CoreWeave launches managed database for AI teams building production apps

CoreWeave built its reputation renting out GPUs to companies training and running AI models. Now it wants to hold their data too.

The company has introduced CoreWeave Database, or CWDB, a managed database platform meant to help AI teams build, run and scale production applications.

The service is PostgreSQL-compatible and currently in limited availability. It runs inside CoreWeave Kubernetes Service (CKS) clusters, the company’s managed environment for orchestrating containerized workloads.

What CoreWeave actually shipped

CWDB handles high availability, daily backups and monitoring through a Kubernetes operator.

An operator is software that encodes the know-how of a human administrator. It watches the database and steps in when something needs fixing. That leaves engineers free to work on the application rather than babysit the infrastructure underneath it.

The database is defined using Kubernetes Custom Resources. Developers describe the database they want in a configuration file, and the system works out how to build and maintain it.

Data lives on CoreWeave’s Distributed File Storage (DFS) system, which handles both live data and backups. Each instance gets 50 GiB of storage by default.

The default setup runs three or more instances. That number matters. With three copies, the cluster can reach quorum, meaning a majority can agree on the current state of the data even if one node goes down.

Compute and storage can be adjusted. Instances range from 2 to over 8 vCPUs, with memory running from 8 to over 32 GiB of RAM.

Teams can manage the database through the Cloud Console, kubectl commands or GitOps tools.

Why a GPU company wants your tables

CoreWeave was founded in 2017 and went public in March 2025. It operates more than 50 data centers.

Its core business is GPU-centric cloud computing. AI applications, though, are rarely just a model sitting on a GPU. They also need somewhere to keep user records, session data, application state and the many other things that do not belong in model weights.

CWDB puts that layer inside the same Kubernetes environment where those teams already operate. The research findings describe the goal as reducing operational overhead for customers already familiar with the CoreWeave ecosystem.

There is no formal announcement yet of a broad public launch. That fits the company’s broader approach of rolling out features in phases and making some capabilities available through customer requests first.

What this means for CoreWeave and its customers

The choice of PostgreSQL softens vendor lock-in concerns somewhat. It is a widely used open-source database, so applications written for it are not locked into proprietary query languages.

The specs also offer a hint about the intended use. Instances topping out in the range of over 8 vCPUs and over 32 GiB of RAM suggest a service aimed at application backends that sit alongside AI workloads, rather than a heavyweight analytics warehouse.

What to watch next is straightforward. First, whether CWDB moves from limited availability to general access, and on what timeline. Second, whether CoreWeave keeps adding managed services around its GPU core. Third, how customers respond to having one more piece of their stack handled by the company that already runs their compute.