The AI infrastructure trade (aka the Power-Up America basket which we recommended one year ago before it soared into the stratosphere), had taken a back seat in recent weeks, with some marquee names such as a Vertiv, Contellation, Oklo and others, tumbling from record highs amid growing speculation that China’s DeepSeek – and other cheap LLM alternatives – will lead to far lower capex demands than what is currently projected.
But while the occasional hiccup is to be expected, the endgame for US infra/nuclear stocks looks (millions of degrees) bright. Consider Texas, where demand on the state power grid is expected to expand so immensely that it would take the equivalent of adding 30 nuclear plants’ worth of electricity by 2030 to meet the needs. That’s according to the Electric Reliability Council of Texas (ERCOT), which manages the grid.
The forecast is based on the addition of new data centers needed to power artificial intelligence. And it’s raising concerns about whether infrastructure in the state, which last week we said wants to be “king of nuclear power as the Next AI trade unfolds” – will be able to expand fast enough…. and at what cost.
Coming out of the pandemic, electricity demand on the Texas grid was already growing faster than anywhere else in the country, Bloomberg reports. And now that’s being supercharged by AI, with the state vying to become the data-center hub of the country, if not the world.
Individual projects are already starting to request 1 gigawatt of power and they pose new risks to maintaining a stable grid, said Agee Springer, Ercot’s senior manager of grid interconnections. A gigawatt is typically enough to power 250,000 homes in Texas. The data centers “present a reliability risk to the Ercot system,” said Springer, who spoke on a panel at Infocast’s ERCOT Market Summit in Austin this week.
“We’ve never existed in a place where large industrial loads can really impact the reliability of the grid, and now we are stepping into that world.”
Hellbound and Down 