MariaDB Claims 56% Throughput Gain in Platform for AI Agents

MariaDB has shipped a major platform release that posts 56 percent higher peak throughput than the last major version in third-party testing. The company announced MariaDB Enterprise Platform 2027 as a single database for enterprise AI agents. It is generally available now.

The platform combines faster transactional performance, real-time analytics, in-memory speed, AI tooling and multi-region scale in one product. MariaDB now positions it as the unified open database platform for what it calls the agentic enterprise.

CEO Rohit de Souza framed the launch as an infrastructure argument, not a model race. He said: "The AI race won't be won by the smartest model. It will be won by the companies that unify the infrastructure underneath their agents. Every system you bolt on for AI adds another seam, and seams break." He added that agents need a durable foundation that does not shift underneath them, so companies can trace every action agents take. "MariaDB has moved far beyond being just a database," de Souza said. "We are the platform the agentic enterprise is built on."

Why one database instead of many systems?

MariaDB argues agents deliver results faster when they draw on a single source of multi-modal data. The alternative is data spread across multiple operational databases, a warehouse, vector stores, ETL engines and caches. Every integration between those layers, the company asserts, adds latency, synchronization lag and another point of failure.

The platform's answer is a set of integrated capabilities:

• AI Retrieval grounds agents in documents and live operational data through one managed service. It turns natural-language questions into read-only SQL. It is available now in public preview on MariaDB Cloud.
• AI plugins bring MariaDB expertise into coding agents such as Claude Code and Codex, so developers get correct code the first time.
• Agent Memory gives agents a durable, in-memory store shared across sessions. Long-running agents pick up where they left off.
• Semantic Cache returns stored answers to questions already asked in different words, cutting response times and token costs.
• MariaDB Exa runs real-time analytics on live operational data without complex pipelines.

What are the measured performance numbers?

Vector search, built natively into the core engine, now delivers up to 30 percent more queries per second than the last major release. Peak throughput is 56 percent higher in third-party testing, a figure the company attributes to external benchmarking rather than internal claims. Raft clustering and conflict detection and resolution add scale and resilience for multi-region deployments.

Security and deployment options target regulated and sovereign-cloud buyers:

• A unified set of role-based access controls spans every workload.
• Bring Your Own Cloud lets enterprises keep data where they choose.
• FIPS 140-3 certification lets federal and other regulated enterprises meet their cryptography requirements.

What ships in the bundle?

MariaDB Enterprise Platform 2027 packages six components under one license framework: MariaDB Enterprise Server 12.3, MariaDB Exa, MariaDB GridGain, MariaDB MaxScale, MariaDB Enterprise Manager and the MariaDB Enterprise Kubernetes Operator. The stack runs on premises, in any cloud, or as a managed service on MariaDB Cloud.

The launch lands as enterprises move AI agents from pilots into production workloads that touch mission-critical transactional systems. MariaDB's bet is that consolidation onto a transactional core it already runs for customers will beat the stitched-together alternative of separate vector stores, caches and warehouses, each adding its own latency and failure surface. Whether that bet pays off will show first in adoption of AI Retrieval once it moves from public preview to general availability.