Our Take
Microsoft solved the closed-data problem in power grid research by proving open sources can build electrically coherent models spanning 21,697 buses across the Eastern Interconnection.
Why it matters
Grid planners and AI researchers have been stuck with toy networks or restricted datasets when analyzing transmission capacity, datacenter siting, and infrastructure upgrades. Open models enable physics-based analysis without lengthy approval cycles.
Do this week
Infrastructure teams: Download Microsoft's grid dataset this week so you can run AC-OPF analysis on realistic transmission networks without data licensing delays.
Microsoft built continental-scale grid models from public data
Microsoft Research released an open dataset of US transmission grid topology derived entirely from publicly available sources (per Microsoft Research blog). The pipeline combines OpenStreetMap's physical infrastructure data with EIA energy statistics and Census demographic data to create geographically grounded models spanning 48 states.
The models range from 11-bus systems to the full Eastern Interconnection with 21,697 buses. All models pass alternating current optimal power flow (AC-OPF) analysis, the standard test for electrical coherence in power systems. The team validated the approach by successfully solving AC-OPF across the entire Eastern Interconnection spanning 36 states.
The pipeline identifies 31,488 distinct transmission corridors across the contiguous US. Most corridors (27,506) carry single circuits, while roughly 4,000 already carry multiple parallel circuits, with maximum density reaching ten circuits per corridor (company-reported).
Restricted grid data has blocked modern power research
Realistic transmission-level grid data is classified as critical infrastructure in the US, requiring lengthy approval cycles and non-redistribution agreements. Researchers have been forced to choose between small toy networks with dozens of buses or synthetic models that don't correspond to real infrastructure.
The data shortage particularly limits AI-based grid analysis methods, which require large volumes of physically plausible training data. Market price signals don't reveal how close transmission lines are to capacity limits or where new demand can be added without triggering congestion.
Microsoft demonstrated the models' practical value by analyzing where high-temperature superconducting cables could relieve Boston-area congestion, showing energy prices dropping 42% from $22.7/MWh to $13.1/MWh when bottlenecked generation becomes deliverable (company analysis).
Open models enable infrastructure planning without licensing
The dataset supports physics-based analysis of transmission expansion, datacenter siting, and capacity constraints using standard power flow tools. Models preserve geographic structure of real corridors while estimating electrical parameters from engineering references.
Microsoft tested datacenter placement scenarios by modeling a hypothetical 500MW facility at two Maryland locations with similar market conditions but different physical constraints. The analysis revealed how identical demand loads create different transmission impacts depending on network position.
The models are not exact replicas of operational grids and should not be used for real-time dispatch or market forecasting. Electrical parameters are estimated and parallel circuits approximated rather than exhaustively enumerated. The goal is structural realism for research and planning analysis, not operational precision.