Most of the AI buzz is about chatbots, models, and the firms making them. Most of the money is going somewhere else.
The firms powering AI are signing 10 and 20-year deals. They're posting backlogs at record highs.
The list is broad. It includes power firms, gas turbine makers, and nuclear firms.
Add fuel cell makers and transformer suppliers.
Investors who only watch the chip and software side are looking at half the board.
McKinsey thinks meeting data center demand through 2030 will need roughly $500 billion in new gear. That covers power plants, grid work, and water cooling.
That money lands with power makers and gear makers. Not Nvidia.
Some of the receipts are public:
These are not guesses. They are signed deals.
Cloud giants are building AI supply now.
AI data centers run hot and run all day. Bain thinks U.S. power firms need to add 100 to 200 gigawatts of new power by 2030.
That's the same as 100 to 200 large nuclear plants. All built in just 5 years.
That's not happening through the existing grid. Lawrence Berkeley National Lab data shows the average new project now waits about 5 years to plug into the grid.
So cloud giants are building power on-site. Turbines, fuel cells, and small reactors are getting bought in bulk.
The buyers are Amazon, Microsoft, Oracle, and Google direct.
The buildout under the AI boom is real, signed, and already in motion.
For investors, the power layer is where the AI boom shows up first. Chips and software get all the press.
But the firms making turbines, running nuclear plants, and selling fuel cells are winning. Their backlogs stretch into 2029.
That doesn't mean the chip story is over. It means there's a second leg to the AI trade.
It gets less press but just as much cash.
Several names are tied to this story. The list includes GE Vernova (GEV), Bloom Energy (BE), Constellation Energy (CEG), Vistra (VST), and Talen Energy (TLN).
Each has a piece of the buildout.
The IEA thinks the U.S. will lead global data center demand growth this decade. Worldwide, data center power use is set to nearly double by 2030.
The cash flowing into U.S. power gear is one slice of that.
The chatbots get the headlines, but the power plants get the contracts.
For long-term investors, the question isn't whether AI is real. It's where to find the cleanest play on the buildout.
The chip names trade at high price tags. They have big expectations baked in.
The power names trade at lower price tags. They get less press too.
That gap is what makes this story worth watching now.
