Menu

Use case / 03 · gFabric

Place the workload. Keep the policy.

Identity first. Eligible capacity next. Authorized execution.

All use cases

Configured identity, policy & capacity sources

Place a workload under policyA workload needs capacity, but its identity, data rules and operating requirements limit where it can run. Apply policy before comparing permitted destinations, check capacity, then authorize the selected placement plan. Execute through configured tools and record usage afterwards. Execution requires authorization and configured customer or partner tooling. Defer / review is an illustrative proposed exception flow, subject to deployment scope. Policy filters candidate destinations before placement selection. The branches are placement plans, not workload execution. An AI Factory, regional Grid or Edge-AI Hub is selected as an alternative, not a mandatory hop. The no eligible destination scenario stops at a proposed defer / review step. gFabric governs; configured customer or partner tools execute. Animation repeats the illustration, not a workload.Eligible placement plansSelection only · No executionProposed exception flowConfirm in deployment scopeIllustrative sequence · Repeatsidentify to policypolicy to capacityauthorize to executeexecute to recordPlan AI Factory placement after policy eligibility and capacity checks. No workload executes here.AI Factory placement plan requires action authorization before execution.Plan Regional Grid placement after policy eligibility and capacity checks. No workload executes here.Regional Grid placement plan requires action authorization before execution.Plan Edge-AI Hub placement after policy eligibility and capacity checks. No workload executes here.Edge-AI Hub placement plan requires action authorization before execution.No eligible destination: illustrative proposed defer / review flow. No execution.Establish the workload identity, model needs and operating requirements, including latency, data location and resource demand.01IdentifyIdentity + requirementsApply configured identity, access and data rules to exclude destinations before selection. Available capacity does not override those rules.02PolicyData + access rulesCompare capacity and operating constraints only within the policy-permitted set. Use configured resource signals to form a placement plan for an eligible AI Factory, regional Grid or Edge-AI Hub.03CapacityEligible resourcesAuthorize the selected placement plan under the configured action policy. Selecting an eligible destination alone does not permit execution.04AuthorizeAction permissionHand the authorized workload to the configured customer or partner execution system at the selected destination. The supported action and integration are defined for the deployment.05ExecuteApproved destinationRecord available usage and cost evidence against the workload identity and its selected destination. Accounting coverage depends on the configured sources.06RecordUsage + GPU costAI Factory is a possible placement plan only after policy and capacity checks. Execution still requires authorization.AI FactoryCentral computeRegional Grid is a possible placement plan only after policy and capacity checks. Execution still requires authorization.Regional GridPermitted regionEdge-AI Hub is a possible placement plan only after policy and capacity checks. Execution still requires authorization.Edge-AI HubLocal environmentIllustrative proposed exception: defer the workload for operator review. No authorization, execution or silent fallback. Confirm this handling in the deployment scope.Defer / reviewNo executiongFabric / Workload placementPlace a workload under policyA workload needs capacity, but its identity, data rules and operating requirements limit where it can run. Apply policy before comparing permitted destinations, check capacity, then authorize the selected placement plan. Execute through configured tools and record usage afterwards. Execution requires authorization and configured customer or partner tooling. Defer / review is an illustrative proposed exception flow, subject to deployment scope. Policy filters candidate destinations before placement selection. The branches are placement plans, not workload execution. An AI Factory, regional Grid or Edge-AI Hub is selected as an alternative, not a mandatory hop. The no eligible destination scenario stops at a proposed defer / review step. gFabric governs; configured customer or partner tools execute. Animation repeats the illustration, not a workload.Eligible placement plansSelection only · No executionDefer / review: proposed exceptionIllustrative sequence · Repeatsidentify to policypolicy to capacityauthorize to executeexecute to recordPlan AI Factory placement after policy eligibility and capacity checks. No workload executes here.AI Factory placement plan requires action authorization before execution.Plan Regional Grid placement after policy eligibility and capacity checks. No workload executes here.Regional Grid placement plan requires action authorization before execution.Plan Edge-AI Hub placement after policy eligibility and capacity checks. No workload executes here.Edge-AI Hub placement plan requires action authorization before execution.No eligible destination: illustrative proposed defer / review flow. No execution.Establish the workload identity, model needs and operating requirements, including latency, data location and resource demand.01IdentifyIdentity + requirementsApply configured identity, access and data rules to exclude destinations before selection. Available capacity does not override those rules.02PolicyData + access rulesCompare capacity and operating constraints only within the policy-permitted set. Use configured resource signals to form a placement plan for an eligible AI Factory, regional Grid or Edge-AI Hub.03CapacityEligible resourcesAuthorize the selected placement plan under the configured action policy. Selecting an eligible destination alone does not permit execution.04AuthorizeAction permissionHand the authorized workload to the configured customer or partner execution system at the selected destination. The supported action and integration are defined for the deployment.05ExecuteApproved destinationRecord available usage and cost evidence against the workload identity and its selected destination. Accounting coverage depends on the configured sources.06RecordUsage + GPU costAI Factory is a possible placement plan only after policy and capacity checks. Execution still requires authorization.AI FactoryCentral computeRegional Grid is a possible placement plan only after policy and capacity checks. Execution still requires authorization.Regional GridPermitted regionEdge-AI Hub is a possible placement plan only after policy and capacity checks. Execution still requires authorization.Edge-AI HubLocal environmentIllustrative proposed exception: defer the workload for operator review. No authorization, execution or silent fallback. Confirm this handling in the deployment scope.Defer / reviewNo executiongFabric / Workload placement

An eligible local plan. Authorization before execution.

RequirementsSelectionAuthorized workDefer / review

Execution requires authorization and configured customer or partner tooling. Defer / review is an illustrative proposed exception flow, subject to deployment scope.

Policy permits local processing and suitable capacity is available at the Edge-AI Hub. The local placement plan passes the action gate, the configured executor runs the workload there, and usage is recorded. The workload does not need to traverse the regional Grid or AI Factory.

gFabricIdentity, policy, placement & usageiFabricResource evidence where integrated
Customer / partnerExecution tools & exception handling

Where it applies

AI Factory · AI Grid · Edge

An AI workload needs execution capacity.

Measures to define

  • Policy-conforming placements
  • Time to eligible capacity
  • Deferred workload rate
  • Attributed usage & cost
Workflow & operating boundaries

A workload needs capacity, but its identity, data rules and operating requirements limit where it can run. Apply policy before comparing permitted destinations, check capacity, then authorize the selected placement plan. Execute through configured tools and record usage afterwards.

  1. Identify. Establish the workload identity, model needs and operating requirements, including latency, data location and resource demand.
  2. Policy. Apply configured identity, access and data rules to exclude destinations before selection. Available capacity does not override those rules.
  3. Capacity. Compare capacity and operating constraints only within the policy-permitted set. Use configured resource signals to form a placement plan for an eligible AI Factory, regional Grid or Edge-AI Hub.
  4. Authorize. Authorize the selected placement plan under the configured action policy. Selecting an eligible destination alone does not permit execution.
  5. Execute. Hand the authorized workload to the configured customer or partner execution system at the selected destination. The supported action and integration are defined for the deployment.
  6. Record. Record available usage and cost evidence against the workload identity and its selected destination. Accounting coverage depends on the configured sources.

This is an illustrative operating workflow, not a customer case study or live deployment. Define identity sources, data boundaries, eligible environments, capacity signals and action permissions before enabling execution. The destination branches show placement plans; they do not send a workload before authorization. AI Factory, regional Grid and Edge-AI Hub are alternatives, not mandatory serial hops. Actual placement and usage coverage depend on supported integrations. Defer / review is a proposed exception flow to scope with the customer, not confirmed native behavior. Animation timing is illustrative. Repetition restarts the illustration, not a workload.

Scenarios

Local processing. Policy permits local processing and suitable capacity is available at the Edge-AI Hub. The local placement plan passes the action gate, the configured executor runs the workload there, and usage is recorded. The workload does not need to traverse the regional Grid or AI Factory.

Approved regional capacity. The workload's configured data rules permit the required data to cross to an eligible region. Capacity checks select a regional Grid plan. Only after action authorization does the configured executor run it there and supply usage evidence. Regional capacity cannot bypass the data boundary.

No eligible destination. No destination satisfies both the configured policy and capacity requirements. This illustrative proposed exception flow defers the workload for operator review, with no authorization or execution and no silent fallback across the data boundary. Confirm exception handling in the deployment scope; this is not a claim of confirmed native product behavior.