Why OpenAI chose ClickHouse for petabyte-scale observability

OpenAI faces the formidable challenge of managing massive volumes of log data, equivalent to 500 Libraries of Congress daily, as it supports various high-growth AI systems like model research, ChatGPT, and enterprise APIs. To handle the rapid growth and complexity of their observability needs, OpenAI selected ClickHouse, an open-source, cloud-native database, for its flexibility, scalability, and ease of integration into AI workflows. When a surge in user activity overwhelmed their systems during the launch of image generation in GPT-4o, the team optimized performance by reducing CPU usage through a simple code change in Bloom filter indexing, showcasing ClickHouse's adaptability. OpenAI continues to refine its observability infrastructure, focusing on smarter query planning and integrating AI agents for autonomous incident management, all while contributing to the broader ClickHouse community through shared improvements.