Data, But Off Center

Everyone is apparently worried that we are overbuilding data centers, and that we are essentially investing on very shaky speculative demand. Stocks are dipping uncomfortably in the tech world, and AI is being discussed with even more vitriol. The thing is, we're talking about the kind of companies that often build with planned obsolescence — and the kind that have been doing heavy R&D with full knowledge of creeping global competitors. So I don't really see this as just a hype-gone-wrong tech story.

Because you only have to shift your perspective a little to see that this is actually a bank story.

Since Covid, commercial real estate loans have been sitting on bank books at valuations that don't reflect reality. Office vacancy never quite recovered, and those numbers need somewhere to hide. We're building the future of intelligence is a better pitch than we're trying to keep regional bank portfolios from imploding. But the underlying mechanics are the same. The data center is the instrument — a way to extract developer fees from new projects and cover losses on failing properties.

Giants in private equity have legacy real estate portfolios — offices and older retail — losing 20% to 40% of their original value. If a PE firm can't deliver high returns, institutional investors pull their money. To keep investors happy, PE firms need an asset class generating massive, double-digit returns to average out the losses from their office portfolios. Right now there is only one real estate asset on earth capable of generating that kind of explosive growth: data centers.

And considering the biggest investors in these private equity companies are public pension funds — well, you can see why the folks at the top aren't governing against them.

So it doesn't really matter if we actually need all those data centers for their functionality. We need them as an economic tool.

That said, people aren't wrong to point out that this boom feels off. We only have to look across the Pacific. In 2023 and 2024, over 500 data center projects were announced across China, from Inner Mongolia to Guangdong. They also publicly assumed great demand. It did not work out that way. Those data centers are now distressed assets running at 20–30% capacity, and investors are quietly unloading them at below-market rates.

Quite bad, right?

To be fair, China built it all really fast — in hot places with shrinking water supplies. So perhaps a financial loss now, but in the long term it would have caused more bleeding. That's why, despite Xi Jinping himself having to ask "Do all provinces in the country have to develop industries in these directions?" — because every town was trying to capitalize on the AI market — China didn't give up. They pivoted, relocating data centers to cooler, water-rich regions.

They went and built an undersea data center just off the coast of Shanghai. The idea was originally Microsoft's — "Project Natick" — but they abandoned it. China went ahead and built it anyway: an underwater data center cooled by seawater and powered by offshore wind.

Pretty cool.

Meanwhile, the US has been approving new gas plants to power data centers for the next thirty years. There's a data center going up in Childress, Texas right now. Population 6,000. Hot and dry, with a fragile grid, far from the fiber connections and technical talent that the latest AI actually needs.

So when Wall Street panics about tech stocks dipping and asks if we’re overbuilding for AI, they’re asking the wrong question. We didn't build these for AI. We built them to save commercial real estate portfolios.

When the AI hype inevitably cools, the tech giants will just write it off as an R&D expense. The private equity funds will have collected their fees and restructured. The banks will have survived.

All Childress gets is the building.