ClickHouse 24.10 performance and the OLAP race

Speed advantages in real-time query execution

I see ClickHouse executing wide-table aggregations 10x to 100x faster than Snowflake on 1 billion rows of event logs. The 120ms response from ClickHouse beats the 1.2s Snowflake response on a 16-core machine. While Snowflake handles 10TB aggregations via its elasticity, ClickHouse uses a shared-nothing architecture with local SSDs to minimize the network latency inherent in S3-backed storage. You likely already know that Snowflake excels at batch-heavy, internal reporting. However, Pinot supports 100,000+ QPS at millisecond latency for companies like Uber and LinkedIn. Pinot utilizes a three-tier architecture with Controller, Broker, and Server nodes. Druid supports sub-second interactive dashboards through continuous Kafka ingestion and segment-based handoffs. Pinot serves the advertising technology space through low-latency, high-concurrency analytics for user behavior. Druid handles observability, clickstream, and metrics through rollups at ingest. ClickHouse handles log and event data analysis through high-performance query engines. It also provides real-time reports and dashboards for IoT and data warehousing workloads. Pinot provides fraud detection and recommendation engine capabilities for high-concurrency needs.

ClickHouse performance remains a factor in the competitive OLAP market. In the latest 26.10.1.2121 build, the encrypt_decrypt test showed a +1.003x speedup. The bitmap_cardinality test in that same build showed a -1.312x slowdown. Pinot achieves p99 query latency under 100 ms for high-throughput workloads.

Scaling for the AI era and global expansion

ClickHouse closed a $400 million Series D round led by Dragoneer Investment Group in January 2026. This funding follows a $350 million Series C from May 2025, bringing total funding above $650 million. The company also secured a $100 million credit facility led by Stifel and Goldman Sachs. ClickHouse grew 300% over the past year and manages 4,000 customers. Annual recurring revenue grew more than 250% year over year. Large organizations including Tesla, Meta, and Anthropic use the engine for real-time analytics and observability. Recent customers include Capital One, Polymarket, Airwallex, Lovable, and Decagon. The company announced its product updates at Open House 2026 in Amsterdam and San Francisco.

The company launched ClickHouse Agents, which runs on Anthropic’s Claude for no-code agent building. This tool uses the Model Context Protocol to connect to external data and works with the AWS Agent Registry. As AI agents generate queries at higher rates than human analysts, ClickHouse aims to serve these high-throughput workloads. The company also acquired Langfuse, which reached 26 million SDK installs per month by the end of 2025. ClickHouse continues to discuss a new funding round at a $6 billion valuation. The company also announced its House Mates partner program with over 25 technology partners. The company reported that its cloud service surpassed $250 million in annual run-rate revenue.

When to bypass standard analytics engines

I find ClickHouse JOIN performance problematic for complex queries because the query planner does not reorder joins. High-cardinality dimensions build large in-memory hash tables that trigger OOM errors instead of spilling to disk. ClickHouse also struggles with update-heavy workloads because its engine assumes append-mostly data. Mutation costs on update-heavy workloads remain high as ALTER TABLE operations are async and rewrite-based. Does the increased query volume from AI agents eventually force a move toward more specialized engines?

For decision-time context, Tacnode maintains freshness within 10 to 50ms latencies through a distributed context infrastructure that maintains complete, consistent, and current context for automated decisions to evaluate against, ensuring the decision remains valid. Tacnode handles structured tables, semi-structured JSON, and vector embeddings in one engine. This provides a coherent view across structured state and vector retrieval.

Monday.com addressed similar scaling needs by building MondayDB. They increased capacity 10x, which enables 500 boards per dashboard, 1 million items per board, and 5 million items per dashboard. Their solution provides 5 nines of data integrity and 5x the speed from the previous database. The team began exploring new database ideas in January 2021. Monday.com uses a combination of technologies including Cassandra, Redis, and MySQL to manage these requirements. They tested the new system internally in September 2022. The engineering organization modified every core component to work with the new database.