In two years, training a frontier model will need a nuclear reactor
Greg Osuri | Trump's impact on crypto x AI, why DePIN is inevitable, and Akash Network revenue ATH's (Proof of Coverage Media)
“The next big LLM will need [a] 250 megawatt data center, and a year after that is going to be 1.2 gigawatt data center, and the only place to get 1.2 gigawatts of capacity is a nuclear reactor. So in two years we’re going to need a new nuclear reactor to train the most powerful state-of-the-art model, and I think that is a huge bottleneck we don’t talk much about.” — 00:22:29
Context: Extrapolating from GPT-4’s ~10MW training data center; he notes the last US reactor (Georgia) cost ~$32B and took ~14 years, so optimistically seven years to build.