ClickHouse vs Datadog for AI Observability

This is a fight over who becomes the system of record for AI exhaust before enterprise buying habits harden. AI apps generate huge streams of append only events, prompts, traces, tool calls, evals, and cost logs, that look more like clickstream data than classic app monitoring. ClickHouse is winning these workloads with much lower storage economics and by bundling purpose built AI observability tools, while Datadog still owns the easier, all in one DevOps workflow.1 2 3 4 5 6

• ClickHouse has turned a database advantage into a product wedge. It bought HyperDX in March 2025 to add an open source observability UI, then bought Langfuse in January 2026 to add LLM tracing, evaluations, and prompt monitoring, effectively packaging database, telemetry backend, and AI observability in one stack.2 3 4 5
• The cost gap matters because AI telemetry is unusually voluminous. Internal interviews describe ClickHouse as 10 to 20 times cheaper than Datadog for high cardinality observability data, and ClickHouse has publicly framed Langfuse workloads as moving from minute level queries to near real time analytics on rapidly growing ingestion volumes.1 4 7
• Datadog is not absent, it is defending from a stronger control point. Datadog reported about 5,500 customers using one or more AI integrations by February 2026, and positioned LLM Observability as something customers can fold into existing APM, logs, and user experience contracts, which is powerful where engineering teams already run incident response inside Datadog.6 8 9

The next step is a stack split. AI native teams will start with ClickHouse for cheap long retention telemetry and model specific workflows, while incumbent enterprises will extend Datadog into AI because it already sits inside on call, alerting, and dashboards. The company that best connects raw agent traces to daily engineering operations will own the larger budget over time.1 2 5 6 9