Sovereign AI and HPC, Defined
Plain definitions of the terms behind sovereign AI and HPC, from Vantage Compute: secure enclaves, GPU chargeback, fair share, cloud bursting and more.
One sentence each.
Secure research computing enclave
A secure research computing enclave is a boundary around compute and data where every user has one identity, access is granted by policy, and every action is audited.
Sovereign AI
Sovereign AI means running AI on infrastructure you control, in your own jurisdiction, so that data never leaves your boundary.
GPU chargeback
GPU chargeback is the practice of attributing GPU usage, and what it cost, to the user, team or project that consumed it.
Fair-share scheduling
Fair-share scheduling divides a shared cluster so that each team gets its agreed portion of capacity over time, rather than whoever submits first.
Slurm and Kubernetes identity
Slurm and Kubernetes identity means one user identity that works on both schedulers, in place of a separate account on each cluster.
Cloud bursting
Cloud bursting is sending work to cloud capacity when your own clusters are full.
Deny-by-default access
Deny-by-default access means a user can do nothing until a policy grants it.
Audit trail for GPU usage
An audit trail for GPU usage is a record of who ran what, where, and on whose approval.