A lot of the AI data centers being announced right now will never be built. On a recent episode of TechCrunch’s Equity podcast, hosts Rebecca Bellan and Tim De Chant talked with Brent Nelson, senior managing director of markets and strategy at Ascend Analytics, about this problem. As TechCrunch AI reports, the buildout already faces local opposition, permitting delays and a scramble for electricity. But before any of that comes into play, many of these projects may not be real in the first place.
This matters because billions of dollars in AI infrastructure investment depend on telling the realistic projects apart from the speculative ones.
🏗️ Why so many announcements won’t hold up
Announcing a data center is cheap. Building one isn’t. That gap leaves room for a lot of noise.
The forces behind it are fairly easy to see:
- Power shopping. Developers often ask for grid connections in several places at once and plan to keep only the best one. Every request looks like real demand on paper.
- Land and option plays. Some projects are really bets on real estate or on power access. The plan is to lock in a site and then sell it or partner later.
- Headline value. A huge capacity number makes good press for local officials, developers and tech companies, even when no financing is behind it.
- No end customer yet. A data center without a signed hyperscaler or AI lab as a tenant is still mostly a hope.
A firm like Ascend Analytics, which forecasts power markets, sits right where this problem lands. When utilities and grid planners can’t tell real load from wishful load, they risk overbuilding generation or charging the wrong people for it.
⚡ Why it matters now
What stands out here is the effect of all this inflated demand. Phantom projects don’t just disappear without a trace. They clog interconnection queues, which slows down the real projects waiting behind them. They also push utilities to plan new power plants and transmission lines, and ratepayers could end up paying for them.
They shape public opinion too. People in towns fighting a proposed facility are often reacting to a project that has no money or customer behind it. That means some of the fierce local opposition is aimed at buildings that would have fallen apart on their own.
And for investors, a pipeline built on announcements sends the wrong signal. If you size the AI infrastructure trade by headline gigawatts, you’ll probably overestimate what actually gets built.
🔮 The next 1-3 years
I expect a sorting process over the next few years rather than a sudden collapse. Here’s what that will probably look like:
- Utilities get stricter. More grid operators will ask for bigger deposits, financial proof and firm timelines before they reserve capacity for a large load. That alone will wash out a lot of the speculative projects.
- Power becomes the real moat. Projects with secured electricity, whether from the grid, on-site generation or long-term contracts, will pull far ahead of everything else.
- Consolidation. Well-funded players will buy stalled sites that already have permits and power rights. The developers who were flipping land will either sell or fade away.
- Pipeline numbers get revised down. Expect forecasts of data center electricity demand to come down as analysts discount projects without tenants or financing.
🧭 What practitioners and businesses should do
If you’re buying compute, planning capacity or investing in this space, treat announcements with some skepticism:
- Check the basics. Before you count a project as real, ask whether it has a tenant, financing, permits and a power agreement.
- Diversify capacity plans. Don’t bet your AI roadmap on a single facility that hasn’t broken ground yet.
- Watch utility filings, not press releases. Interconnection agreements and regulatory filings tell you much more than ribbon-cutting promises.
- Price in delays. Even real projects are running into permitting and grid bottlenecks, so build slack into your timelines.
The AI infrastructure boom is real, but it’s smaller and slower than the headline totals suggest. The companies that can tell signal from noise will make better bets as the hype gets checked against what actually gets built. The full conversation with Brent Nelson is available on TechCrunch’s Equity podcast.