newsroom
AI Infrastructure · Modal: Sandboxes for RL/coding agents driving 2x/mo growth; alternative accelerators (TPU/AMD) face CUDA switching costs but will matter in 2-3 years
now playing · AI Infrastructure
Tabarrok: AI models commoditizing 1000x faster than prior tech, gains widespread but economic impact lagged
Frontier models are only 6 months ahead of open source, and even older free models are 100x better than prior gen, so AI gains will diffuse widely; however, like electricity, economy-wide a…
Modal: Sandboxes for RL/coding agents driving 2x/mo growth; alternative accelerators (TPU/AMD) face CUDA switching costs but will matter in 2-3 years
Code execution sandboxes are the fastest-growing primitive for reinforcement learning and agentic workflows; GPU market stays tight for 1-2 years; speech-to-speech latency breakthroughs are…
Schneider: China's humanoid manufacturing scale will outpace West once AI software matures, Taiwan remains critical regardless of chips
China leads in hardware manufacturing for embodied AI (humanoids) and will scale faster once economic utility unlocks; US retains software/chip advantage but Taiwan risk persists beyond sem…
Secret Level: AI-native studios slash film costs 10x but storytelling and workflow pipelines remain the differentiator
AI can reduce production costs dramatically (e.g., $500k film with $400k compute), but sustainable businesses require proprietary workflow pipelines and storytelling excellence, not just mo…