The AI infrastructure gold rush has spawned an entire category of vaporware so common it has its own name: bragawatts.

The Summary

The Signal

The distance between announcement and construction in AI infrastructure has become a chasm. Bernstein analyzed over 7,000 data center projects across North America and concluded that most will never break ground. The culprits? Newcomers with zero construction experience making billion-dollar promises. Kevin O'Leary, the Shark Tank investor, earned specific mention as a "bragawatt" producer. The analysts were blunt: "Press releases are cheap…concrete is not."

This matters because capital is getting allocated based on fantasy. When 65% of announced capacity belongs in the vaporware category, investors, utilities, and local governments are making decisions on false signals. Data Center Map tracks only facilities with at least 10 cabinets of initial capacity, a modest threshold that still reveals stark geographic concentration. Virginia dominates, but even states with minimal current infrastructure are fielding hyperscale proposals.

"Press releases are cheap…concrete is not."

The real infrastructure buildout is happening, just not where the press releases say it is. Nokia's Q3 profit beat came directly from data center networking equipment sales, proving someone is buying physical gear for actual facilities. That's a better signal than any announcement. Follow the equipment sales, not the headlines.

Meanwhile, Texas has become ground zero for the environmental cost of what does get built. The state could see air pollution return to levels not seen since the 2000s if current and proposed data centers all come online. The determining factor: how these facilities get their power. Gas turbines versus renewable energy will decide whether Texas trades AI leadership for respiratory problems.

The power equation is forcing unconventional solutions. Tech companies and developers are building their own generation capacity because the grid can't keep up. But the rush to secure electricity is prioritizing speed over sustainability. When Data Center Map compiled its October 2026 snapshot of US operations, it showed a system already strained. Add another 170 gigawatts of real capacity, and the grid math gets brutal.

Key contrasts across the buildout:

  • Announced capacity: 500 gigawatts
  • Likely actual capacity: 170 gigawatts
  • Current market leader: Virginia (operational centers)
  • Highest environmental risk: Texas (air quality regression)

The Implication

If you're tracking AI infrastructure, ignore the announcements. Watch equipment orders, utility capacity agreements, and actual construction permits. Nokia's earnings beat tells you more about the real buildout than any press conference. For states courting data centers, the Texas situation is a warning. The economic development pitch looks great until you're explaining why air quality just regressed three decades. The facilities that do get built will need power solutions that don't require choosing between computation and breathable air.

The bragawatt problem also signals a maturation point. When an industry spawns its own term for vaporware, the easy money phase is ending. Real infrastructure takes expertise, capital discipline, and years of patient work. The gap between announcement and reality is a filter. What emerges on the other side will be the actual foundation of the agent economy, not the fantasy version.

Sources

Business Insider Tech | Business Insider Tech | Bloomberg Tech | The Verge AI