EDB Previews Horizontal Scaling for Postgres AI Distributed Ahead of 2027 Launch

Sharding to spread writes across node groups

EDB announced the preview on Oct. 29, 2026. The capability uses sharding to distribute transactional writes across node groups. Data and writes are spread over shard groups that process work in parallel, and each group keeps the replication, automatic failover and configurable durability of PG AI Distributed, according to EDB.

The company frames the move around a familiar scaling problem. Once writes reach the limits of a single server, read replicas cannot absorb them. Buying a larger machine adds cost and only postpones the ceiling, and replacing the database turns growth into a migration project, EDB says. The new approach is meant to offer a path past that ceiling within Postgres.

EDB also points to AI agents as a source of additional load. Every new customer and request adds demand on the database, and agents working around the clock add more. In EDB's words, availability alone is not enough when a system stays online but cannot keep pace, leaving agents unable to complete their work.

Same operational foundation, according to EDB

EDB stresses that shard groups are built the same way as the groups customers run today, with the same replication, failover and durability mechanisms. Scaling out works by adding groups, with the same runbooks, extensions and in-place failover in seconds, the company says. Postgres stays Postgres, and quorum durability with asynchronous telemetry is set as policy for each workload. The architecture is identical on-premises, in the cloud and in hybrid setups.

Max Romanenko, Chief Technology Officer at EDB, said: "A write ceiling should not force a database replacement or a retreat from the availability the business depends on."

The company argues that hitting a database limit can trigger months of work: buying more capacity, evaluating another platform, adapting applications, rebuilding availability protections and retraining the team. Extending the architecture teams already operate is meant to avoid that.

Track record cited for PG AI Distributed

EDB says PG AI Distributed already runs always-on systems for more than 75 financial institutions and is backed by more than a decade of distributed Postgres engineering. According to the company, top banks globally rely on it for up to 99.999% availability across clouds, data centers and air-gapped systems.

Its production footprint includes a 250-node cluster and a cluster managing more than 70 TB of transactional data, EDB says. One customer operates more than 350 distributed clusters across over 10,000 cores. Horizontal scaling builds on the replication, consensus, failover and routing mechanisms behind those deployments.

The target workloads are those that cannot pause, such as payments gateways, core banking, high-ingest applications and multi-tenant SaaS.

Preview status and what it means for customers

Horizontal scaling is currently in preview with customer design partners. The initial release is designed around transactions that operate within a shard group. EDB has published a technical blog and a demo preview.

For enterprises, the pitch is that growth in write volume could be handled without leaving Postgres or giving up the availability practices and infrastructure choice they already use. EDB ties the feature to its broader vision of a sovereign AI data system for the agentic enterprise, and cites a figure that 86% of enterprise leaders expect an AI data system to be critical within three years. Until general availability arrives in February 2027, the capability remains a preview, and the performance gains in practice will depend on how it performs beyond the design partner stage.