AWS's $200 Billion Year Is Quietly Squeezing Out Tier-2 Hosts

A single AWS campus in New Carlisle, Indiana, that stretches across 30 buildings on 1,200 acres now qualifies for 2.2 gigawatts of electricity — enough to power about 1.6 million average American homes. Locals call it Project Rainier, and Amazon calls it just another site.

With about 900 similar projects in the works, and a budget for 2026 capital expenditures of $200 billion, that isn’t a typographical error or a predictive statement. It’s a line item in Amazon’s publicly disclosed guidance for 2026 capital expenditures, representing the largest private investment ever committed to a single year of infrastructure development.

For comparison, the figure approaches New Zealand’s annual GDP and equals roughly 200 Hoover Dams in 2025 dollars. It marks the moment at which a hyperscaler ceased to be a tenant in the global markets for power and real estate and became the markets themselves.

This is where the attention of hosts should be focused. The size of the investment has drawn intensive coverage by the financial press over several months. But little attention has been paid to the ways in which AWS is tying together land, power, water, and silicon to build a vertically integrated back end for the AI economy.

Project Rainier is a 2.2-gigawatt, 30-building, 1,200-acre campus based entirely on AWS’s internally developed Trainium2 chips. But it isn’t unique. AWS has committed to multigigawatt projects in Mississippi, Pennsylvania, Ohio, Virginia, and Oregon with construction lead times of 18 to 36 months and orders for transformers placed well in advance of project completion.

Most importantly, AWS is no longer merely buying power off the grid. Instead, it’s negotiating long-term agreements for the purchase of power directly with utilities and is heavily involved in the design of substations, systems for treatment of water, and roads to support the loads served by each of these projects.

From the perspective of the host, this represents complete integration of a power plant. Look at the resulting geography. AWS used to chase the fiber backbones at Ashburn, San Jose, and Hillsboro. The 2026 buildout is happening elsewhere at New Carlisle, Indiana; Madison County, Mississippi; Salem Township, Pennsylvania; Hermiston, Oregon, and many other locations. This migration map represents an inventory of areas previously more familiar with soybean production and subsidies for rural broadband than with hyperscale tenants.

This is not an accident. AWS and, frankly, the entire hyperscale industry to a much greater extent are pursuing locations where open capacity for power still exists. Data centers will account for 46% of Virginia’s electricity consumption by 2030, according to Dominion Energy’s own numbers. There’s no additional capacity for substation construction.

Consequently, we’re following AWS with additional capacity for substation construction into counties whose zoning boards had previously been primarily concerned with grain elevators.

The resulting experience isn’t one of large financial commitments. Rather, it represents complete vertical integration. AWS buys the land, provides financing for the substation, arranges for operation of the gas peaker or nuclear unit, and provides for fiber connectivity. Finally, we install computing equipment on top of all this. All resources are owned and subject to a 20-year amortization schedule. Other providers of managed hosting or colocation services receive no competitive advantage in the market for land. Rather, they receive no competitive advantage at all for the entire stack of services.

This represents a completely new competitive situation for providers of managed hosting or colocation services attempting to obtain a lease for 10 MW of capacity in any market previously identified as a candidate for future development. Now, these providers compete with the largest buyer of infrastructure in the history of mankind for the same allocations of capacity.

This isn’t an exaggeration, but simply a description of the results of the December 2025 capacity auction of the PJM Interconnection. The auction resulted in a shortfall of 6,625 MW and an additional cost of approximately $14.7 billion to electricity consumers in the region.

The prospects for future development of power as a valuable currency for providers of managed hosting or colocation services are therefore extremely grim. The entity that owns the majority of this valuable resource is no longer a utility. It’s Amazon.

Consequently, we’re witnessing a near-complete absence of announcements regarding geographic expansion by providers of managed hosting or colocation services. Instead, we’re observing an extremely high degree of activity with regard to development of strategic alliances and implementation of operational changes for edge computing.

Global spending on data center construction rose 32% year over year in 2025, but so did resistance from local communities. In a research project, Data Center Watch tracked 142 organized activist groups in 24 states, $18 billion in U.S. data center projects that have been blocked, and another $46 billion in project development that has been delayed.

Maine Governor Janet Mills vetoed a hyperscale construction moratorium bill in late April, but only after the bill cleared the legislature. Pennsylvania lawmakers are weighing a three-year moratorium of their own.

The political leverage generated by these actions isn’t ideological, but reflects the cost of electricity. When ratepayers in Northern Virginia receive their monthly utility bills and see the line item for AWS capital expenditures, the project is no longer an industry story. Rather, it becomes a story for the kitchen table that ultimately influences action by state legislatures.

AWS expenditures in 2026 won’t be a one-year event. Instead, they represent a continuing operating rate of $200 billion. Additional expenditures of approximately $200 billion will be incurred for 2027, and equipment orders for diesel backup generators at Caterpillar and Cummins are fully committed through 2028.

As a consequence, the pressure on markets for Tier-2 power won’t be a temporary fluctuation of weather conditions. Rather, it represents a fundamental change in climate. The hosting companies that survive the next three years will be those that identify areas where AWS won’t expand, and achieve an earlier market position.