By Aurora SolarLearning Objectives
- Understand how Aurora Solar lets installers design a residential solar system remotely, without ever visiting the roof.
- Separate the genuine machine learning in the product from the physics simulation that the marketing also labels AI.
- Evaluate where Aurora Solar fits against drone-based and free satellite-only competitors, and what the current solar market headwinds mean for it.
What Is Aurora Solar?
Aurora Solar is cloud software for designing, selling, and proposing residential solar installations. Give it a street address and it builds a three-dimensional model of the home's roof from high-definition aerial imagery and LIDAR, simulates shading and energy production across the year, and produces permit-quality engineering designs plus customer-ready proposals. The problem it solves is a logistical one: before remote design tools existed, quoting a rooftop system meant sending a person with a ladder and a shade-measurement instrument to every prospective home, which made the sales funnel slow and expensive.
The company was founded in 2013 and is headquartered in San Francisco. Founders Christopher Hopper and Samuel Adeyemo met at Stanford business school, and the origin story is instructive: a 2011 solar installation on a school in Kenya took two weeks to physically build but seven months to plan from a distance. Aurora Solar has raised more than 500 million dollars in total private funding across several rounds. It reached unicorn status on a 250 million dollar Series C in May 2021, followed by a landmark 200 million dollar Series D in February 2022 co-led by Coatue and Energize Ventures. The company remains private with no initial public offering.
💡Key Concept
The machine learning makes it fast; the ray tracer makes it right. Aurora Solar markets a lot of capability under the AI umbrella, but only one part is a learned model. Automated detection of roof planes, obstructions, and trees from aerial imagery and LIDAR is genuine computer vision and machine learning — Aurora describes it as proprietary, and it generates a full three-dimensional site model from an address in under about 10 seconds, improved from roughly 30 seconds. That learned perception layer is what makes remote design possible at all. But the shading and irradiance analysis everyone credits for Aurora's accuracy is deterministic physics: the software simulates the sun's path for every daylight hour of the year, runs a ray-tracing algorithm to compute irradiance at each point on the roof, and then varies each module's electrical characteristics by the irradiance falling on it. No training data, no learned weights — just optics and geometry. Understanding this split is the difference between evaluating the product honestly and buying the brochure.
✅Tip
Visit Aurora Solar: aurorasolar.com — pricing is quote-only, so expect a sales conversation rather than a self-serve signup.
Core Capabilities
Aurora AI — automated three-dimensional model generation
The headline feature and the genuine machine-learning core. From an address and a single utility bill, Aurora AI detects the roof planes, identifies obstructions such as vents and chimneys, maps surrounding trees, and assembles a proposal-ready design. Aurora reports it has been run more than 1.6 million times.
Design Mode — permit-quality engineering
The full engineering environment, where a designer refines the automated model into a plan set that a jurisdiction will accept. This is where the module-level simulation engine runs, computing per-panel production from ray-traced irradiance rather than a rule of thumb.
Sales Mode — remote proposal and quote generation
A lighter-weight surface for sales reps to build and send a proposal on a call, including production estimates, savings projections, and financing options. This is automation and templating rather than machine learning, but it is the workflow that most seats actually live in.
HelioScope — commercial and industrial design
Aurora's separate product line for larger commercial and industrial arrays, with the layout and stringing tooling those projects require. It is a distinct product, not a mode inside the residential tool.
Supporting products
Contract Manager handles the paperwork stage. Expert Design Services is a human-in-the-loop offering where Aurora's own staff produce outsourced plan sets. Financing integrations connect proposals to lease and loan providers.
Strengths
- The strongest shading simulation in United States residential design software, grounded in hourly sun-path modeling and ray-traced irradiance rather than approximation.
- Genuinely fast remote modeling — a usable three-dimensional site model in about 10 seconds removes the site visit from the front of the sales funnel.
- The deepest design-to-proposal workflow available: the same model carries from automated detection through engineering plan set to customer quote without re-entry.
- Real external validation exists in at least one case — the Connecticut Green Bank accepted Aurora's remote shade reports in place of an on-site inspection for rebate approval.
- Vendor-claimed scale is substantial: Aurora reports more than 7,000 companies served and more than 20 million project designs processed. Treat these as company figures, not audited ones.
- HelioScope gives the company a credible answer in commercial and industrial work, which most residential-first competitors lack.
Limitations and Considerations
- The "NREL-validated" claim is doing heavy lifting on old work. The National Renewable Energy Laboratory validations date to 2015 and 2017, and they tested Envision — a discontinued predecessor product that measured roofs from street-level photographs, not today's imagery-and-LIDAR pipeline. The 2017 study covered 15 roofs in a single metropolitan area. That is a small sample, one climate, roughly a decade old, and it predates the machine-learning product entirely. The validation is real; it just does not describe the software you would buy today.
- Physics is not machine learning. The accuracy comes from ray-traced simulation and good input data. The learned models mainly buy speed. If you are evaluating Aurora Solar as an AI purchase, be clear about which half you are paying for.
- Output quality is bounded by imagery quality and freshness. Remote modeling inherits whatever the aerial data captured — a stale or low-resolution image can miss a recently added dormer, a new air-conditioning unit, or several years of tree growth. Drone-photogrammetry rivals claim more accurate on-site models for exactly this reason. Users also report that production calculations tend to run conservative.
- Pricing is opaque and widely criticized. Nothing is published; every deal is quote-only. Secondary sources report figures around 159 to 259 dollars per user per month, plus a credit-based consumption model that can push a high-volume month well past the sticker price. Treat those numbers as unconfirmed reporting, not vendor pricing.
- Severe industry headwinds are the dominant story. The United States residential solar tax credit for homeowner-purchased systems expired at the end of 2025, and California's net-metering changes cut export compensation sharply. Aurora cut staff three times starting in 2024. Co-founder Christopher Hopper became executive chairman in October 2025, with Charlie Herche taking over as chief executive. The 2022 funding mark predates all of this, which is why no current valuation should be quoted. The industry is shifting toward third-party-owned lease and power-purchase models, and Aurora is shipping financing integrations to follow it.
- There are no large language model or agent features. Whatever "AI" means in this product, it means computer vision. Do not expect a chat interface or an autonomous design agent.
| Capability | Machine learning or physics? | What that means for you |
|---|---|---|
| Roof plane, obstruction, and tree detection | Machine learning (computer vision) | Learned from imagery and LIDAR — fast, and occasionally wrong in ways a human must catch |
| Shading and irradiance analysis | Physics (hourly sun path plus ray tracing) | Deterministic and accurate, bounded by input data quality, not by training data |
| Weather and climate inputs | Neither — lookup datasets | Reference data, not a learned or predictive model |
| LIDAR elevation data | Neither — a sensor data source | Raw input that feeds the vision model; not itself AI |
| Sales proposal generation | Neither — automation and templating | Workflow speed, no intelligence claim to evaluate |
Best Use Cases
| Task | Why Aurora Solar |
|---|---|
| Quoting a residential rooftop without a site visit | The vision model returns a usable three-dimensional roof from an address in about 10 seconds, removing the truck roll from the top of the funnel |
| Shade analysis on a complex or heavily treed roof | Hourly sun-path plus ray-traced irradiance is the strongest simulation in the residential category, and at least one green bank accepts the reports in place of inspection |
| Producing a permit-quality plan set | Design Mode carries the same model from automated detection through to a jurisdiction-ready engineering document |
| Commercial and industrial array layout | HelioScope is a purpose-built product line rather than a residential tool stretched past its range |
Getting Started
- Request a quote through aurorasolar.com — there is no self-serve signup, and pricing depends on seat count and design volume. Ask specifically how the credit-based consumption model behaves in your busiest month.
- Run Aurora AI on a handful of roofs you have already surveyed in person. Compare the automated roof planes, obstructions, and tree map against what you measured on-site. This tells you how good the aerial imagery is in your service territory, which is the single biggest determinant of output quality.
- Take one of those designs all the way through Design Mode to a permit-quality plan set, and check the production estimate against a system you have already installed and metered. Expect the number to run somewhat conservative.
- If your work is mostly commercial or industrial, evaluate HelioScope separately rather than assuming the residential experience transfers.
Key Takeaways
- Aurora Solar is the incumbent premium standard for United States residential solar design, and its moat is the depth of the design-to-proposal workflow plus the quality of its shading simulation.
- The genuine artificial intelligence is a computer vision model that detects roof planes, obstructions, and trees from aerial imagery and LIDAR in about 10 seconds. That perception layer is what makes remote design possible.
- The accuracy Aurora is famous for is not machine learning. It is hourly sun-path modeling and a ray-tracing algorithm — deterministic physics, bounded by input data rather than training data. The machine learning makes it fast; the ray tracer makes it right.
- The NREL validation everyone cites tested Envision, a discontinued street-photo predecessor, on 15 roofs in one metropolitan area in 2017. It says nothing about the current pipeline.
- Competitive pressure is real: OpenSolar is free but satellite-only with weaker shade analysis, Scanifly claims more accurate on-site models via drone photogrammetry, and PVcase owns large-scale ground-mount terrain design. Combine that with the expired residential tax credit, California's net-metering changes, three rounds of layoffs, and a 2025 chief-executive transition, and Aurora is a strong product operating in a contracting market.
- The broader lesson for renewables: the genuine AI is perception and prediction — the physics engines are what deliver the accuracy.
