Case Study: OpenAI achieves petabyte-scale observability with ClickHouse

Why OpenAI chose ClickHouse for petabyte-scale observability

OpenAI, the research lab and product engineering organization behind ChatGPT and its enterprise APIs, needed an observability system that could handle petabytes of logs per day, rapid ingestion spikes, and highly searchable, low-latency access across multiple high-cardinality workloads. Its teams needed reliable visibility into everything from billion-model research experiments to consumer traffic surges and mission-critical API incidents, making scale, speed, and flexibility essential.

OpenAI chose ClickHouse for its open-source, cloud-native observability backend, using ClickHouse Cloud-style horizontally scalable infrastructure with sharded clusters, replicas, and tiered storage for recent and historical logs. The team also optimized Bloom filter indexing in ClickHouse with a small code change that cut CPU usage by 40%, helping them absorb a 50% overnight log-volume surge from GPT-4o image generation while preserving fast query performance and giving OpenAI a more efficient, future-proof observability stack.

OpenAI

Akshay Nanavati

Engineering Manager