Snowflake Makes Hybrid Tables Up to 8x Faster by Default

Performance improvements apply automatically

The updated Hybrid Tables engine is designed for transactional reads and writes inside the Snowflake database. According to Snowflake, single-statement query execution for repetitive point operations can deliver up to 8x higher throughput. For certain workloads, the company says a task that previously required eight XS warehouses can run on one, potentially cutting compute costs by more than 80%.

Snowflake also reports 3x lower median latency and 2x lower tail latency in benchmark results. The benchmarks used the Yahoo Cloud Serving Benchmark with a 100% read workload on a Gen2 XS warehouse. Lower p50 latency means faster typical responses, while lower p99 latency is intended to reduce the number of especially slow requests during peak load.

Bulk inserts have also been optimized at the storage engine level. Snowflake says benchmark results show up to 10x faster batch writes at 10x lower cost, including on tables that already contain data.

A transactional layer in the same database

Hybrid Tables are aimed at workloads such as pipeline state, AI agent context and internal applications. Snowflake says they provide fast, ACID-compliant reads and writes alongside analytical data, allowing both types of data to remain under the same governance framework and be joined in one query.

The company says customers no longer need to maintain a separate operational database, a synchronization pipeline and a second set of governance controls for these use cases. Existing applications and workflows can use the performance improvements without changes to their architecture.

Snowflake has also simplified the pricing model by removing one billing dimension. It says customers have seen costs fall by 15% on average, with reductions of 40% or more for high-throughput workloads.

Use cases include apps, pipelines and AI agents

For workflow state, Hybrid Tables can store job status, checkpoints, sessions and intermediate results. Snowflake says MarketWise replaced MySQL and DynamoDB for tracking data engineering pipeline state in real time, reducing infrastructure costs by 35% and eliminating unnecessary data movement. Elementum uses Hybrid Tables to coordinate AI agents, business rules and human steps in multistep processes.

Application data serving is another focus. Snowflake says Grailed serves personalized style recommendations to millions of users from precomputed machine learning results in Hybrid Tables, without a separate serving database. In manufacturing, logistics and supply chain operations, the tables can handle current-state lookups while standard tables support large analytical joins on the same platform.

Hybrid Tables also support high-concurrency writes through row-level locking. Writer, an enterprise AI platform, writes roughly 1.3 million LLM trace records a day to Snowflake. Snowflake says that after Writer switched the destination to a Hybrid Table, p50 insert latency dropped roughly 15x and write failures fell to near zero.

What the update means for Snowflake users

For existing Hybrid Tables customers, the main change is operational simplicity: Snowflake says there is nothing to do because the improvements are enabled by default. The company positions the update as a way to support more transactional volume on the same compute while keeping operational and analytical data together.

Snowflake also highlights Appriss Retail, which uses Hybrid Tables for real-time return and claims scoring. The company says Appriss handles more than a million lookups a day for its biggest retailer without exporting data from Snowflake. New users can create a Hybrid Table with a single statement, use it for point writes and lookups, and join it with existing tables in one query.