The Suboptimal Present of Data Centers...and the Bright New Future
People are figuring out what hyperscale datacenter campuses really are: land and resource grabs with compute attached.
The Utah fight around Kevin O'Leary is just the latest flare-up. A proposed AI campus in Box Elder County, called Stratos, has been described as roughly 40,000 acres with a full-buildout energy appetite of about 9 gigawatts, more than twice Utah’s current average consumption. Residents showed up furious, not because they hate technology, but because they can do the math. They see the water risk, the air risk, the heat, the noise, the tax breaks, the rushed approvals, and the promise that all this destruction is somehow “innovation.”
That story is going to repeat.
For a decade, the public mostly treated datacenters like magic internet warehouses. They were out of sight, abstract, and wrapped in green language. But AI has ripped the mask off. These are not just buildings. They are industrial-scale energy consumers, and in many places they are being dropped onto communities that did not ask to become sacrifice zones for AI margins.
The International Energy Agency says global datacenter electricity consumption is projected to roughly double from 2025 to 2030, while AI-focused datacenter consumption grows even faster. In the United States, the Department of Energy has pointed to estimates that datacenters could consume up to 9% of annual U.S. electricity generation by 2030. That is no longer a niche load. That is a grid-planning problem, a ratepayer problem, a water problem, and a political problem, and the public is catching on.
Hyperscale campuses also have a geography problem. Bigger is not automatically better. Putting gigantic compute warehouses in remote locations often means bigger transmission needs, more local environmental pressure, longer permitting fights, and worse latency for the workloads that actually need to happen close to users, markets, devices, and machines.
The next era of compute will look different.
It should be smaller, faster, more distributed, and more honest about power. Ultra-low latency compute paired with behind-the-meter power changes the conversation. The operator has to own the economics, reliability, and footprint instead of quietly socializing the burden through utilities, ratepayers, and local politics.
That is the direction we are building toward at Vailinor: low-latency compute, integrated power strategy, and infrastructure that does not require pretending a 40,000-acre megaproject is the only way forward.
This should not be a left-versus-right issue, or an AI-versus-no-AI issue. It is an infrastructure design issue. People want better technology, but they are increasingly unwilling to accept blank-check extraction in the name of progress.
This backlash we are seeing is not anti-compute. It is anti-stupid compute.
The future belongs to builders who understand that power, latency, land, water, and community consent are not side quests. They are the whole product.
Originally published on X and LinkedIn by Jason Brink, May 7, 2026.