TiDB X

TiDB X is a cloud-native distributed SQL database from PingCAP that re-architects the classic TiDB design around object storage as a shared persistent layer. It is built for teams that need horizontal scalability, strong consistency, and the ability to run both transactional (OLTP) and analytical (OLAP) workloads on a single unified query engine. The target audience includes enterprises managing multi-region deployments, AI applications that combine vector search, knowledge graphs, and JSON with SQL, and organizations seeking to reduce total cost of ownership by separating compute from storage. TiDB X is positioned as an AI-native database, supporting integrations with OpenAI, Hugging Face, Cohere, Gemini, Jina, and NVIDIA for retrieval-augmented generation and reasoning workloads.

TiDB X introduces a shared-storage architecture where object storage serves as the backbone, allowing compute and storage to scale independently. Classic TiDB already separates TiDB servers (SQL parsing and query planning) from TiKV (key-value storage) and TiFlash (columnar analytics), but TiDB X pushes this further by making storage stateless and elastic. The system provides context-aware scaling based on QPS, latency, query mix, and data types. It supports long-term memory with versioned, branchable storage, enabling time-travel queries and data branching for development and testing. The unified query engine fuses vectors, knowledge graphs, JSON, and SQL, making it possible to perform hybrid search (vector + full-text + SQL) in a single query. Benchmark data from 2024-2025 shows TiDB achieving 52K TPS on OLTP mixed workloads, 95K QPS on read-heavy workloads, and 40K TPS on write-heavy workloads on comparable cloud instances (c5.4xlarge, three regions).

In the distributed SQL market, TiDB competes directly with CockroachDB, YugabyteDB, and Google Spanner. Compared to CockroachDB, TiDB offers higher raw throughput (52K vs 45K TPS on OLTP mixed) and better analytical performance thanks to TiFlash. Against YugabyteDB, TiDB provides automatic geo-partitioning and lower P99 write latencies for geo-distributed writes (30-80ms vs YugabyteDB's higher range). However, CockroachDB has a simpler operational model with a single tightly coupled architecture, while TiDB requires managing two distinct services (TiDB + TiKV). YugabyteDB offers closer PostgreSQL compatibility, while TiDB is MySQL-compatible. TiDB X's shared-storage architecture is a differentiator, enabling instant elasticity and cost optimization that competitors' shared-nothing designs do not match.

The honest trade-offs: First, operational complexity is real — running TiDB means managing separate TiDB, TiKV, and TiFlash components, plus now object storage configuration. Second, while TiDB X reduces storage costs through object storage, the compute layer can still be expensive under sustained high QPS due to the need for many TiDB nodes. Third, MySQL compatibility is not perfect; some MySQL features (e.g., stored procedures, triggers, full-text indexes in older versions) have limited support, which can complicate migrations. Fourth, multi-region deployments require careful placement rule tuning to avoid cross-region latency penalties; the 30-80ms P99 write latency is good but not as low as single-region setups.

TiDB X is available via TiDB Cloud Premium in public preview with unlimited growth and instant elasticity; specific per-node or per-credit pricing is not publicly disclosed. Classic TiDB offers a free tier (TiDB Serverless) and paid tiers (TiDB Dedicated) with hourly compute and storage costs.