The environmental conversation about AI tends to arrive at aggregate energy figures, which are contested and hard to make decision-relevant. The conversation that actually determines whether a facility gets built is local, procedural and much more concrete: can the grid connect it, where does the cooling water come from, and who is entitled to object.
The interconnection queue is the constraint
In most US regions the limiting factor on new data centre capacity is not land or capital but interconnection — the queue to connect to the transmission network, which in several markets is measured in years. This has consequences that get missed in coverage focused on emissions.
It means siting decisions are driven substantially by where spare grid capacity exists, which is not where demand is. It means large loads compete with generation projects for the same queue positions. And it means the electricity a facility draws is, at the margin, whatever the local system operator dispatches to meet it, regardless of what the operator has contracted for on paper.
Water, and the disclosure gap
Cooling water is consistently the most contentious local input, particularly in water-stressed regions, and it is the least consistently reported. Practice varies from detailed disclosure of consumption and source to nothing at all, and comparison across operators is largely impossible.
For an organisation with disclosure obligations, the practical difficulty is that its own reported figures depend on data its cloud provider may not supply at facility level. This is worth establishing before a reporting deadline rather than during one.
What participation requirements do
The benefits of a data centre are distributed nationally or globally; a substantial share of the costs — grid capacity, water, traffic, land use, noise — falls within a few miles. Public participation in permitting is the principal mechanism by which the local side of that ledger enters a decision that is otherwise made by parties who do not bear it.
Narrowing participation therefore does not make the objection go away. It relocates it to litigation, to state politics, and to the point at which a project is far enough advanced that opposition is more expensive to accommodate. Developers have generally learned this the hard way in other infrastructure classes.
Where this lands for a risk function
An organisation with material compute commitments should know which facilities serve it, what their grid and water position is, and what the permitting status of any expansion it depends on looks like. That is a supply-chain resilience question as much as an environmental one: a facility that cannot expand is a capacity constraint that will surface as a procurement problem long before it surfaces as a disclosure one.