Snowflake as Record, ClickHouse for Speed
AI program manager at AstraZeneca on running self-hosted ClickHouse
This split shows that fast analytics engines are being added to the enterprise data stack, not replacing the warehouse at its center. At AstraZeneca, Snowflake is where teams do the heavy cleanup and governed reporting work that regulators and finance teams depend on. ClickHouse sits downstream where AI agents need answers in milliseconds, for example pulling patient journey context, ranking relevant records, or grounding responses for oncology workflows without waiting minutes for a warehouse query.
-
The line is basically batch control plane versus live execution plane. Snowflake handles multi step transformations, cross domain joins, clinical operations, and regulatory reporting. ClickHouse handles sub second metrics, logs, conversational retrieval, and the orchestration layer for agentic AI.
-
The coexistence is partly organizational. Snowflake is easier for analytics and BI teams because scaling and infrastructure are abstracted away. ClickHouse delivers much higher speed, but it needs engineers who understand sharding, replication, indexing, and cluster tuning.
-
That makes ClickHouse additive TAM inside large enterprises. It is winning latency sensitive workloads that previously sat awkwardly in Databricks, Snowflake, Elasticsearch, or Datadog, especially user facing analytics and retrieval heavy AI systems where every extra second breaks the product experience.
The next step is a more modular stack built on open storage formats, where the warehouse remains the system of record and engines like ClickHouse are swapped in for the slices of work that need instant response. That favors vendors that can plug into existing governed data estates instead of forcing full migrations.