Microsoft Opens Up Power Grid Data to Boost Energy AI

Microsoft Research announced on Friday, May 8, 2026, the release of an open-source dataset that maps the transmission grid topology of the United States. This initiative aims to eliminate one of the biggest hurdles in energy systems research: the lack of realistic and accessible data for training artificial intelligence models.

Addressing the Shortage of Critical Data

Traditionally, information about electrical infrastructure is classified as critical data, protected by national security secrecy. This has forced researchers to use simplified test networks or synthetic models that do not reflect the system's real complexity. Microsoft's new pipeline uses public data from OpenStreetMap and energy statistics to build geographically accurate and electrically coherent models.

According to an official blog post by Microsoft Research, the dataset covers 48 US states and includes interconnections ranging from 11 to over 20,000 buses. The ability to simulate power flow on a continental scale allows engineers to study grid congestion and system resilience amid new demand peaks.

AI and the Future of Data Centers

The release comes at a time when the generative AI industry is facing an energy hunger crisis. With new models requiring gigawatts of power, knowing exactly where the grid has spare capacity is vital for placing new data centers. The dataset enables analyses of Optimal Power Flow (AC-OPF), revealing physical margins that market price data cannot show.

In addition to the dataset, Microsoft has made available GridSFM, a machine learning-based substitute that can predict grid operating points in milliseconds. This combination of open data and neural models paves the way for other companies and research centers to develop solutions for autonomously managing electrical loads, reducing wait times for connecting new AI infrastructures.

The code and data are already available on GitHub and Hugging Face by Microsoft.