The AI Buildout's Next Constraint Is Not Chips. It Is Neighbors.

A city council in Edgewater, Florida is considering a ballot measure to ban new data centers outright. Population: roughly 24,000. It will not change the trajectory of AI on its own. It is worth your attention anyway, because it is the first constraint on this buildout that money cannot route around.

Everything else so far has been solvable with capital. Chips are a queue. Land is a purchase. Power is a contract. A referendum is none of those.

## The scale that provoked it

The same day Edgewater floated its ban, Hut 8 announced a second 15-year IT lease worth $9.8 billion, covering 352 MW at its Beacon Point campus in Texas, part of a site planned for roughly 1 GW. The tenant is described only as an existing investment-grade customer.

Also that day, the Department of Energy's National Nuclear Security Administration selected a contractor for a 1 GW AI data center in South Carolina.

Two gigawatt-class commitments, announced within hours, in a single country.

For context on where this actually sits today: Our World in Data estimates data centres consumed around 485 TWh in 2025, roughly 1.5% of global generation, with AI-focused facilities accounting for about 0.5%. That is the number worth holding onto. AI's electricity footprint is currently small. The commitments being signed this month are not sized for a small footprint.

## The financing is getting more visibly strained

The demand side of this looks confident. The balance sheets funding it look less so.

Oracle launched a $20 billion equity offering to fund its AI infrastructure pivot, following an S&P downgrade to BBB-. Issuing equity after a downgrade is not the move of a company that finds this cheap to finance.

Further down the stack, Super Micro's shares have fallen about 60% from their 52-week high as the server business proves lower-margin than the AI framing implied. Selling the shovels only works while the shovels have pricing power.

## The part that should bother you

Return to Hut 8's unnamed tenant. Look across the deals we track and the pattern repeats: a striking share of the largest compute commitments involve a counterparty that is not disclosed.

The supply side of this buildout is highly legible. We can count megawatts, racks, campuses and lease terms. The demand side is substantially anonymous. You can verify that the capacity is being built. It is much harder to independently verify who has committed to paying for it over fifteen years, or how concentrated those commitments are among a handful of buyers.

That asymmetry is fine while demand holds. It is exactly the asymmetry that makes a correction hard to see coming, because the first sign of trouble would be a counterparty you were never able to name.

## What to do with this

**Watch permitting, not press releases.** The Edgewater measure is a leading indicator. Track whether similar measures appear in Texas, Virginia and Arizona over the next two quarters. Local power politics is where the buildout's real timeline gets set, and it moves slower than a funding announcement but binds harder.

**Ask who the tenant is.** When you read the next multi-billion-dollar compute lease, check whether the counterparty is named. If it is not, treat the announcement as evidence about supply, not demand. Those are different claims, and the second one is the one the valuation rests on.