AI Buildout Frontier

ai-buildout-frontier.vercel.app

A public county-level tracker of AI compute buildout against real power infrastructure, showing where infrastructure can realistically expand and where energy constraints become the limiting factor.

Project Details

While most research in the field of AI focuses on creating benchmarks, testing models, alignment, foresight, policy, and governance, there remains a clear gap in the empirical collection of information regarding the physical location of AI infrastructure and its interaction with the surrounding world—not only institutionally but also geographically. We lack sufficient data on how AI infrastructure is geographically distributed, how it affects local communities and the environment, where it is concentrated, and what the picture looks like at different scales—country, state, and county. These were precisely the questions I asked myself while working on my independent project—a geographical study of AI infrastructure. For example, The Epoch AI maintains datasets on GPU clusters and Frontier data centers, but these do not address their interaction with the outside world; they are simply published as-is. Semianalysis develops its own corporate tools, but they are very expensive and designed to solve business problems rather than serve the public good. Moreover, they view the issue at the national level, whereas increasingly heated debates and political battles are taking place at the subnational level (states, counties). The deployment of AI infrastructure and its geographical concentration are becoming important factors in politics and the economy. And I have seen firsthand just how little accessible and systematic data there is on this issue.

Even in the U.S., which has perhaps the most robust system of government statistics in the world, information has to be gathered bit by bit from hundreds of different websites, reports, and datasets. So I came up with the idea of creating this AI observatory, which would look at data at the state and county levels. Right now, I’m running this project on my own, in my spare time, and on a very limited budget. I actively use AI assistants to search for and process data, write scripts, and work on the web platform’s technical stack. I’ve developed a methodology for the Grid Constraint Index (GCI), a framework for data integration, and a web tool for visualizing and making the results publicly available. Currently, a minimum viable product (MVP) based on Virginia’s counties is ready. 133 Virginia counties and its equivivalents ingested into Supabase backend, methodology page published, validation experiment against verified Prince William County data center sites. At this stage, before scaling up further, I am focused on calibrating the methodology and testing the hypotheses that emerged during the research. To continue this work, I will need significant time and resources that will allow me to focus on this project.

Expected outputs. A public tracker spanning PJM and MISO (another ISO and RTO can be added in case of full-time work on the project); an open county-level compute-concentration dataset with a web tool; a published, citable methodology; and a research roadmap for the observatory's next stage.

Link to project MVP:

https://ai-buildout-frontier.vercel.app/

Link to GitHub:

https://github.com/Geofuturist/ai-buildout-frontier

Theory of Impact

One of the central concerns in AI safety is the concentration of power. If a single actor — a state, a corporation, or another entity — comes to believe it holds a decisive and lasting advantage in AI-driven capability, that belief may increase incentives for dangerous escalation, reduce opportunities for coordination, and contribute to catastrophic conflict between major nuclear powers. This is a known instability in strategic thinking about asymmetric technological advantage, and AI-driven compute concentration is a new and underexamined version of it.

Assessing and governing that risk requires data — specifically, the ability to see where compute-driven advantage is actually accumulating, not only at the national level, but at the level of the physical infrastructure that produces it. No public tool currently tracks how AI infrastructure is geographically concentrated at the subnational resolution where siting, energy access, and permitting decisions actually happen.

AI Buildout Frontier is built to close that gap: a public, county-level dataset and index that lets governance and safety researchers see infrastructure concentration as it happens, rather than reconstruct it after the fact. You cannot govern what you cannot measure — this project gives the field a measurement layer that doesn't yet exist.

People

Volodymyr (Vladimir) Ilin

Team Member

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$5,000

Ideal

$45,000

How the money will be spent

Minimum funding ($5,000) would cover the project's operating costs (hosting, software, API credits, and data acquisition) and allow me to continue developing AI Buildout Frontier on a part-time basis. At this level, the focus would be on completing the Virginia methodology recompute (v2.0), improving the platform, and beginning expansion to the PJM region.

Ideal funding ($45,000) would allow me to work on the project full-time for approximately one year. In addition to project expenses, it would fund a modest living stipend, enabling substantially faster data collection, methodology development, validation, and platform improvements. The goal at this level would be to expand coverage beyond PJM to additional US electricity markets (ERCOT, MISO, and others), with the long-term objective of building a county-level observatory covering the entire United States.

Budget categories

-Software, hosting, cloud services, and APIs

-Data collection and processing

-Platform development and maintenance

-Living stipend (ideal funding level)

-Accounting, administrative costs, and contingency

Discussion

No comments yet. Be the first to share your thoughts.