Nvidia grabs 20% of $10B firm to lock down 15 gigawatts for AI's next phase

The energy math behind the AI boom is getting a hard number attached to it: Nvidia is spending up to $3 billion for a roughly 20 percent stake in Lancium, the power infrastructure developer valued at around $10 billion, according to The Information. This isn’t a chip deal or a software play. It’s a raw power play. Lancium already has four gigawatts under contract in Texas and is scouting sites to add another 15 gigawatts—the kind of capacity you need when a single next-generation AI training cluster can draw the load of a small city.

The deal puts a spotlight on an uncomfortable reality the industry is finally admitting out loud. Amazon is backing a separate 7.65-gigawatt gas-fired plant in Pecos County, Texas, a facility the New York Times reports could belch up to 33 million tons of CO₂ annually and earn the title of dirtiest power plant in the country. Amazon spokesperson Margaret Callahan told the Times the company’s climate goals are intact but conceded that AI data centers are making them harder to reach. That’s corporate speak for “we’re building a giant fossil fuel plant because the timeline for renewables doesn’t match the urgency of the AI race.”

Climate scientist Zeke Hausfather has a comparison that makes the scale visceral: a heavy user of agentic AI already burns through roughly as much energy per year as running a clothes dryer, and that figure isn’t flattening out. Environmental groups are now raising alarms about health and pollution impacts around the Pecos County site, which sets up a conflict that’s only going to spread as more of these facilities break ground.

What’s worth watching here is how this reshapes the competitive landscape beyond the obvious environmental fight. Nvidia securing a direct stake in power infrastructure—not just signing a power purchase agreement—signals that the bottleneck for AI scale is shifting from silicon to electrons. If access to gigawatt-scale power becomes the new moat, the companies writing equity checks for generators today are the ones who get to train the GPT-5s and Gemini Ultras of 2027 while everyone else waits in an interconnection queue.