The most valuable input to AI right now isn’t a smarter model. It’s a gas turbine. According to Hacker News, which surfaced a widely discussed analysis titled “Silicon Valley Has Lost Its Biggest Advantage,” the AI boom has quietly turned into a construction boom, and the winners include companies that make yellow trucks and giant engines.
Here’s the detail that reframes everything: Caterpillar’s stock has more than doubled in the past year, making it worth six times as much as Nike. Not because of bulldozers. Because of its massive gas-powered engines, which now help power the nation’s data-center build-out. What stands out here is the shift in what actually constrains AI. It’s not bits. It’s atoms and electrons.
The Economics Just Flipped
Silicon Valley built its empire on a simple advantage: software costs almost nothing to make and prints money at scale. Data centers break that model. Running one is closer to running a steel mill than shipping a smartphone app.
The numbers are staggering:
- Amazon, Google, Microsoft, Meta, and Oracle are on track to spend more on data centers this year than they earn from operations.
- Combined capex since ChatGPT launched has already blown past half a trillion dollars.
- J.P. Morgan projects AI investment could top $1.1 trillion next year.
To keep paying, these giants will likely take on debt. That’s a real change. Companies that mint tens of billions in profit are now borrowing to fund the physical layer of AI.
Power Is the New Moat
Five years ago, a standard data center needed 10 or 50 megawatts. This week, Meta said it’s more than doubling its flagship AI facility to a peak of five gigawatts. A proposed site in Utah would demand nine. That’s several large cities’ worth of electricity pumping through a few warehouses full of chips.
Grids can’t keep up, so the fastest path online is to build your own power plant. That’s why Caterpillar and every major turbine maker are drowning in back orders. In May, Elon Musk reportedly spent at least $1 billion on an energy company with a fleet of combustion turbines, likely to feed Grok.
Listen to how the executives talk now. Sam Altman says his biggest constraint is “electrons.” Nvidia’s Jensen Huang says a shortage of skilled “plumbers and electricians” will be the boom’s hardest bottleneck. As the article puts it, you cannot vibe-code a data center.
Why It Matters Now
Two pressures are converging. First, the strangest alliances in tech are forming out of pure necessity. Anthropic, the industry’s most safety-focused firm, is reportedly paying Musk $1 billion a month to rent a data center, months after Musk called the company “evil.” When rivals who trade public insults still sign billion-dollar deals, that tells you compute scarcity beats grudges.
Second, the backlash is arriving. New York just became the first state to place a moratorium on new hyperscale data-center construction. Communities are pushing back on the noise, water use, and power draw. Regulation is no longer a future risk. It’s here.
What to Do About It
For practitioners and businesses, the takeaways are practical:
- Treat compute as a supply-chain problem, not a cloud line item. Availability and location now matter as much as price. Lock in capacity early.
- Watch the power map. The next AI hubs will follow cheap, abundant energy, not just talent. Regions with spare grid capacity gain leverage.
- Model efficiency is now a cost strategy. Smaller, cheaper models that do the same job reduce your exposure to the electricity crunch.
- Expect permitting friction. If your roadmap assumes unlimited data-center growth, build in delays from local opposition.
Looking one to three years out, AI increasingly resembles drilling an oil well. Copper, silicon, labor, and electricity go in. Tokens come out. The State Department has even announced an international pact to shore up the AI supply chain, called Pax Silica.
The race for smarter models will keep making headlines. But the quieter race, for turbines, transformers, copper, and electricians, is the one that decides who can actually build. Full details are at the original source.