The ‘BitTorrent for LLMs’ Dream Is Dead. Physics Killed It.
The dream of a ‘BitTorrent for LLMs’—pooling idle GPUs to run massive models—sounds like the ultimate democratization of AI. But the metaphor is a category error. LLM inference is a real-time, latency-sensitive sequential computation, not a static download. The cold truth? Physics doesn’t care about your democratic ideals. Here’s why the P2P dream died, and where the real AI revolution is actually happening.