Residency first.
Then deploy.
Most “sovereign AI” pages sell a cloud badge. This one asks the only questions that matter: where the data must sit, who the buyer is, what the licence path is, and whether the model can run inside that boundary. Compute and a local team follow that. They do not precede it.
What this program is for
Enterprise and industrial AI
Models that sit on a factory, a hospital file or an operations stack — not a consumer chatbot.
Platforms that must stay in-country
Software that only sells if residency, access control and audit can be shown.
Applied, not generic
A named use case and a named buyer type. Not “AI for the UAE.”
What we will not write on this page
We do not hold a G42 contract
Residency is a requirement to test. Naming a hyperscaler as if we sit inside it is a claim we will not make.
We do not implement AI 2031
National strategy is context. 360Disruption does not award government AI status.
We do not sell defence adjacency
If a file is actually defence-adjacent, that is a separate clearance path. It is not marketing copy.
We do not start with a data centre
A node is a destination. First: data class, buyer, licence, and whether the model can run locally at all.
How an AI file moves
Is there a UAE case
Use case, data class, residency constraint, buyer type, licence path, IP ownership. Go / hold / redirect.
Proof, not presence
A contained pilot with a real counterpart. Evidence that the model works inside the boundary the buyer requires.
The footprint the file earned
Entity, hosting arrangement, support bench — only what the pilot justifies.
People and compute after demand
Engineering presence and in-country capacity when the offtake is real. Not a knowledge-hub slogan.
The method is Prove before you invest. The desk that owns the file is the FDI Execution Desk.
Sovereign is a constraint. It is not a brand.
If the data cannot leave, say so in Discovery. If it can sit in a commercial UAE cloud, say that too. Pretending every model is a national system wastes the buyer and the company.