Aranya Raises $11M for Software It Says Can Ready GPU Clusters in 48 Hours

Aranya has raised $11 million to build software for a stubborn stage of AI deployment: turning delivered GPU servers into clusters that can run production workloads. The company says its open-source clusterdOS engine can convert racked bare-metal hardware into customized AI clusters in under 48 hours.

The financing includes a $9 million seed led by First Round Capital and a $2 million pre-seed led by Asylum Ventures. Box Group, Vermilion Cliffs and Asylum joined the seed, while Founder Collective, Parable VC and Uncommon Ventures participated in the pre-seed.

Turning hardware into an operating system

ClusterdOS is built on Kubernetes, software used to coordinate applications running in containers. Aranya says its engine uses configuration files to deploy and maintain AI infrastructure, coordinating workloads, networking and storage across different hardware environments.

The company says the product also works with virtual machines and Slurm, a system for scheduling computing jobs. Its pitch is to handle failures closer to their source: Aranya says clusterdOS can identify and resolve problems such as overheating GPUs and network faults, rather than only moving work away from a failing server.

A control layer beyond one cluster

Aranya says clusterdOS provides federated control across multiple clusters. It plans to use the funding to expand engineering, sales and marketing, and to launch a full multicluster interface.

What the planned interface is meant to change

• Engineers could create inference endpoints using plain-language requests instead of directly editing configuration files.
• They could add or remove nodes through the same interface.
• Aranya says requests would remain limited by each user’s existing permissions, preserving an auditable action set.

The company’s public performance evidence remains limited. Aranya says a deployment with Hydra Host cut cluster setup from six weeks to less than 48 hours and reduced outages by 90%; those are company and partner claims, not independent benchmarks. The funding now tests whether that approach can become a repeatable control layer for varied GPU fleets.