Why we're building Gridsurf.

U.S. load growth is outrunning the grid. Major ISOs revised their long-term forecasts up by tens of gigawatts in two years, mostly from data centers and electrification. Every major ISO/RTO is now planning around demand spikes nobody saw coming twelve months ago.

The generation to meet that demand exists. Roughly 2,600 GW of projects sit in US interconnection queues today. Average wait times exceed four years. Completion rates hover around 20%. The plant that should be online in 2026 is still in a study that started in 2022. Multiply that across seven ISOs/RTOs and you have the largest infrastructure bottleneck of the AI era.

The Interconnection data exists. Every ISO publishes queue updates, cluster studies, transmission upgrade allocations, FERC filings, and stakeholder meeting transcripts every week. None of it is built for AI agents.

Cost allocations live in messy PDFs. Cluster results arrive as Excel attachments to meeting agendas. Stakeholder positions sit in 200-page transcripts. The information is public but reading it is the bottleneck.

Today, every developer, investor, and consultant tracking interconnection hires analysts to read it manually. The best teams dedicate headcount to it. The cost of being wrong is months of lost time and millions in stranded capital.

Gridsurf has three layers. A data foundation normalizes the queue: we turn unstructured ISO source files (queue reports, cluster studies, cost allocation sheets, change logs) into clean, machine-readable records with stable identifiers across every cycle.

A market intelligence layer keeps it current. New filings, cluster results, stakeholder meetings, and tariff changes flow in continuously and index against the foundation below.

An agent layer answers questions. Purpose-built AI agents read across the layers, run code in sandboxed environments, and return cited answers grounded in the underlying files.

Every answer cites its source. You can always follow a claim back to the file that settled it, and you can always rerun the analysis yourself.

Foundation models commoditize fast. The wrapper around them gets thinner every quarter. What stays valuable is context: what you know about the customer’s domain, how current that knowledge is, and how hard it would be for the next builder using the same APIs to replicate it.

We have spent months inside the workflow. We sit with generation developers when cluster results land. We watch consultants rebuild the same Excel sheet for a fifth client. We see how a single partner withdrawal cascades cost onto a 200 MW project that did nothing wrong. None of that is in a foundation model’s training data, and none of it surfaces in public datasets.

That immersion is the moat. The next person to clone our API surface will still need to clone our customers, our queue history, and our domain depth. Slow on purpose.

With our Agents, every interconnection diligence cycle that takes six weeks will take six hours. Every developer can simulate a project withdrawal cascade before the decision window ends. Every M&A buyer can price queue risk before signing a term sheet. Every site selector knows which substation has headroom before calling the utility.

Analysts will still be analysts. They will just spend less time reading PDFs and more time on the calls only they can make. If you are navigating through the interconnection process and cluster studies across the U.S. ISOs/RTOs, let’s chat.