# GPUaaS vs Armada Bridge Gap Matrix

As of: March 6, 2026

External reference:
- Armada Bridge product page: https://www.armada.ai/product/bridge
- Source content captured from product page on March 6, 2026.

Scope note:
- Armada claims are marketing-level statements from the product page.
- This matrix compares those claims to currently implemented and documented GPUaaS capabilities in this repository.
- Backlog IDs listed under "Closes via" are seed identifiers — create and track them as discrete work items.

---

## What Armada Bridge Claims

### Three Named Use Cases

| Use Case | Bridge Description |
|---|---|
| **AI Factory Orchestration** | Deploy + scale large AI workloads with dynamic optimization across GPU clusters |
| **GPU-as-a-Service** | Provision resources, monetize capacity, deliver self-service cloud for end users |
| **Platform-as-a-Service** | Run AI models via APIs and dashboards with customizable compliance controls |

### Two Deployment Models

| Model | Description |
|---|---|
| **Bridge + Galleon** | Turnkey: Armada's ruggedized hardware + Bridge software; production-ready in weeks |
| **Bridge on Your Infrastructure** | Deploy on existing GPU/DC infra; no vendor lock-in, full flexibility |

### Five Capability Pillars

1. Hard isolation for multi-tenant environments
2. Elastic resource allocation across clusters
3. GPU monetization and revenue generation
4. Unified billing and real-time observability
5. Air-gapped security, data sovereignty by default, regulatory compliance

### NVIDIA Certifications Claimed

- NVIDIA Cloud Partner (NCP)
- NVIDIA Spectrum-X (RoCE / AI fabric network)
- NVIDIA Quantum-2 (InfiniBand)
- NVIDIA Base Command Manager (cluster management)

### Target Operators (not end-users)

Data centers, land & power operators (monetizing underutilized assets), universities/research institutions, telcos/MSPs.

---

## Gap Analysis — By Bridge Pillar and Use Case

### Pillar 1: Hard Multi-Tenant Isolation

| | |
|---|---|
| **Bridge claim** | Network + workload-level hard isolation with multi-tenant security controls |
| **GPUaaS current** | Tenant/project ownership enforced at API layer; node isolation via `user-revoke` (OS user revoke on release, node stays enrolled) |
| **Evidence** | `doc/architecture/Tenant_Project_Ownership_Baseline.md`, `doc/architecture/Role_and_Policy_Lifecycle_Model.md`, `doc/architecture/db_schema_v1.sql` |
| **Gap** | `full-reimage` isolation path (MAAS) not yet implemented; no network-fabric segmentation (VLAN/IB) modeled at provisioning layer; IAM management UX not fully deployed end-to-end |
| **Closes via** | `IAM-PHASE2-UX-API-COMPLETE` (P0), `MAAS-INTEGRATION-NODE-LIFECYCLE-001` (P1), `NODE-LIFECYCLE-RETIRE-REACTIVATE-001` (P0) |

### Pillar 2: Elastic Resource Allocation

| | |
|---|---|
| **Bridge claim** | Cross-cluster placement, dynamic optimization |
| **GPUaaS current** | Single-pool `BareMetalAdapter`; allocation policy via `platform_policy_values` table; capacity projection in inventory service |
| **Evidence** | `doc/architecture/State_Machines.md`, allocation/provisioning flows in `doc/api/openapi.draft.yaml` |
| **Gap** | No `SchedulerAdapter` abstraction for multi-pool or priority-class placement; no auto-rebalancing; manual node lifecycle touchpoints |
| **Closes via** | `SCHEDULER-ABSTRACTION-MULTIPOOL-001` (P1), `NODE-LIFECYCLE-RETIRE-REACTIVATE-001` (P0) |

### Pillar 3: GPU Monetization + Revenue Generation

| | |
|---|---|
| **Bridge claim** | Monetize underutilized capacity; tenant billing and revenue reporting |
| **GPUaaS current** | Immutable ledger + Stripe payments + usage accrual + balance thresholds + policy-driven auto-release |
| **Evidence** | Billing/payments contracts in `doc/api/openapi.draft.yaml`, `doc/architecture/Seed_Data_Spec.md`, `scripts/seed.sql` |
| **Gap** | No per-tenant/project revenue reporting or operator billing dashboard; no prepaid/postpaid commitment support; no invoicing depth; project-level budget/quota controls missing |
| **Closes via** | `BILLING-ACCRUAL-CONFIDENCE-001` (P0), `BILLING-MONETIZATION-DEPTH-001` (P2) |

### Pillar 4: Unified Billing + Real-Time Observability

| | |
|---|---|
| **Bridge claim** | Unified cross-cluster view; real-time monitoring |
| **GPUaaS current** | OTel traces + Prometheus metrics + structured JSON logs + correlation-ID runbook path; ops admin pages for basic monitoring |
| **Evidence** | `doc/operations/Observability_Baseline.md`, `doc/operations/Ops_Runbook_Architecture.md`, `packages/web/app/admin/ops/page.tsx` |
| **Gap** | No standard dashboard pack; no tenant/project drill-down in metrics; no operator-facing billing diagnostics panel; alert coverage incomplete |
| **Closes via** | `OBS-DASHBOARD-PACK-001` (P0) |

### Pillar 5: Air-Gapped, Sovereign, Compliant

| | |
|---|---|
| **Bridge claim** | Air-gapped installs, data sovereignty by default, full regulatory compliance support |
| **GPUaaS current** | Local-dev docker-compose exists; no explicit deployment profile documentation; no compliance evidence packs |
| **Evidence** | `doc/operations/local-dev/` (local compose + Keycloak realm) |
| **Gap** | No documented deployment modes (connected / restricted-egress / air-gapped); no artifact mirror strategy; no KMS integration guidance for air-gapped environments; no compliance evidence pack |
| **Closes via** | `DEPLOYMENT-PROFILES-SOVEREIGN-001` (P1) |

---

### Use Case A: AI Factory Orchestration

| | |
|---|---|
| **Bridge claim** | Deploy/scale large AI workloads; dynamic optimization across clusters |
| **GPUaaS current** | App catalog + entitlement contracts (Phase 1 done); app instance runtime not yet built |
| **Evidence** | `doc/architecture/App_Control_Plane_v1.md`, IAM app entitlements UI |
| **Gap** | No app instance lifecycle (deploy/upgrade/rollback/decommission); no operator pipeline; no NVIDIA/AMD runtime conformance tests |
| **Closes via** | `APP-RUNTIME-INSTANCE-LIFECYCLE-001` (P1), `GPU-REFERENCE-PROFILES-CONFORMANCE-001` (P2) |

### Use Case B: GPU-as-a-Service (Self-Service + Operator Monetization)

| | |
|---|---|
| **Bridge claim** | End-user self-service provisioning AND operator-facing monetization controls |
| **GPUaaS current** | Marketplace → allocation → terminal flow (core implemented); basic admin billing views |
| **Evidence** | `doc/api/openapi.draft.yaml`, marketplace/allocations/admin contracts and UX |
| **Gap** | Operator revenue dashboard missing; auto-discovery and MAAS node onboarding not implemented; manual bootstrap friction present |
| **Closes via** | `BILLING-ACCRUAL-CONFIDENCE-001` (P0), `MAAS-INTEGRATION-NODE-LIFECYCLE-001` (P1), `BILLING-MONETIZATION-DEPTH-001` (P2) |

### Use Case C: Platform-as-a-Service

| | |
|---|---|
| **Bridge claim** | Run AI models via APIs and dashboards with customizable compliance |
| **GPUaaS current** | App entitlement foundation + service accounts — contracts in place; no hosted API-serving layer yet |
| **Evidence** | `doc/architecture/App_Control_Plane_v1.md`, service account model |
| **Gap** | No hosted AI model API layer; `policy_overrides` at project level exists in schema but runtime enforcement not implemented |
| **Closes via** | `APP-RUNTIME-INSTANCE-LIFECYCLE-001` (P1) |

---

### NVIDIA Certifications

| | |
|---|---|
| **Bridge claim** | NCP, Spectrum-X, Quantum-2, Base Command Manager certified |
| **GPUaaS current** | Hardware-aware node model exists; no certified integration with NVIDIA fabric components |
| **Evidence** | `doc/architecture/Node_Operations_and_Agent_Lifecycle_v1.md` |
| **Gap** | No validated reference profiles for NVIDIA (or AMD); no InfiniBand/Spectrum-X provisioning hooks; no BCM integration |
| **Closes via** | `GPU-REFERENCE-PROFILES-CONFORMANCE-001` (P2) |

### Deployment Models

| | |
|---|---|
| **Bridge claim** | Turnkey (Bridge + Galleon hardware) OR BYOI (bring your own infrastructure) |
| **GPUaaS current** | Single-node docker-compose for dev; no production install playbook |
| **Evidence** | `doc/operations/local-dev/` |
| **Gap** | No documented HA production topology; no install automation; no Galleon-equivalent hardware bundle or reference BOM |
| **Closes via** | `DEPLOYMENT-PROFILES-SOVEREIGN-001` (P1) |

---

## Priority Ranking (Competitive Urgency)

| Priority | Backlog ID | Gap Summary | Why Urgent |
|---|---|---|---|
| **P0** | `IAM-PHASE2-UX-API-COMPLETE` | IAM management UX completeness, role enforcement sweep | Demo-blocking; first thing operators check |
| **P0** | `BILLING-ACCRUAL-CONFIDENCE-001` | Usage accrual accuracy, release-stop billing boundaries | Revenue trust critical for operators |
| **P0** | `OBS-DASHBOARD-PACK-001` | Standard dashboards, alert coverage, tenant drill-down | Support-readiness; shows operational maturity |
| **P0** | `NODE-LIFECYCLE-RETIRE-REACTIVATE-001` | Explicit retire/reactivate/remove node states | Core reliability; drives elastic allocation claim |
| **P1** | `APP-RUNTIME-INSTANCE-LIFECYCLE-001` | App instance deploy/upgrade/rollback | Most visible PaaS gap vs Bridge |
| **P1** | `ENTERPRISE-ONBOARDING-OIDC-SAML-001` | Tenant federation self-service (OIDC/SAML), invite lifecycle | Sales-blocking for enterprise operators with SSO requirements |
| **P1** | `DEPLOYMENT-PROFILES-SOVEREIGN-001` | Connected / restricted-egress / air-gapped deployment modes | Required for regulated customer conversations |
| **P1** | `SCHEDULER-ABSTRACTION-MULTIPOOL-001` | SchedulerAdapter interface for multi-pool placement | Needed to credibly claim elastic cross-cluster allocation |
| **P2** | `BILLING-MONETIZATION-DEPTH-001` | Invoicing depth, prepaid/postpaid, project budgets/quotas | Revenue model completeness for commercial operators |
| **P2** | `GPU-REFERENCE-PROFILES-CONFORMANCE-001` | NVIDIA + AMD reference bootstrap/runtime/fabric checks | Required for formal NVIDIA partnership claims |

---

## Demo Positioning

### What We Can Credibly Claim Today (vs Bridge)

- Contract-first, multi-tenant GPU control plane with tenant/project ownership isolation at API layer
- End-to-end GPU provisioning (request → SSH terminal), billing, service accounts, app catalog + entitlement foundations
- Correlation-based observability workflow (error → logs → trace) and CI guardrails for reliability
- Single-node BYOI deployment via docker-compose (dev/test ready)
- IAM: platform roles, service account management, app entitlement management (Phase 1 foundations implemented; Phase 2 UX/API completion pending)

### What We Should Not Claim Yet

- Full AI Factory / PaaS orchestration maturity (app instance runtime not built)
- Turnkey sovereign / air-gapped deployment certification
- Formal NVIDIA fabric/certification alignment (NCP, Spectrum-X, Quantum-2, BCM)
- Cross-cluster elastic placement (single-pool only today)
- Operator revenue dashboard or invoicing depth

---

## Implementation Backlog Seeds

Create and track these as discrete work items:

| ID | Priority | Description |
|---|---|---|
| `IAM-PHASE2-UX-API-COMPLETE` | P0 | IAM management UX completeness — role enforcement sweep, membership admin, audit coverage |
| `BILLING-ACCRUAL-CONFIDENCE-001` | P0 | Usage accrual accuracy, release-stop billing boundaries, operator diagnostics panel |
| `OBS-DASHBOARD-PACK-001` | P0 | Standard dashboard pack (API, provisioning, billing, notifications, node fleet, auth); alert coverage; tenant/project drill-down |
| `NODE-LIFECYCLE-RETIRE-REACTIVATE-001` | P0 | Explicit node states for retire/reactivate/remove; agent behavior post-retire; placement defaults |
| `APP-RUNTIME-INSTANCE-LIFECYCLE-001` | P1 | App instance deploy/upgrade/rollback/decommission lifecycle; operator pipeline; service account identity integration |
| `ENTERPRISE-ONBOARDING-OIDC-SAML-001` | P1 | Tenant federation self-service (OIDC/SAML provider config), invite lifecycle, work/personal login UX |
| `DEPLOYMENT-PROFILES-SOVEREIGN-001` | P1 | Deployment modes: connected / restricted-egress / air-gapped; artifact mirror strategy; KMS integration guidance |
| `SCHEDULER-ABSTRACTION-MULTIPOOL-001` | P1 | SchedulerAdapter interface for multi-pool/priority-class placement; capacity-class policy keys |
| `MAAS-INTEGRATION-NODE-LIFECYCLE-001` | P1 | MaaS-backed node provisioning/deprovisioning/reactivation integration; lifecycle hooks for retire/reimage/re-enroll |
| `BILLING-MONETIZATION-DEPTH-001` | P2 | Invoicing depth, prepaid/postpaid commitment options, project-level budget/quota controls, exportable billing packages |
| `GPU-REFERENCE-PROFILES-CONFORMANCE-001` | P2 | NVIDIA + AMD validated reference profiles; InfiniBand/Spectrum-X provisioning hooks; BCM integration; conformance test matrix |