Durantic launches as managed infrastructure layer for AI workloads across fragmented GPU fleets

Durantic, founded by infrastructure engineers from Meta and Hudson River Trading, emerged publicly as a managed infrastructure service for customers operating heterogeneous GPU fleets across customer-owned, colocated, leased, and hybrid environments.

The company offers a 30-day pilot followed by recurring monthly platform-and-operations fees scaled by GPU and server count. A bare-metal-native control plane with agents on every machine handles provisioning, networking, storage coordination, telemetry collection, and the full infrastructure lifecycle from hardware acceptance through RMA (return merchandise authorization — the process for damaged hardware claims).

Durantic targets the operational bottleneck in AI infrastructure: as workloads fragment across hyperscalers, GPU clouds, colocated bare metal, and sovereign clusters, operationalization of mixed GPU generations, Kubernetes and Slurm orchestration, custom network fabrics, and hardware lifecycle management has become the constraint rather than hardware availability itself.