Vinod Khosla: AI’s energy crisis has a fix — and it doesn’t need the grid

A single large AI data center now consumes as much electricity as a city of 80,000 people. The companies building them are putting up hundreds simultaneously, in places the grid was never designed to reach, on timelines that utility companies, with their seven year queues for new connections, simply can’t comprehend. The AI revolution is stretching the nation’s electricity grid to its limits. So why not unplug? So it’s not that there’s anything wrong with the US power grid. It simply can’t provide what the country needs right now at the scale, flexibility and speed required. More than 2,600 gigawatts of proposed projects are waiting to connect to a grid whose total installed capacity is less than half that number. I truly believe that AI will create a future of abundance so profound it is almost unimaginable. But that future runs on power—enormous amounts of power, delivered to where compute is being built, when it’s needed, at a scale the existing infrastructure will never be able to meet. The industry’s response so far has been to reach for the same gas turbines and engines that have powered the grid for fifty years. While widely understood, and deeply wrong for this moment. When you need power in months and the interconnection queue stretches seven years, you reach for what is available. But building the AI century on fifty year old combustion technology is like building the internet on dial-up. It works until it does not, and the companies that accept that constraint will spend the next decade competing against the ones that did not. There is a better solution. It’s one that has been commercially deployed for years, and which has the potential to change the economics of where and how AI infrastructure gets built. Linear generators convert fuel directly into electricity through a low-temperature, flameless reaction. There’s no combustion and no complex mechanical parts. They can run on natural gas, biogas, hydrogen,