Training sets, model weights, inference caches and data lakes each live on their own storage; copies multiply and compute spends longer waiting for data.
AetherStore is Jumborca's own AI data foundation and the foundation layer of the Storage Power Center's three-layer architecture. It is designed for AI workloads: file, object and block storage in one namespace; separate performance tiers for training, inference, data lakes and archives; built-in data protection, access control and end-to-end observability. Data stays on the foundation and is scheduled next to whichever compute needs it, so large-scale copying is avoided.
Capabilities
- One namespace: file, object and block storage in a single namespace, one copy of the data reached in several ways
- Performance tiers: the right performance and cost tier for training, inference, data lakes and archives
- Built-in data protection: snapshots, replicas and access control ship with the foundation
- End-to-end observability: capacity, throughput and access paths are visible, with capacity forecasting through SupInsight
- Not tied to hardware: a software foundation we own, running on a range of hardware and public-cloud object storage