Written by Jeffrey A. Newman, Esq. MBA with the assistance of Claude.ai
I said it before and I’ll say it again: we’re building data centers in the wrong place. Cooling is one of the highest costs of running them, so stop building them where it’s hot. Move the computers to the cold. Keep the windows open. Lease federal land along Alaska’s coast, pull in near-freezing ocean water, and let nature do the cooling we’re paying compressors to do. The water off Alaska sits between 32 and 45 degrees year-round. That’s free. We’re choosing to ignore it.
But there’s a second problem, and it’s worse. It’s the clock and it is ticking.
Data centers already use about 4.4% of the country’s electricity, and that could reach 12% by 2028. This is a real load on a real grid. And here’s the part nobody planned for: we can build the building in a year, but we can’t power it for five to seven. Construction isn’t the slow step anymore. A gigawatt facility now goes up in twelve months. Then it sits dark, waiting for a grid connection, for half a decade. Google says transmission delays are the single biggest threat to its expansion. The utilities say the wait is even longer than the developers think, and it’s getting worse. There is talk about building micro generators but the timing is slow and that is not going to supplant the grid drain.
China v U.S.
We are in a race against time and against China. One we cannot lose. Charles Jennings detailed just how committed China’s government is to AI development in his book, AI: Rise of the Lightspeed Learners. In my opinion, this book should be read by everyone. China is underwriting the development of AI in its Universities and grade schools. It is sponsoring and funding thousands of AI research companies, and AI is in critical development within China’s military-industrial establishment. This is a national security issue. Compute is power now, in the literal and the strategic sense. The country that wins the development phase wins. The country that waits and allows things to unfold slowly loses. We are in a race, and we are letting finished, paid-for capacity sit idle for years over a permitting bottleneck we refuse to fix.
The data center dilemma, controlled in this country by a small group of business leaders, has dragged on for years — and, as things stand, it places us at the will of China’s leadership. Our economy will suffer, as will our citizens, if we do not prevail in the AI race. As we speak, idle facilities drive up the cost of computing, and that cost flows into everything: medicine, defense, jobs. To dodge the delay, developers are bolting on their own gas generators just to turn the lights on. We’re burning fuel we didn’t need to burn because the grid can’t keep up. What we need is another Government-sponsored Manhattan project that integrates cooperation between AI scientists and corporate leaders. How else are we going to resolve the binding constraint resulting from utilities, grid interconnection, transformers, and power procurement not keeping pace? Goldman Sachs said in February 2025 that data center power demand could rise 50% by 2027 and 165% by 2030, while noting that supply growth has already been constrained by utilities’ inability to expand transmission fast enough, as well as by permitting and supply-chain bottlenecks. Goldman also warned that transmission projects can take years to permit and build, creating bottlenecks for data center growth
A June 2026 Wall Street Journal report, summarizing a JPMorgan analysis, said more than 60% of expected data center capacity due by 2027 was not yet underway, and another 7% was delayed. Other June 2026 reporting said power availability, not capital, is the primary constraint on development
Deloitte notes that if U.S. infrastructure cannot keep up, “power and grid capacity constraints could hamstring AI advancement,” with AI data center growth increasingly limited by interconnection queues, substation capacity, and transmission build‑out rather than capital availability. Brookfield similarly observes that deployment timelines are now dictated “less by hardware readiness and more by grid interconnection queues, substation capacity, and utility coordination cycles,” meaning projects slip not because GPUs are late, but because power is. Brookfield and other infrastructure investors note that “modern AI infrastructure is following electricity, not population centers,” with hyperscale campuses gravitating to regions with cheap power, available land, and existing transmission headroom. That dynamic favors cross‑border and cross‑state siting strategies and encourages vertically integrated structures (e.g., co‑locating data centers with new gas, nuclear, or renewables plus storage) to escape local grid bottlenecks.
Heat
The physics of heat transfer is currently one of the greatest impediments to data center development. With the rise of intensive artificial intelligence (AI) workloads, the computing capacity per rack has skyrocketed from the legacy 5–10 kW to 30–100 kW+, making it physically impossible to cool these facilities using conventional air conditioning. Hyperscale data centers rely heavily on air-based or water-based cooling. They dump massive thermal exhaust into the environment, raising local temperatures by up to 4°F within half a mile. This “Data Heat Island Effect” is sparking severe public pushback and environmental pushback. Hardware Failure: Because nearly every watt a processor consumes becomes waste heat, chips will quickly malfunction or degrade if temperatures are not strictly managed. To bypass these thermal limits, the industry is transitioning to specialized liquid and direct-to-chip cooling setups. However, retrofitting legacy sites or designing entirely new plants for these multi-megawatt cooling demands requires massive capital, complex modular engineering, and huge amounts of freshwater
Alaska
Alaska has free cooling and untapped power — wind, tidal, geothermal — plus the option to build on-site generation from day one and skip the interconnection line entirely. A site like that doesn’t wait five years for someone else’s wire. It builds its own runway. And if you’re already pumping seawater to cool the racks, you’re one step from desalination — fresh water for a parched West, shipped south while the data moves through undersea fiber.
This needs to be treated like the security priority it is — federal land, fast-tracked, built where the cold is free and the power doesn’t take half a decade to reach.
The build takes a year. To this date no unified decision as to where data center should be built, based on the needs of the nation and the speed of completion has even been considered as a national security issue. Yet it is a national security decision. We don’t have years to waste.
Jeffrey Newman, JD, MBA, a former prosecutor, is a lawyer whose firm represents whistleblowers. Jeff@Jeffnewmanlaw.com
