A six-month-old company run by a former coal miner just became AI's riskiest bet, planning to go from $1.5 billion to $30 billion in revenue in two years by building what could be America's largest data center.

The Summary

  • Nscale acquired American Intelligence & Power, a seven-week-old company developing a massive West Virginia data center complex, after raising $2 billion at a $14.6 billion valuation
  • The company projects 20x revenue growth in two years (from $1.5B to $30B) and has signed an LOI to provide 1.35 gigawatts of computing power to Microsoft
  • Nvidia invested, Sheryl Sandberg and Nick Clegg joined the board, and the entire play rides on West Virginia power infrastructure that doesn't exist yet

The Signal

The AI infrastructure land grab just entered its absurdist phase. Nscale appears to have been independent for roughly six months. Its founder came from coal mining, not tech. The company it acquired, AIP, was literally created seven weeks ago through a merger between an investment firm and a West Virginia site developer. And yet here we are: $14.6 billion valuation, Microsoft as anchor tenant, Nvidia money, Meta executives on the board.

The West Virginia angle matters more than it looks. This isn't about cheap land or tax breaks. It's about power. The site promises 2 gigawatts of computing power online by 2028, which would make it America's largest data center complex. That's not a data center. That's a small city's worth of electricity dedicated to running AI training clusters. The coal miner background suddenly makes sense. This is energy infrastructure play dressed up as a tech company.

The revenue projection tells you everything about the current moment. Going from $1.5 billion to $30 billion in two years isn't a business plan, it's a belief system. It assumes Microsoft and others will keep buying compute capacity at whatever price, that West Virginia can actually deliver gigawatt-scale power on an aggressive timeline, and that the AI training arms race won't pause for breath. Goldman Sachs underwrote these numbers, which means someone believes them enough to stake reputation on it.

What's fascinating is the inversion. A year ago, CoreWeave looked risky for betting everything on GPU infrastructure. Now you've got SoftBank borrowing aggressively, Oracle making massive capex commitments, and a six-month-old company planning the largest data center in America. The risk ladder keeps extending. Each new entrant has to be bigger, faster, more levered than the last to get attention and capital.

The Implication

Watch the power deals, not the press releases. If Nscale can actually bring 2 gigawatts online in West Virginia by 2028, it proves that energy infrastructure, not chip supply or software talent, is the real constraint on AI scale. If it can't, you'll see the first major write-down in the AI infrastructure buildout and a reality check on these valuations.

For anyone working in AI: your training runs are increasingly dependent on whether a former coal miner can build power infrastructure in Appalachia faster than utilities have ever moved. The agent economy's foundation is being laid by people who understand electricity, not algorithms.

Source: The Information