Menu

Architecture / 01

Intelligence, connected.

Our solutions
Reference architecture
Distributed AI architectureReference architecture. AI Factory, regional Grid, Edge-AI Hubs and Physical AI are linked by networks. iFabric operates across the estate; gFabric governs relevant identities, placement and actions. Teal circular signals carry evidence, approved workloads and results; copper signals carry authorized actions; dashed violet routes carry approved model packages from customer or partner systems. Diamond checkpoints indicate policy authorization before regional work or physical action. A local path does not require upstream processing. Animation timing is illustrative.Configured local operating boundaryConfigured local operating boundaryConfigured local operating boundaryConfigured local operating boundaryapproved model factory to grid 0approved model factory to grid 1approved model factory to grid 2approved model grid to edge 0approved workload edge to grid 0result grid to edge 0field evidence to local edge 0authorized local response 0approved model grid to edge 1approved workload edge to grid 1result grid to edge 1field evidence to local edge 1authorized local response 1approved model grid to edge 2approved workload edge to grid 2result grid to edge 2field evidence to local edge 2authorized local response 2approved model grid to edge 3approved workload edge to grid 3result grid to edge 3field evidence to local edge 3authorized local response 3AI Factory01AI FactoryPrepareAI Grid02AI GridDistributeEdge03EdgeInferPhysical AI04Physical AIActgFabric governance spans the systemgFabric / GoverniFabric operations span the estateiFabric / OperatePolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicyNetworksConfigured local boundaryEdge-AI Hub 1, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointEdge-AI Hub 2, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointEdge-AI Hub 3, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointEdge-AI Hub 4, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointAI Factory: central compute and customer or partner model preparationCentral computeRegional AI Grid capacity 1Region 01Regional AI Grid capacity 2Region 02Regional AI Grid capacity 3Region 03Distributed AI architectureReference architecture. AI Factory, regional Grid, Edge-AI Hubs and Physical AI are linked by networks. iFabric operates across the estate; gFabric governs relevant identities, placement and actions. Teal circular signals carry evidence, approved workloads and results; copper signals carry authorized actions; dashed violet routes carry approved model packages from customer or partner systems. Diamond checkpoints indicate policy authorization before regional work or physical action. A local path does not require upstream processing. Animation timing is illustrative.Configured local operating boundaryConfigured local operating boundaryapproved model factory to grid 0approved model grid to edge 0approved workload edge to grid 0result grid to edge 0field evidence to local edge 0authorized local response 0approved model grid to edge 1approved workload edge to grid 1result grid to edge 1field evidence to local edge 1authorized local response 1AI Factory01AI FactoryAI Grid02AI GridEdge03EdgePhysical AI04Physical AIgFabric governance spans the systemgFabric / GoverniFabric operations span the estateiFabric / OperatePolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicy authorization before the path continuesPolicyNetworks · Local boundariesEdge-AI Hub 1, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointEdge-AI Hub 2, local inference and nLLM reasoning where deployedEdge-AI HubnLLM · optionalPhysical AI evidence sourcePhysical AI authorized action endpointAI Factory: central compute and customer or partner model preparationCentral computeRegional AI Grid capacity 1Regional capacity

Local inference. Regional capacity. Model delivery.

Data / resultsAuthorized actionModel package

Model preparation & delivery: customer / partner systems.

AI Factory, regional Grid, Edge and Physical AI, connected through governed networks.

iCore: private cellular where selected. Other access integrates by design.

Next: incident recovery ↗Download diagram ↓
Architecture notes

The system. Independent reference paths show local evidence and authorized response, approved work using regional capacity, and customer or partner model distribution. They are not one mandatory path through every environment.

Local inference. Field evidence reaches the configured Edge-AI Hub. Local reasoning supports a response through an authorization gate. This path completes locally without an upstream request.

Regional capacity. A policy gate precedes crossing the configured local boundary. Only an eligible, authorized workload uses regional capacity. Its result returns to the edge. If no destination qualifies, the workload is deferred or escalated.

Model delivery. Customer or partner systems prepare and distribute an approved model through the configured regional path to the edge, unless an Invences integration is separately confirmed. This is model delivery, not automatic training from live telemetry.

Illustrative flows, not live telemetry or measured performance. Connectivity, model distribution and action integrations depend on the deployment.