Summary
Cerebras CEO Andrew Feldman discusses the unprecedented AI infrastructure buildout, with data centers the size of cities and insatiable demand from hypescalers. He explains how Cerebras inference chips break Moore’s Law, why reasoning will consume massive compute, and why open source and sovereign AI are reshaping the market. Black Forest Labs CEO Robin Rombach then explains generative video capabilities, partnerships like Martin Scorsese, and how multimodal models will power both content creation and robotics.
- AI buildout is on a scale not seen since the pyramids or the war effort, with data centers using more power than entire cities.
- Cerebras has a $25B backlog and its inference chips deliver reasoning speed way above Moore’s Law, making token-heavy reasoning tractable.
- Open-source models are catching up to frontier models, and domestic open-source options are needed for sovereignty and regulation.
- AI safety debates center on whether powerful reasoning models should be rolled out in steps and red-teamed; partisan noise complicates clear thinking.
- Black Forest Labs’ generative video tools are being used by filmmakers like Scorsese for pre-visualization and set replacement, lowering production costs.
- Multimodal models trained on images, video, and audio are converging toward world action models that can serve both movie-making and robot control.
- The future may see interactive fan content and AI-empowered storytelling built on licensed IP libraries.