Grow fast and overload things

OpenAI and Anthropic are struggling with reliability, with ChatGPT failing to even reach two nines of uptime. The root cause isn't high development velocity but rather rapid, unpredictable growth in user demand — a phenomenon resilience researcher David Woods calls 'florescence.' GPU capacity constraints prevent simple scaling solutions, so these companies must invest in load shedding, resource redistribution, and graceful degradation. The author expects loads to become even more unpredictable as users continue to innovate with LLMs.