When Is a Site Actually Ready for AI Infrastructure?
A few weeks ago, I wrote about what I saw as a missing layer in Atlanta’s AI infrastructure strategy: coordination.
When I say AI infrastructure, I mean the physical layer behind artificial intelligence: the places where computing happens, along with the required power, fiber, cooling, and land. AI infrastructure can include large data centers, smaller compute facilities, or computing systems inside existing buildings and institutions.
These pieces of AI’s physical infrastructure increasingly have to be considered together, even though they do not always enter the conversation at the same time.
Over the past few weeks at Mosaic Ventures, we have been testing some of these questions against real Atlanta properties through the Atlanta Distributed Infrastructure Initiative, or ADII. ADII is an early effort that we started to better understand site readiness, utility coordination, and the practical work that has to happen before an infrastructure opportunity can move from an idea to something real. In this piece, when I use “site,” I mean a property being evaluated for AI infrastructure: the land, building, and the access and control conditions around it.
That work raised a foundational question: When is a site actually ready?
At first, determining site readiness sounds straightforward. Find a property. Check the land and building. Look at the zoning. See whether power and fiber are nearby. Work through the economics.
But once you start verifying those factors, “site readiness” gets harder to define.
Our ADII work has reinforced something simple: a property can look promising and still be a long way from being ready.
What the Map Doesn’t Tell You
One thing we’ve learned is that site diligence is often less about finding information than figuring out which information you can rely on.
A property record can identify an owner without telling you who can authorize a sale, lease, or redevelopment. Existing electrical infrastructure does not tell you how much power can be delivered today. And a telecommunications carrier may be able to reach a building without offering the cost, reliability, or redundancy the intended use requires.
We ran into the control issue directly in one ADII review. Public records pointed us toward an entity that appeared to hold the title. Direct outreach clarified that the entity held title only as part of a tax-incentive structure and did not control the property. The actual transaction path ran through the owner’s authorized real-estate representatives. The site had not changed; our understanding of who could actually make a deal had.
The deeper we got into the process, the clearer the lesson became:
Infrastructure readiness is not a property characteristic. It is a chain of things that must be verified.
But the technical pieces are only part of that chain.
The Community Question Comes Earlier
A site can be technically feasible and still be a poor fit for the community around it.
Communities are asking how AI infrastructure could affect their homes and neighborhoods: noise, water use, backup generation, land consumption, visual footprint, growing electricity demand, and whether new infrastructure costs could eventually show up in household utility bills. They are also asking what benefits, if any, remain local.
A property being considered for AI infrastructure near housing, transit, retail, or walkable development may have another use that creates more value for the nearby community and surrounding area. Another property may make sense for smaller-scale compute without the footprint of a large data center campus.
Another question is who the AI infrastructure is being built for and who benefits from it.
Does it create useful computing capacity or infrastructure access for local companies, universities, or institutions? Does it contribute meaningfully to the local economy? Or are the costs of AI infrastructure concentrated in one place while most of the value goes elsewhere?
Communities also live with these infrastructure choices long after market conditions change. A longer-term question worth asking is what happens if demand changes?
Infrastructure decisions tend to last longer than market cycles. A highly specialized facility may occupy land and draw utility investment for decades, even if the assumptions that justified it change. That makes long-term adaptability a community question as well as an economic one.
None of that means a project should not move forward. It means those questions belong in the early phase of site selection, not at the end.
Community fit is becoming financially relevant too. Reuters reported in August that lenders are paying closer attention to permitting and local opposition when evaluating data-center projects. Community resistance, approval delays, and legal challenges can affect financing and, ultimately, whether projects move forward at all.
Community fit is not separate from feasibility. Increasingly, it is part of it.
Nearby Doesn’t Mean Ready
Power and connectivity have taught us a similar lesson.
Seeing electrical infrastructure near a property is useful, but it does not tell you how much power can actually be delivered, what upgrades might be required, how long they would take, or who would pay for them.
We saw that distinction in our ADII work. At one Atlanta property, a prior electrical permit confirmed that the building had historical panels and branch circuits. That was useful evidence, but it still did not tell us whether service was active, how much capacity was available today, whether the transformer was adequate, or what upgrades and lead time a new load would require. We still treated power as an open gate.
Utilities face a similar problem on a much larger scale. In its 2025 Long-Term Reliability Assessment, the North American Electric Reliability Corporation distinguishes speculative or exploratory large-load requests from projects that have advanced into real development commitments when assessing future electricity demand. The distinction matters because utilities cannot treat every inquiry as firm future load.
A developer’s inquiry is not the same thing as committed electricity demand, and a power line near a property is not the same thing as usable capacity.
Georgia recently saw that play out at a much larger scale with OpenAI’s planned data-center campus in Effingham County.
Georgia Power’s agreement to provide the project with up to 3,200 megawatts went through review by staff at the Georgia Public Service Commission. The agreement ultimately moved forward, but not without additional protections. Georgia Power agreed that if a large data-center customer leaves early, related shortfalls would be allocated to other large customers rather than residential customers and small businesses. The utility also agreed to additional public reporting around large data-center contracts.
The project is on a completely different scale from the smaller opportunities we’ve been looking at through Mosaic Ventures in Atlanta. But the process makes a broader point: even when a customer is real, and a project has been announced, questions about power, cost, risk, and who carries that risk still have to be worked through.
Fiber has its own version of this problem.
At one Atlanta property we studied through ADII, AT&T Business confirmed that dedicated internet service was available at the building, that bandwidth could scale with the project’s needs, and that installation could take up to 60 days.
That sounded like a major box checked. But we still needed written pricing, service-level terms, the selected bandwidth, installation scope, the building’s network handoff point, and evidence of a truly independent backup route. Availability improved the site’s connectivity case, but it did not make the site “fiber ready.”
That fiber review reinforced an important rule for how we think about infrastructure information: new evidence should reduce uncertainty, not give us more confidence than it deserves.
Atlanta Is Working Through the Same Readiness Questions
In August 2026, the Atlanta City Council’s City Utilities Committee approved a proposal to create a Data Center Task Force. The proposed group would examine siting and regulation alongside grid capacity, water and cooling needs, noise, tax impacts, and costs to utility ratepayers.
When the proposal reached the full Council on August 17, members sent it back to committee. The proposal was still being held in the City Utilities Committee in mid-September, so the task force had not yet been established as of that meeting.
The exact structure of Atlanta’s data-center policy process is still taking shape, but the underlying questions remain.
Power decisions affect cost. Cooling decisions affect water. Land use affects communities. Permitting affects financing. And all of those decisions influence whether a project that looks possible on paper has a realistic path to being built in a way Atlanta can support.
A Better Definition of Ready
Atlanta does not need another list of available properties.
What would be more useful is a better way to tell the difference between a property that is interesting and one that is actionable.
That means knowing enough about site control, power, connectivity, land use, community fit, and economics to understand whether there is a realistic path forward and being clear about what is still unknown.
Not every question has to be fully resolved before a project moves forward, but everyone involved should know which answers are still missing.
That may be the biggest thing we have learned from doing our ADII work over the past several weeks.
Better infrastructure intelligence is not mainly about collecting more information. It is about knowing what has been verified, what is still an assumption, and which unanswered question could change the decision.
Atlanta will see more proposals tied to AI, data centers, compute, and the infrastructure that supports them.
Some will be good opportunities. Others will look attractive until the harder questions are asked. And some may be technically possible but simply wrong for the place being considered.
The point is not to make AI infrastructure harder to build. It is to ask those questions early enough that developers, utilities, governments, and communities are not spending months fighting over projects that were never ready or never right for the location in the first place.
Atlanta does not need to say yes to every available site. It needs to know enough to understand what it is saying yes to, and why.
Author Bio:
Nsikan Uboh is the founder of Mosaic Ventures, an Atlanta-based venture platform focused on AI infrastructure, regional innovation systems, and technology for public benefit. Through Mosaic Ventures, he is developing the Atlanta Distributed Infrastructure Initiative, a research and coordination effort focused on responsible AI infrastructure growth in the Atlanta region.