The Power Plant Is Becoming a Product

We've spent ten years arguing about fuels. The argument that matters now is about time.

Something is changing in how the world gets electricity, and it's not the usual fight over fuels. One level below that argument, the architecture itself is shifting, toward generation built next to the customer who'll use it. 

A New Category: On-Site, Always-On Power 

The industry has a name for this now: compact on-site firm power, generation that runs around the clock, at the facility it serves, never touching a transmission line. 

Remote military bases, mining sites, offshore rigs, and islands have worked this way for decades. What's new is the scale: it's becoming a serious architecture for serious loads.

• American data center developers have announced roughly 101 gigawatts of behind-the-meter natural gas capacity: plants built to sit on campus and sell power to nobody else. 

• More than half of hyperscalers and colocation providers expect to run fully on-site-powered campuses by 2030. 

• One AI company already runs something like a gigawatt and a half of its own turbines outside Memphis. 

  
  

It's Not a Data Center Story. It's a Speed Story. 

It's tempting to read this as a story about AI, or gas. We think it's really about speed. 

Here's why: not ideology, not even price. Interconnection queues in major markets now run for years, and utilities' delivery estimates run well behind what developers plan around. 

When time is the binding constraint, customers stop asking which electron is cheapest and start asking which power they can get. Gas has won so far because its turbines are off-the-shelf, buyable, and truckable while a grid connection is still queued. 

The Twist: Gas Is Now the Slow Option 

Here's what makes this moment interesting: the incumbent of the speed category is now failing on speed. 

Only three manufacturers build heavy-duty gas turbines at scale, and all three are sold out for years. The largest has a backlog of over a hundred gigawatts against annual production of roughly twenty, already booking slots for 2031. 

Gas won the opening round because it could ship first. Increasingly, it can't ship at all. 

Who's Racing to Fill the Gap 

The candidates are lining up. 

In fission, a wave of microreactor and small-modular developers is designing for the on-site model from day one, with factory-built units for remote industry and military bases. The US Air Force has already committed to a dedicated microreactor in Alaska. 

Fusion is splitting along the same line. The best-funded programs are building utility-scale plants in Virginia, Tennessee, and the UK. Others, nt-tao among them, are building machines an order of magnitude smaller, meant to be manufactured in numbers. 

Different physics, different fuel cycles, same underlying bet: stop treating a power station as a bespoke, decade-long project, and start treating it as a product. 

A Different Way to Pay for Power 

That bet, if it pays off, quietly rewrites how energy gets funded. 

A conventional power station runs on project finance: a decade-long undertaking where risk sits in one place, backed by sovereign guarantees and offtake contracts signed years before concrete pours. 

A manufactured power unit is a different financial animal. Risk spreads across an order book instead of concentrating in one site, and capacity comes in increments a customer can budget for. 

Project finance built the twentieth-century grid and will keep building the big plants. Manufacturing finance is what built cars, aircraft, and solar panels. Which kind of money wins will matter as much as any reactor design. 

We've Been Here Before 

Energy has done this dance before, just pointed the other way. The grid was built on the conviction that generation belongs far away and at maximum scale. For a hundred years that was right, because central plants were the fastest way to add supply. 

The physics hasn't changed. The bottleneck has. Once permitting and interconnection become the slowest links, distance stops being an efficiency and becomes a tax. 

Bigger Than AI 

Filing this under artificial intelligence would be a mistake. Data centers are the loudest early customer, but the same logic applies to any load that's large, constant, and unable to wait: 

• Desalination and water treatment plants run continuously, often in water-stressed regions where grids are already strained. 

• Steelmakers and chemical plants need industrial heat that transmission lines can't deliver anyway. 

• Military bases treat grid independence as a security requirement, not a line item. 

• Ships can’t even connect to a grid if it was available now. They need a solution for a clean scalable propulsion power. 

  
  

The data center boom is this category's first market. We doubt it'll be its largest. 

The Catch 

None of this is a finished argument for the distributed model. Thousands of scattered generation units raise licensing, security, and oversight questions that large, watched stations don't. 

And the grid isn't going anywhere. Central generation will carry most of the world's demand for decades. On-site power is becoming a layer of the system, like personal computers alongside the mainframe, not its replacement. 

The Speed Crown Is Vacant 

Layers can end up defining an era, though. Customers driving today's demand growth are choosing power that behaves like a product: sited on their land, sized to their load, on their schedule, not the grid planner's. 

Gas won the opening round because it could ship first. Now even gas can't ship fast enough. The speed crown in energy's fastest-growing category is, for the moment, vacant, and it'll go to whoever can manufacture firm power fastest. 

We've spent ten years arguing about fuels. The argument that matters now is about time.