.
Amazon's data centers consumed 2.5 billion gallons of water in 2025, the company revealed. That figure is likely to grow as more facilities come online and AI workloads demand more cooling infrastructure. The environmental cost of this expansion is becoming harder to ignore.
Inference Changes Everything
The driver behind this buildout isn't traditional cloud workloads. It's AI inference.
McKinsey projects that inference workloads will surpass training to become the dominant AI compute type by 2030, representing more than half of all AI compute and roughly 30 to 40 percent of total data center demand. The firm expects US data center capacity to grow from about 30GW in 2025 to 90GW or more by 2030, a compound annual growth rate of 22 percent.
That growth is larger than the entire power demand of California today.
Training and inference have radically different infrastructure needs. Training clusters require 100 to 200 or more kilowatts per rack, specialized interconnects, and advanced liquid cooling. Inference workloads run at 30 to 150 kW per rack, closer to enhanced cloud compute than high-performance computing. The result is two different build archetypes, each determining where and how hyperscalers invest.
This split is reshaping where companies build. Training facilities are going up in remote, power-rich areas where grid capacity and land are available. Inference facilities are popping up near metro areas where latency matters. Hyperscalers are no longer building one type of data center. They're building two, and the investment decisions behind each are fundamentally different.
The McKinsey analysis argues that hyperscalers are capturing about 70 percent of forecast capacity through owned or leased options. Their infrastructure decisions will define how the broader data center ecosystem develops over the next decade.
Hardware advances are lowering the amount of energy needed per compute. Software optimizations, including shifts toward smaller, fine-tuned models, are reducing runtime requirements. These efficiency gains could moderate growth toward compound annual growth rates of 4 to 7 percent, particularly because permitting timelines and regional grid constraints cap how quickly new capacity can come online.
The Subsea Squeeze
Cloud capacity isn't just about data centers. It's also about the cables that connect them.
Around 40 submarine cable systems are expected to enter service in 2026, a record. That's up from about 15 systems launched in 2025. Industry estimates suggest more than $13 billion worth of new cable systems will come online between 2025 and 2027, the largest expansion since the dot-com era.
Much of that capacity is reserved for hyperscale companies' own traffic. Amazon, Microsoft, and Google are investing heavily in private cable networks to move data between their cloud regions. The Subsea Cables report notes that these deployments are unlikely to significantly impact the open bandwidth market, which means the public internet may not benefit as much as the cloud providers themselves.
This concentration matters because it gives hyperscalers control over the entire stack. When a company owns both the data centers and the cables connecting them, it can optimize performance in ways that smaller competitors can't replicate. It's another dimension of the same capacity race that's playing out on land.
What Happens Next
The constraint isn't demand. It's power.
McKinsey's research shows that permitting timelines, regulatory limits, and regional grid constraints are the main bottlenecks preventing even faster expansion. Utilities across the US are struggling to connect new data centers to the grid fast enough. Some projects are facing multi-year waits just for interconnection studies, and that's before a single shovel hits the ground.
Amazon's data centers consumed 2.5 billion gallons of water in 2025. As facilities get bigger and more power-hungry, their environmental footprint grows. Liquid cooling is becoming standard for AI workloads, but water usage remains a flashpoint in communities where data centers are being built. The tension between AI growth and resource consumption is only going to get worse.
The companies that solve the power problem first will define the next phase of cloud computing. AWS is betting that its scale and capital advantage will let it stay ahead. The 3.8GW added in 2025 suggests that bet is paying off. Whether it can sustain that pace through 2027 is the question that will determine the cloud industry's next chapter.
Data center construction is no longer just a technology story. It's an energy story, a regulatory story, and a geopolitical story. The companies that navigate all three will dominate the next decade of cloud computing. The ones that don't will be left competing for scraps in a market that rewards scale above everything else.