Infrastructure Permitting at the Speed of the Moment

I grew up in factories. My parents ran an import business, and at the dinner table they talked about shipping containers, customs paperwork, and the USD-to-BRL exchange rate — which moved overnight and changed their bottom line. Infrastructure has been the language I've thought in for as long as I can remember.

The world is now in the middle of one of the largest infrastructure buildouts in human history. Driven by AI, electrification, and industrial reshoring. We are trying to build the physical layer that the next century of compute and energy will run on. We are not building it fast enough.

The Empire State Building went up in 410 days. The Hoover Dam in five years. A project's interconnection queue averages five. A high-voltage transmission line today takes thirteen. We used to build at the speed our ambition demanded. We don't anymore.

The constraint is not capital. ACP estimates $10–18M of extra cost on a single 100 MW solar project from permitting delays alone. Data Center Watch has tracked $64B in data center projects blocked or delayed over the last two years. The money is there. The demand is there. The constraint is the system that turns intent into the right to build — and it was designed for a slower era.

How electrical infrastructure is permitted

The United States has roughly 40,000 AHJs (Authorities Having Jurisdiction) across federal, state, county, and local layers. Each one has its own rules, its own process, its own approval chain. Zoning ordinances. Battery storage ordinances. Fire codes. Conditional use permits. Stormwater. FEMA. CEQA. FAA. FERC. PUC.

Step one is figuring out who the applicable authorities are. Then figuring out what their requirements are — thousands of pages of them. Then drafting paperwork to meet those requirements. Then showing up to defend your project.

Delays cost millions per month; tens of millions on the largest projects. The two biggest causes today are power and permitting. And this won't get any easier.

Historically, permitting has been done through human labor: local counsel, experts, lawyers, professionals who have spent their careers on this work. But the work doesn't compound or scale with the buildout. And nothing updates when an ordinance changes overnight. Under deadline pressure, humans miss requirements that block projects, or over-flag requirements that don't apply.

What is now possible

Someone once told me there was no way to actually map out this patchwork of jurisdictions. I thought, why not?

But even in 2024, the technology wasn't quite there. We tried RAG, classification, chunk tagging — every retrieval pattern in the book. None of it held up to the structure of these documents.

Now we can process hundreds of thousands of tokens in one shot — an entire piece of regulation loaded into context.

But models are still not good at knowing what permitting professionals expect to see. The structure. The citation density. The jurisdictional quirks worth surfacing. Without the right scaffolding, the output misses real requirements and adds noise where the expert needs precision. In permitting, missing one requirement can cost six months.

The missing step is codifying the judgment of local counsel, permitting experts, and independent engineers — turning their tacit knowledge into structure the model can act on. Human taste, machine-level precision, and the scale to keep up.

What doing the work taught us

The thing you notice from doing this work by hand, and from speaking to hundreds of practitioners, is how much of it is intelligence that should compound.

Finding the ordinances. Reconciling federal, state, county, and local layers. Different jurisdictions, same shape of work. We set out to do the intelligence layer once and well, so the people who know permitting can spend their time on what only they can do.

How the system delivers today

Today, Spark delivers a permit matrix in days, not weeks, at a fraction of what the same engagement costs from traditional methods. We've delivered nearly 200 in the past year — work that would have cost ~$1M at consultant rates.

The model produces the first pass. A permitting expert reviews every line before it ships. The project developer gets a document they can act on, with every requirement cited to source. And we sit down with developers afterward, to talk through which agencies will be sticklers, where they have existing relationships, how sequencing might play out on the ground. The conversation moves to where it should have been all along. And the system improves with use. Every ordinance ingested is durable across every project that ever touches that jurisdiction.

Instead of grinding through hours of research, experts spend their time flagging the agencies that will block a project, surfacing the political read on a room, talking about sequencing. We're not replacing any of the human work that matters. We're giving back the hours they should always have been spending on the parts only they can do.

What this unlocks

When identifying permit requirements and pathways is scalable, when diligence catches what's wrong with a project before it costs the developer a quarter, when drafting permits becomes a system instead of a service — permitting collapses from a 12-month gate to a 12-week one.

The infrastructure of the AI era — the data centers, the generation, the transmission, the storage — gets built at the speed the moment demands. The bigger shift is that the people building got the right tools at the right time.

If you're building energy or compute infrastructure and the permitting layer is in the way, reach out — we'd love to hear about your project.

Julia