What it takes to get a data center built — and where software helps

When a developer builds a data center, they aren't just buying land and pouring concrete. They are getting permission from utilities, state regulators, county boards, federal environmental agencies, and air-quality boards. Each of those approvals has its own paperwork, deadlines, math, and history of past decisions across thousands of similar projects.

For decades, the only way to handle those approvals was to hire specialists — engineers, environmental attorneys, former agency staff — who had spent years learning how each piece worked. They charged by the hour because the work required human judgment that software couldn't replicate. The regulatory documents themselves were thousands of pages of unstructured text spread across dozens of agencies and 50 states. That's why the first three layers (where the data is structured and the workflow is repeatable) got software early, and the middle five did not.

Two things changed. First, AI language models in 2023 and after got good enough to read those unstructured documents and extract the relevant rules cheaply and accurately. Second, the post-2025 regulatory landscape introduced new rules — ERCOT Batch Zero, PJM PCLR, EPA Title V updates, state-level zoning changes like Loudoun County ending data-center by-right approval in March 2025 — that even the experienced consultants haven't fully digested yet. That combination finally makes some of the consultant-only work tractable as software, while leaving the parts that require political judgment and site-specific negotiation to the people who already do it well.