Intelligence Per Kilowatt Hour: The New AI Metric | Deep Geeks ep04

Demand for intelligence looks unlimited. The supply of kilowatts is very much not. That gap is the metric nobody is putting on the board deck yet. https://www.weka.io/podcast/?utm_source=youtube&utm_medium=social In this clip from DeepGeeks Episode 4, an investor who came to AI through energy technology explains a thesis that started six or seven years ago. Renewables got remarkably cheap, one of the genuine wonders of the modern economy, but you still have to connect them. There is no grid yet that moves at the speed of renewables, let alone at the speed of AI. When utilities were asked back then how they were preparing for the coming AI power wave, the answer was that they did not see anything yet. That led to a thesis called green data, and it operates on two levels. First, how do you add grid capacity quickly and in a low carbon way. Second, how do you get more tokens per kilowatt hour out of the capacity you already have. GPUs burning power without doing useful calculation, because data is not being fed in fast enough, is pure waste. Feed the cluster properly and the numbers move: one WEKA customer took a training run from two weeks down to four hours. The same pattern is now showing up in inference through WEKA's Augmented Memory Grid. The question worth asking at any scale: am I getting the most intelligence out of every kilowatt hour I draw? Learn more: Listen to the full episode: https://www.weka.io/podcast/?utm_source=youtube&utm_medium=social&utm_content=full-episode WEKA NeuralMesh: https://www.weka.io/product/neuralmesh/?utm_source=youtube&utm_medium=social&utm_content=product-page Subscribe for more AI infrastructure content: https://www.youtube.com/@WekaIO?sub_confirmation=1