DeepSolar Project, using machine learning and 1B+ satellite images, claims it found 1.47M solar installations in the US, much more than commonly cited estimates
Meghan Brown / ENGINEERING.com :
Context & Ripple Effects
DeepSolar's claim is a census story, not a hardware story: by training machine learning on 1B+ satellite images, it counts rooftop solar directly instead of relying on the commonly cited estimates that understate adoption at 1.47M installations. The method matters because distributed solar is exactly the kind of infrastructure that official surveys miss panel by panel.
The finding lands in the middle of a software-for-solar arc. Days later, Aurora Solar raised a $20M Series A to apply lidar and computer vision to panel layout and installation design, and by 2021 the same company had raised $250M at a $2B valuation on software that sizes installations per property — a market that only exists if the installed base is large and growing, which is precisely what DeepSolar's census argues.
First-order effects
- Analysts, utilities, and policymakers tracking US solar adoption now face a higher baseline: the commonly cited installation counts they build forecasts and interconnection plans on appear to undercount the actual fleet by a wide margin.
- The solar industry gains a measurement tool it lacked — a satellite-derived census that can be re-run to track adoption without waiting for slow survey cycles.
Second-order effects
- Software vendors in the solar design chain, most directly Aurora Solar, see their addressable market expand with every undercounted installation, reinforcing the investor thesis behind its $250M raise at a $2B valuation.
- Satellite imagery and computer vision become a competitive measurement layer for energy infrastructure, pressuring survey-based estimate providers and opening adjacent counting markets (storage, EV charging, other rooftop equipment) to the same technique.
Third-order effects
- If official estimates systematically undercount distributed generation, grid planning, net-metering policy, and utility rate design built on those estimates are mispricing rooftop solar's scale — a structural argument for satellite-verified data in regulatory proceedings.
- The pattern points toward energy infrastructure being measured from orbit rather than from filings, a shift that extends from counting rooftop panels to sourcing power itself, as in Meta's deal for up to 1GW of beamed space solar — solar's frontier moving beyond what any ground census can see.
The trend: Computer vision over satellite imagery is replacing survey-based estimates as the way the energy industry measures distributed infrastructure, making solar's true scale visible and investable.