Aranya

Aranya turns raw bare metal GPU hardware into custom, production-ready clusters in under 48 hours, targeting AI inference and training providers, neoclouds, and sub-20MW data centers that have the servers but not the platform layer to sell them as usable AI compute. Its flagship product, ClusterdOS, transforms Kubernetes into a self-healing system for AI supercomputers, handling cluster lifecycle management through simple feature flags , and handles the full cluster lifecycle, bootstrapping, maintaining, and upgrading with minimal effort . In its deployment with Hydra Host, it reduced production cluster setup from 2 to 6 weeks to under 48 hours and cut cluster downtime by 90%. The launch matters because inference workloads are quickly outgrowing the tooling built for training runs, and operators are stuck gluing together Kubernetes, kernel drivers, firmware, and federation by hand. Aranya positions inference as the core value-extracting workload of the AI era, with infrastructure demands that are just as unforgiving as the crypto era of GPU compute . The company is now expanding beyond ClusterdOS into an AI-native multicluster interface that lets teams spin up inference endpoints, provision VMs, and add or remove nodes in plain language rather than through config files or CLIs , with autonomous agents that live inside the cluster and adapt to what it needs rather than executing pre-scripted jobs . Aranya was founded by Christian Bhatia Ondaatje, Sasivarnan Kanaghasalam Sathyapriya, and Aryamika Bhatia Ondaatje , with Ondaatje as CEO after earlier stints as an early engineer at Crusoe and a founding platform engineer at Hyperbolic. The company is based in San Francisco and reports operating more than $500M of compute across its partners, with roughly a year of build time behind the public debut and a fresh round supporting hires across platform engineering, SRE, and developer relations for the open-source ClusterdOS community.