KDB-X is GA: Meet the unified compute engine for high-performance AI and time-series analytics

KDB-X is a new generation of kdb+, designed to support modern AI and data workloads in capital markets without the complexity of fragmented systems.

It combines time-series analytics, vector search, and GPU-accelerated compute in a single runtime, allowing teams to build and operate real-time systems across research, backtesting, and production.

The general availability release of KDB-X focuses on five areas that matter most to engineering and quantitative teams: performance, cost efficiency, GPU acceleration, support for AI workloads, and openness.

1. Performance: Built for high-volume, low-latency workloads

KDB-X extends the performance characteristics of kdb+ into modern AI and data pipelines, with a focus on predictable performance under real-world conditions.

In STAC-M3 benchmarking—the industry standard for tick-level analytics—kdb+ continues to deliver leading results across mean response time and multi-user throughput, outperforming alternative architectures used by major banks and trading firms. As kdb+ remains the compute engine behind KDB-X, these benchmarks continue to be directly relevant to capital markets workloads.

More recently, KX benchmarked KDB-X against QuestDB, ClickHouse, InfluxDB, and TimeScaleDB using the TSBS benchmark, a de facto community standard for time-series databases. KDB-X won 58 of 64 benchmark scenarios across aggregation, filtering, and group-by workloads under defined test conditions. In worst-case queries, ClickHouse was observed to be up to 1,100x slower, highlighting the impact of architectural differences in time-series processing.

These results were achieved with significantly lower resource utilization. KDB-X delivered faster query performance while using a fraction of available compute, running with just 4 threads (1.5% of total CPU capacity) and 16 GB of memory (8% of system resources), while competing systems required full hardware utilization. This level of efficiency allows teams to scale high-performance workloads without proportional increases in infrastructure cost.

2. GPU acceleration: 10x–25x faster on core workloads

KDB-X introduces native GPU acceleration directly within the compute engine, allowing time-series workloads to execute without leaving the product environment.

This includes core operations such as joins and aggregations, as well as higher-level workloads like backtesting, risk simulation, and large-scale model scoring. Across these use cases, KDB-X delivers 10x to 25x performance improvements, with near-linear scaling across multiple GPUs, based on internal and third-party benchmarks.

The impact is practical. Workloads that were previously constrained by CPU-bound processing—such as intraday backtesting or real-time risk recalculation—can now operate at significantly higher frequency and lower latency. Instead of moving data to external GPU pipelines, compute is brought directly to the data, reducing overhead and simplifying system design.

3. Dual-mode AI: Time-series and vector in one system

KDB-X is designed to support both structured and unstructured data without requiring separate systems. It combines a high-performance time-series engine for market and sensor data with a vector database for embeddings, documents, and alternative data

This “dual-mode” architecture allows teams to work across both data types using a single query model. For example:

• Time-aligned joins can be applied to both structured data and vector retrieval results
• Windowed aggregations and temporal logic operate consistently across all data
• AI pipelines can combine deterministic analytics with generative models without moving data between systems

This reduces architectural complexity and improves consistency across workflows.

4. Cost efficiency: High performance without high infrastructure cost

KDB-X is designed to deliver performance without requiring large-scale infrastructure. In one example, organizations have built a Level 1 equities data system processing over 200GB of data daily using a 64 GiB instance of KDB-X.

That translated to a license cost of just over $5,000 per month.

KDB-X achieves this through:

• Efficient in-memory processing
• Low CPU and memory overhead
• The ability to handle high data volumes without horizontal sprawl

For teams managing large data pipelines, this significantly reduces total cost of ownership compared to multi-system architectures. If performance and efficiency are critical to your firm, KX is making it easy to get started inexpensively with a path to growth, all with the technology the top capital markets firms depend on. You can view transparent list pricing here.

5. Open by design: Formats, languages, and ecosystem

KDB-X is designed to integrate into existing environments without imposing proprietary constraints. It includes:

• Native support for open data formats such as Apache Parquet and Arrow
• Polyglot access via Python, SQL, and q
• REST APIs and standard data interfaces
• Coding assistance (Claude code, Gemini CLI, OpenAI codex)

This allows teams to:

• Query external datasets directly without conversion
• Integrate with existing tools and pipelines
• Avoid vendor lock-in at the data layer

KDB-X also introduces a modular ecosystem, enabling teams to build, share, and reuse components across workflows. KDB-X already has 25 community-contributed modules spanning ingestion, transformation, analytics, and utility functions, with more in development.

In addition, a built-in MCP (Model Context Protocol) server enables natural language querying through AI assistants, expanding access to data without requiring specialized tooling.

What this means in practice

KDB-X simplifies how systems are built. Instead of maintaining separate pipelines for time-series data, vector search, GPU processing, and research and production environments, teams can operate on a single compute layer. This reduces data movement, system duplication, and operational complexity, allowing workflows to move more directly from development into production.

Learn more

KDB-X is now generally available, including GPU acceleration and a free Community Edition.

To see how it works in practice, join us on April 22 for a live session with myself and Michael Gilfix, where we’ll walk through architecture, performance, and real-world deployment patterns.

Register for the webinar here.