Microsoft Research Releases Open Dataset
- Grid Connection & Reliability Policy & Markets Research
Marin Holt
- 0
- 7 minutes read
- Models Released: 54 (48 single-state + 6 multi-state)
- Eastern Interconnection Buses: 21,697 buses
- Transmission Corridors Identified: 31,488
- States Covered: 48
- Validation Method: AC optimal power flow solvability
Microsoft Research released an open transmission-level power grid dataset on May 8, 2026, covering 48 US states and six multi-state interconnections. The dataset includes a detailed model of the Eastern Interconnection with 21,697 buses. There are 54 grid models in total – 48 single-state and 6 multi-state – published on GitHub and Hugging Face. These models were derived from open-source data without proprietary inputs.
The Microsoft Research team developed a pipeline that combines OpenStreetMap geographic data with US Energy Information Administration energy statistics and Census demographic data. This pipeline creates electrically coherent grid topologies. The team identified 31,488 distinct transmission corridors, with 27,506 carrying single circuits and approximately 4,000 carrying multiple circuits. Each model passes AC optimal power flow solvability for both peak and off-peak demand.
Getting here required working around a structural barrier: the Critical Energy Infrastructure Information designation and utility NDAs have blocked public access to actual US transmission planning models for decades, forcing researchers to use small synthetic test cases that rarely exceed a few hundred buses. The Microsoft team’s contribution is solving the at-scale problem without touching restricted data, constructing coherent topologies entirely from OpenStreetMap, EIA statistics, and Census demographics. The AC-OPF solvability check at peak and off-peak demand is the validation hurdle that previous open attempts couldn’t clear at continental scale. Microsoft’s GridSFM follow-on work is already using these grids to train a surrogate model that returns a full AC-OPF operating point in milliseconds — faster by orders of magnitude than conventional iterative solvers.
The dataset has limitations, including incomplete OpenStreetMap line geometry in remote regions. The released models are realistic approximations rather than exact copies of the actual grid. Line ratings, transformer impedances, and protective relay settings are inferred from public information instead of utility planning data. AC-OPF convergence at planning scale is one validation, but it does not cover transient stability or other effects.
No prior open dataset has covered all 48 continental states with a single consistent derivation pipeline — earlier efforts stalled at regional scale or relied on heterogeneous methodologies that made cross-region comparisons unreliable. The dataset’s longevity now depends on two things: whether Microsoft Research sustains the engineering effort to incorporate new EIA data as generation mix and topology change, and whether any utilities quietly cooperate to patch the worst geometry gaps in remote regions without triggering CEII disclosure requirements. The GitHub license terms are still the gating factor for commercial applications; consultants and planning vendors will be watching that decision closely before committing roadmap resources to models built on this foundation.
Critical Perspective
The open dataset’s accuracy and utility depend on cooperation from utilities to provide real-time data updates. Validating these models via AC optimal power flow solvability is crucial but not sufficient without this data. According to Watt-Logic, maintaining accurate grid models requires continuous validation with actual operational data, which many utilities are reluctant to share openly. If utilities do not provide updates, the model’s predictive power could diminish rapidly, leading to misinformed research conclusions and flawed policy decisions.
Sources
- https://www.microsoft.com/en-us/research/blog/building-realistic-electric-transmission-grid-dataset-at-scale-a-pipeline-from-open-dataset/
- https://www.startuphub.ai/ai-news/ai-research/2026/microsoft-builds-open
Why It Matters
The release of Microsoft Research’s open transmission-level power grid dataset has significant implications for the US energy sector, particularly for utilities, ERCOT, PJM, and CAISO operators. This dataset fills a critical gap in AI/ML grid research, enabling more accurate and realistic modeling of the US transmission grid. With 54 grid models covering 48 US states and six multi-state interconnections, this dataset has the potential to inform policy decisions and improve grid operations. The dataset’s impact extends to the data-center demand and BESS economics, as it can help optimize energy storage and transmission planning. Furthermore, the release of this dataset may influence FERC orders and IEEE/NERC standards, as it provides a more accurate representation of the US transmission grid. The open question remains whether utilities will cooperate to refine these models without publishing their own restricted data, which could determine the dataset’s long-term utility and commercial use eligibility. As the energy sector continues to evolve, the availability of accurate and realistic grid models will become increasingly important. The Microsoft Research dataset has the potential to drive innovation and improvement in grid operations, and its impact will be closely watched by industry stakeholders.Related Coverage
Array
