ClickHouse vs Snowflake at AstraZeneca

This split is really about who is expected to shape the database by hand. ClickHouse rewards engineers who think in query paths, sort order, sharding, and background merges, because performance comes from modeling tables around the exact way patient records, genomic events, or agent queries will be read. Snowflake gives analytics teams a safer default, because infrastructure, scaling, and much of the tuning are abstracted away, so BI and warehouse style users can stay focused on SQL and reporting workflows.

• At AstraZeneca, the friction on ClickHouse was not basic SQL, it was operational design. Teams needed to learn indexing strategy, table engine selection, cluster management, auto scaling, backups, and cost tuning. That is familiar territory for backend and SRE profiles, but a steeper jump for warehouse oriented analytics teams.
  1 sacra
• The payoff for doing that engineering work was very large. AstraZeneca used ClickHouse for sub second retrieval across petabytes of patient data, with simple aggregations in 30 to 40 milliseconds and complex groupings under 200 milliseconds, while Snowflake remained the system for regulatory reporting, transformations, and cross domain governance.
  1 sacra
• This maps to the broader market. ClickHouse has spread fastest in engineering led, user facing analytics use cases at companies like Ramp, GitLab, and Statsig, where low latency matters enough that teams will tune the system. Snowflake still fits best where the buyer wants managed elasticity, familiar warehouse workflows, and minimal operator burden.
  2 sacra 3 sacra

Going forward, this persona divide will keep both products in the stack more often than one replacing the other. ClickHouse will keep moving up from observability and real time AI retrieval into more production analytical workloads, but the biggest wins will come in teams that already have strong systems engineers and a reason to care about every millisecond and every dollar per query.