Data residency is the guarantee that data is stored and processed within a specified jurisdiction; sovereign deployment extends the guarantee to the whole system — models, runtime, and operations run inside infrastructure the owning institution controls, up to fully air-gapped environments. For public-sector AI in Canada, residency is the entry requirement, not a differentiator.
Residency claims are only as strong as their weakest subprocessor: a platform hosted in-country that routes inference, telemetry, or support access abroad has not kept the promise. Serious vendors disclose the full chain and contract to it.
Sovereign deployment is an architecture choice made early or never: systems designed around a single vendor's managed cloud rarely retrofit into classified or air-gapped environments. The honest question for any platform is not 'where is the data?' but 'what breaks when the internet is removed?'
This vocabulary is implemented, not aspirational — it describes how the MAIA substrate actually runs. See the platform or read the research.