The Brockovich AI Datacenter Investigation: Mapping America's Expanding AI Infrastructure

A Public Investigation into AI Infrastructure

As the generative AI wave sweeps the globe, the physical backbone supporting model training and inference — data centers — is expanding across the United States at a breathtaking pace. A project called "Brockovich AI Datacenter Reporting" has turned its attention to this often-overlooked issue: where exactly are these power-hungry, water-intensive, land-consuming AI data centers located, and what do they mean for local communities?

The project's name evokes environmental activist Erin Brockovich, signaling that its core focus isn't technology itself, but the environmental and social costs behind the expansion of AI computing infrastructure. The emergence of this kind of investigative reporting marks a shift in public attention toward AI — away from "model capabilities" and toward "physical consequences."

Why AI Data Centers Deserve Tracking

Training and deploying large language models requires massive GPU clusters, which in turn require purpose-built data centers to house them. Compared to traditional data centers, AI data centers operate at higher power densities, generate far more heat, and consume staggering amounts of electricity. A single large AI data center can consume as much power as a small to mid-sized city.

This expansion creates real-world problems across several dimensions:

Energy Pressure

AI data centers are driving a surge in grid demand, and some regions are already experiencing power supply constraints. To secure computing capacity, tech giants have begun signing long-term power purchase agreements and even restarting or investing in nuclear power facilities. Tracking the geographic distribution of data centers helps clarify where regional grid stress is actually concentrated.

Concrete examples of tech companies reviving nuclear power have already emerged: in 2023, Microsoft signed a deal with Constellation Energy to fund the restart of the Three Mile Island nuclear plant in Pennsylvania — the site of the 1979 accident — and reconnect it to the grid. Amazon, meanwhile, acquired a data center campus adjacent to a nuclear plant to access direct power supply. The logic behind this trend is that nuclear power provides stable, around-the-clock baseload power with minimal carbon emissions, aligning with tech companies' net-zero commitments. However, nuclear restarts involve long timelines and enormous upfront investment, making them incapable of relieving immediate grid pressure in the short term. More broadly, the long-term Power Purchase Agreements (PPAs) signed by companies like Google and Microsoft are competing with traditional industrial users and residential consumers for limited renewable energy quotas, indirectly driving up electricity costs in some regions.

Water Consumption

The immense heat generated by high-density computing must be handled by cooling systems, many of which rely on large volumes of water. In water-scarce regions like the American West, the siting of data centers can create competition with agriculture and residential water use.

A data center's cooling approach directly determines how much water it consumes. The three main methods are air cooling, evaporative cooling, and liquid cooling. Evaporative cooling is highly efficient but consumes enormous amounts of water — heat is carried away as water evaporates, and that water cannot be recovered. Research estimates that training GPT-3 consumed approximately 700,000 liters of fresh water, equivalent to the water used in manufacturing roughly 370 cars. Immersion cooling and direct-contact liquid cooling, which have gained traction in recent years, can significantly reduce water consumption but carry high retrofit costs and have yet to become industry standard. In states like Arizona and Texas, local governments have begun requiring data center operators to disclose water usage data and incorporating it as a condition of site approval — a clear example of public oversight driving policy change.

Land Use and Community Impact

Data centers tend to be sited in areas with low electricity prices and abundant land, but the impacts of their construction and operation on surrounding communities — from noise and traffic to the gains and losses of tax incentives — are often poorly disclosed to the public.

The Value of Investigative Data Projects

Projects that aggregate and publicly share information about data center locations, scale, and operators are fundamentally filling a transparency gap. Tech companies typically don't proactively disclose detailed operational data for every data center they run, and independent tracking and reporting can give policymakers, researchers, and the public the evidence they need to make informed decisions.

Methodologically, these projects typically combine satellite imagery, public planning documents, permit records, and corporate announcements to build a national map of AI infrastructure — transforming the abstract concept of "computing power" into a physical reality that can be concretely examined.

This approach to information gathering has a specific name in investigative journalism: Open Source Intelligence (OSINT). When applied to AI infrastructure tracking, satellite imagery analysis is particularly valuable — the cooling towers, solar panel arrays, and parking lot layouts of large data centers have distinctive signatures in high-resolution satellite images that can be cross-referenced against building permit records in local planning databases. Utility companies' interconnection requests filed with regulators are another key source, since any large power consumer connecting to the grid must go through this public process. This multi-source, cross-verification method allows researchers to map the geographic distribution of AI computing capacity with reasonable accuracy, even when companies choose not to disclose it voluntarily.

Balancing Industrial Expansion with Public Oversight

Building AI infrastructure is undeniably necessary to support technological progress, but how to strike a balance between industry development and environmental and community interests is becoming an increasingly pressing public issue. Projects like Brockovich AI Datacenter Reporting represent a form of bottom-up oversight.

For practitioners who care about AI's long-term impact, understanding the geographic distribution and costs of data centers is just as important as understanding model architecture. The physical cost of computing power ultimately feeds back into society in the form of energy prices, water allocation, and environmental impact.

Conclusion

This project hasn't generated much discussion on Hacker News yet, but the issues it touches on — transparency and sustainability in AI infrastructure — are inevitably going to become more important as the industry expands. As we marvel at the capabilities of AI models, we should not lose sight of the vast physical network that underpins it all, nor the real social and environmental costs it carries. Tracking, documenting, and making that information public is where public participation in this technological transformation begins.