Based on survey of academic papers (2022–2026), vendor publications, and production deployments.
Adoption Ranking
| Architecture |
Adoption |
Primary Telecom Use Cases |
| GCN |
★★★★★ |
Fault diagnosis, anomaly detection, RCA, resource allocation, interference management. Default baseline in most papers. |
| GAT |
★★★★★ |
Alarm correlation, traffic prediction, handover optimization, network slicing. Attention mechanism differentiates neighbor importance. |
| MPNN |
★★★★☆ |
Network performance modeling (RouteNet family), scheduling, routing. Theoretical foundation for most telecom GNNs. |
| GNN + DRL |
★★★★☆ |
Resource allocation, power control, O-RAN xApps, connection management. De facto standard for optimization. |
| GNN + Transformer |
★★★★☆ |
Spatio-temporal fault detection, RCA, KPI forecasting. Fastest-growing category. |
| GraphSAGE |
★★★☆☆ |
Large-scale anomaly detection, xApp conflict detection, intrusion detection. Sampling enables scalability. |
| GAE/VGAE |
★★★☆☆ |
Link prediction (handover), unsupervised anomaly detection, topology reconstruction. |
| GIN |
★★☆☆☆ |
Network authentication, malicious node detection. More expressive but less adopted in telecom. |
Key Telco-Specific Models
RouteNet Family (UPC Barcelona — most influential)
| Model |
Year |
What It Does |
| RouteNet |
2019 |
Foundational MPNN for per-flow delay, jitter, packet loss prediction |
| RouteNet-Fermi |
2023 |
State-of-the-art: 3-stage MPNN, handles diverse scheduling + traffic |
| RouteNet-TGNN |
2024 |
Temporal extension — 27.5% better delay prediction |
| QT-RouteNet |
2022 |
Queueing theory + GNN for 5G scalability |
| GNNetSlice |
2025 |
GNN for B5G network slicing |
Vendor Models
| Model |
Vendor |
Architecture |
Application |
| Simba |
Ericsson |
GCN + Transformer |
RCA of concurrent 5G RAN faults (F1 ~0.8+) |
| KE-GNN |
Huawei |
GCN + Knowledge Rules |
5G fault scenario ID (99.15% accuracy) |
| EEPC-GNN |
Nokia Bell Labs |
Bigraph GNN |
Energy-efficient power control in cell-free MIMO |
| GraphIQ |
NetAI + Google Cloud |
Proprietary GNN |
Commercial GNN-native AIOps for deterministic RCA |
| CausalGNN-Net |
Academic |
GATv2 + Transformer |
Causal discovery from irregular telecom alarms |
Dominant Pattern per Use Case
| Use Case |
Dominant GNN |
Production Leader |
| Root Cause Analysis |
GCN + Transformer |
NetAI GraphIQ |
| Alarm Correlation |
GAT / Propagation MPNN |
NetAI / Ericsson |
| Network Performance |
MPNN (RouteNet) |
UPC Barcelona |
| O-RAN xApps |
GNN + DRL |
Intel xApp |
| Power Control |
Bigraph GNN / GCN |
Nokia Bell Labs |
| Network Slicing |
GCN + DRL |
Google Cloud |
| Handover/Mobility |
VGAE |
O-RAN research |