AI will get more efficient. We'll still burn more energy than ever. Jason explains the demand trap that no one's talking about. As models shrink and inference gets cheaper, the conventional wisdom says AI's environmental footprint will shrink too. Jason — who spent years inside ad tech and data infrastructure — says that reasoning misses the core dynamic: efficiency gains don't reduce demand, they enable more of it. His wine box analogy cuts to it cleanly. The box holds 12 bottles. We drink some of it. Make the box bigger, cheaper, more efficient — and people ask for more boxes. Not fewer. More. We've seen this exact pattern before. Cars got dramatically more fuel-efficient over the last 50 years. We have more cars on the road and burn more total fuel than at any point in history. Commercial aviation got radically more efficient per passenger mile. Global aviation fuel use is at record highs. Textiles, consumer electronics, compute itself — every time the unit cost dropped, total consumption went up. AI is going to follow the same curve. The models will get smaller. Inference will get cheaper. And the total number of inference calls, training runs, and server-hours will scale faster than those efficiency gains. The net result: more energy demand, not less. The uncomfortable truth is that efficiency is a tool for growth, not conservation — at least inside a system that has no mechanism for saying "that's enough." Jeremy and Jason dig into what that means for the AI infrastructure buildout, the Pentagon's entanglement with AI power consumption, and whether there's any realistic path to making this sustainable. 🎙️ Brobots is a weekly tech podcast hosted by Jeremy Grater and Jason Sisneros — covering AI, health, and what it means to be a better human in a world that's changing faster than the ethics can keep up. 📍 New episode every Monday → https://brobots.me 🎧 Listen on Spotify, Apple Podcasts, and everywhere else #AI #DataCenters #Sustainability #TechPodcast #AIEnergy