AI × Energy
Unclogging the grid queue
AI-native interconnection engineering, which takes power-flow studies from years to weeks.
Thesis
The bottleneck for clean energy and data-center power in the US is no longer panels or turbines. It’s paperwork plus physics. Projects wait years in interconnection queues while engineers run power-flow and stability studies largely by hand. Roughly 2,600 GW of generation and storage sat in US queues at the end of 2023, more than the entire existing fleet TODO: verify, LBNL “Queued Up”.
A company that can run these studies faster and more reliably, for developers first and utilities later, sits right at the narrowest point of the energy transition.
Why now
- FERC Order 2023 pushed ISOs toward cluster studies and firm deadlines, so speed is now required TODO: verify current implementation status.
- AI data-center demand makes “time to power” a board-level question for the largest buyers in the world.
- Simulation plus ML surrogate models can pre-screen thousands of configurations before a full study runs.
Why it’s big
Every gigawatt delayed is lost revenue for developers and lost capacity for the grid. Whoever owns the study workflow can extend into siting, permitting and grid-upgrade planning.
What has to be true
- AI-accelerated studies match the accuracy of incumbent tools (PSS/E-class) on historical cases.
- Developers will pay for pre-screening on their own, before any utility acceptance.
- The founding team has real power-systems engineers, not just ML people.
Biggest risk
Long, conservative utility sales cycles, and the chance that the true constraint is transformers and transmission, not studies.
How I’d diligence it
Interview three ISO planning engineers and five developers about which delays come from studies and which from upgrades. If studies aren’t at least ~30% of the delay, I’d pass.