Beyond ChatGPT: Why Renewable Energy Needs Specialized AI

Picture this: A sprawling solar farm that could power 10,000 homes is ready to build—but it's trapped inside a 500-page binder. Welcome to the renewable energy industry, where clean energy generators live or die by paperwork.

Every solar, battery storage, or wind project needs dozens of permits and agreements before it can be built. It is also a million-dollar asset that undergoes some form of financial transaction, such as an acquisition or equity/debt financing. Let’s examine the components of this landscape, and the tools we have to navigate these “fat binders”.

The Regulation Overwhelm

Consider the scale: in the United States alone, there are over 3,000 counties, 10,000 building permit jurisdictions, and more than 2,000 energy-regulating entities. Each authority maintains its own set of rules. These agencies include:

• Federal Agencies (Federal Energy Regulatory Commission—FERC, North American Electric Reliability Corporation—NERC) regulate the reliability and security of the power system and the interstate transmission of electricity, natural gas, and oil.

• Regional Transmission Organizations (RTOs) set the interconnection rules, perform grid management, and operate wholesale electricity markets.

• State Utility Commissions (PUC) are state government agencies that regulate utilities.

• Local siting and zoning authorities, known as Authorities Having Jurisdiction (AHJs), enforce land use regulations, zoning and solar- or wind-specific ordinances, and permit requirements. They also hold town hall meetings with locals.

The numbers tell a striking story:

• 3,000+ counties, 10,000+ permit jurisdictions, 2,000+ energy regulators

• 11 GW of new utility-scale capacity in Q2 2024 alone (91% YoY growth)

• ~300,000 distributed generation projects nationwide

• 90M+ pages of project documentation (equivalent to 30,000 feet of stacked paper)

Before it becomes a living and breathing power generator, each project is a large set of documents, including:

• Building permits

• Electrical permits

• Power purchase agreements

• Interconnection agreements

• Environmental impact assessments

• Valuation reports

• Membership Interest Purchase Agreement (MIPA)

Approximately 30% of wind and solar siting applications submitted in the last five years were canceled, while about 50% experienced delays of 6 months or more (Berkeley Lab).

Recent initiatives address the overhead imposed by a patchwork of jurisdictions and regulatory agencies. States like California and New York have implemented comprehensive permitting reforms, with California’s AB 205 giving the California Energy Commission authority for a consolidated permitting process. However, permitting is just one component of the broader development lifecycle, which starts with site selection.

Vertical AI in Renewable Energy: From Weeks to Seconds

For the first time in history, we have the technology to index and organize all this information into formats that are easy to search and action upon in seconds instead of weeks or months. This technology is large language models (LLMs).

Understanding LLMs

Large Language Models are sophisticated machine learning models designed to process and understand human language. They transform words into "vector embeddings" that represent meanings and relationships. They can understand context and nuances across vast amounts of text and enable natural language interactions with complex documents.

The latest models, like Claude 3.5 Sonnet, demonstrate superior capabilities in visual reasoning and document analysis, graduate-level reasoning, and enhanced memory capacity.

LLMs work in the following way:

1. They break down text into smaller pieces (called tokens), like how we break sentences into words

2. They use special "attention mechanisms" that work like a digital spotlight to focus on important connections between words (similar to how humans focus on crucial parts of a conversation)

3. They learn from massive collections of text. Imagine reading millions of books and websites to understand how a language or academic theory works.

Beyond Generic AI: Industry-Specific Solutions

While general-purpose AI tools like ChatGPT have transformed how we interact with information, specialized sectors, including renewable energy, require vertical solutions. Here's what makes industry-specific AI platforms different:

• Specialized Data Pipelines: AI models are pre-trained on trillions of tokens but still lack the context that makes them useful for solar developers out-of-the-box. How long does it take to get permitted in Cook County, Illinois? What are the setbacks for ground-mounted solar in this town's Agricultural District? What are the setbacks, in feet, required for a utility-scale solar project in this municipality? All this information is locked inside PDFs and messy government websites. Before an LLM can begin to answer these questions, we need to scrape this data across thousands of jurisdictions, track changes across every geography, and compile them into a dataset.

• Industry-Specific Search and Retrieval: Not every AI retrieval system is built the same. Getting ChatGPT to answer every solar-adjacent question accurately 99% of the time requires its own solution. As a result, we should carefully measure the accuracy of such systems against our own documents—zoning ordinances, Power Purchase Agreements, Interconnection Agreements, etc.

• Enterprise-Grade Security: Some developers benefit from more rapid analysis of regulatory (public) and project-level (private) documents. When working with foundation model providers such as OpenAI and Anthropic, consider Data Processing Agreements and data retention agreements to ensure that your proprietary data is handled securely and privately.

• Features to enable the renewables developer: Integrating AI in renewable energy isn’t just about chatbots—it’s about presenting and packaging the data in ways that fit the user’s mental model. It’s about creating workflows that can handle the complex interplay of regulatory requirements, policy, finance, and market dynamics to drive operational efficiency. For instance, LLMs can create permit matrices, track county-level moratoria and opposition, integrate incentive programs, and more.

At Spark, we are building the vertical AI agent for the energy infrastructure of the future. If you’re interested in learning more, book time with us.

Product updates

1. SOC2 (Type I) certification: Spark is now SOC2-compliant, demonstrating robust systems to protect customer data in alignment with Trust Service Principles (Security, Availability, Confidentiality, Processing Integrity, and Privacy). See more in our Trust Center.

2. Reports: We have indexed 100k+ documents across 30 US states and significantly improved our results' relevance and speed.

   

3. Document Diligence: You can now download answers from document Q&A as a report and visualize citations. Our answer accuracy is now above 95%

   

Finally, Happy Thanksgiving! We’re grateful for our incredible partners—with a combined 12 GW+ pipeline and hundreds of millions in AUM—who trust Spark with their everyday work.

And thank you, reader, for being part of our journey to accelerate the deployment of a cleaner grid. ⚡️

Julia and Tae