When the Boom Exits

Signals Brief #8 (2026.03.02) | Why Data Center Deals Need More Than Promises

Mar 06, 2026

Follow the capital flows — but also follow what happens when they reverse.

In the past year, Microsoft walked away from more than 2 gigawatts of data center capacity across the U.S. and Europe.1 The Stargate joint venture — OpenAI’s $100 billion infrastructure bet — has functionally stalled amid partner disputes.2 Goldman Sachs’ chief economist assessed AI’s contribution to the U.S. economy in 2025 as “basically zero.”3 Moody’s flagged $662 billion in future data center commitments sitting entirely off hyperscaler balance sheets — obligations that could evaporate if demand projections don’t materialize.4

None of this means AI won’t transform industries. It means the timeline is uncertain, the demand is unproven, and someone will bear the cost if projections shift. Whether AI delivers transformative productivity in 2027 or 2035, communities need protection mechanisms now. Last week I framed the challenge through the history of fiber optics and the early internet boom days. This time, the question is whether communities can secure protection before the deals are signed — and whether we have a window to get it right.

The Policy Landscape Is Fragmenting

Four serious policy frameworks have emerged for governing data center development. Brookings published a playbook proposing Community Equity Endowments and compute vouchers for startups — creative mechanisms, but pitched to governors and economic development officials rather than the communities hosting facilities.5 Harvard Kennedy School dissected OpenAI’s AI Opportunity Zones proposal, warning that in-kind technology donations depreciate rapidly and that exclusive university partnerships risk platform lock-in.6 Public Citizen called for decommissioning bonds and minimum durational commitments.7 And Linn County, Iowa enacted what may be the nation’s first comprehensive local ordinance requiring data center developers to post financial security against abandonment — borrowing directly from wind and solar bonding practice.8

These are substantive contributions. They are also disconnected. What’s missing is a more interconnected framework: one that considers asset decommissioning, protects against downside risk, and establishes intermediary capacity to negotiate and enforce agreements with entities that have vastly more resources and information. This builds upon The Convergence Imperative thesis which addresses these three dimensions. By extension, it may help inform economic policy solutions for this next industrial revolution.

Hyperscalers Are Not Trusted Brokers

A Heatmap national poll found only 44% of Americans would welcome a data center nearby — making data centers less popular than gas-fired power plants or nuclear facilities. In Pennsylvania, 68% of residents oppose data center development. Microsoft used the code name “Project Nova” in Caledonia, Wisconsin, where even a village trustee didn’t know the company’s identity until residents opposed the project and it was rejected.

These examples demonstrate how information asymmetry erodes trust. NDAs shield operational data from public scrutiny. Water usage estimates prove wildly inaccurate. Communities remember Foxconn’s Mount Pleasant catastrophe, where state and local officials spent $1.2 billion in public money on a manufacturing project that promised 13,000 jobs and delivered fewer than 1,200.9

The uncomfortable truth is that hyperscalers are structurally poor trust brokers for their own commitments. Their incentives favor speed, opacity, and optionality — which is precisely why intermediary infrastructure matters.

The Intermediary Gap

This is the Intermediary Gap in action. Good Jobs First has spent 28 years documenting how clawback provisions fail when communities lack capacity to monitor and enforce them. Linn County can pass the best ordinance in the country — and it may be exactly that — but without a funded intermediary organization with legal, technical, and community organizing capacity, enforcement depends on volunteer vigilance against well-resourced corporate legal teams. Local governance works when local intermediaries are funded to make it work.

Lancaster, Pennsylvania negotiated $20 million in direct payments tied to construction milestones, plus a $10 million letter of credit backing a 100% renewable energy requirement — with fund distribution governed by a multi-stakeholder "Innovation Hub" committee.10 That’s more than most communities have secured. But examine what’s missing: no funded technical assistance for ongoing community engagement. No demand-risk protection if the hyperscaler pivots. No mechanism for renegotiation when conditions change. No intermediary organization with the expertise to monitor compliance across a 20-year facility lifecycle.

Contrast this with the NAACP Solar Equity Initiative, where pre-existing organizing infrastructure — relationships, expertise, constituency — enabled effective community participation in solar deployment. The lesson isn't that CBAs are insufficient. It's that legal provisions require funded organizations with the expertise and staying power to make them work — and most communities don't have them, because current policy frameworks don't pay for them.

Toward a Financial Assurance Framework

There’s considerable downside risk to a community when a project is left either unrealized or whose lifespan is cut short. I wonder what it looks like to establish a financial assurance framework for data center development. The idea would come together as a set of interconnected mechanisms — synthesizing existing tools like decommissioning bonds, clawback provisions, and demand-risk collateral — but adding something none of them provide alone: the intermediary capacity communities need to negotiate, monitor, and enforce these protections over 20-year infrastructure lifecycles.

Here’s what most policy frameworks miss: they treat hyperscaler pullback as company-specific failure — bad actors, broken promises, corporate malfeasance. In contrast, I am arguing something different. Microsoft’s 2+ GW capacity pivot was not a misrepresentation. It was a rational response to market signals in an industry where AI agents haven’t yet delivered the productivity gains that justify current infrastructure spending. The risk communities face isn’t bad actors — it’s demand volatility. That’s how free markets work.

A financial assurance framework would have at least these four components:

Decommissioning bonds protect against physical site abandonment — now standard in renewable energy, emerging in Linn County for data centers
Clawback provisions recapture incentives when job or investment commitments aren’t met — Good Jobs First’s core reform agenda
Demand-risk collateral secures energy commitments against hyperscaler pivots — the ComEd/Illinois model Brookings mentions in passing
Funded intermediary capacity ensures communities have technical assistance throughout the project lifecycle — the missing piece everywhere

The Heat Dividend: A Provocation

There’s another component that I imagine would improve outcomes. We often hear of the incredible amount of water data centers require to keep cool, or risk overheating. But even when cooling benefits are achieved, there is still a meaningful level of heat being generated that goes to waste. The provocation then is: Can data center waste heat address community climate adaptation needs in local communities?

What I am calling a “Heat Dividend” incentive has European precedent. Meta’s facility in Odense, Denmark recovers 100,000 MWh annually to heat hospitals and community buildings. Google’s Hamina, Finland center provides 80% of local district heating demand. Germany’s Energy Efficiency Act will require new data centers to utilize at least 10% of waste heat by July 2026, rising to 20% by 2028.

Meanwhile, Pennsylvania faces $15.47 billion in climate adaptation costs by 2040 — stormwater drainage ($7.8 billion), road maintenance ($2.98 billion), urban tree planting ($1.7 billion), school air conditioning ($1.23 billion).11 Rural and high-poverty communities face the highest per-capita costs, often without the fiscal capacity to respond.

Think about this: What if AI Opportunity Zones went further to be reframed around climate resilience? The capital is flowing regardless. The question is whether communities can direct that investment toward infrastructure that serves them even if AI timelines stretch or demand shifts.

Heat capture may not be viable everywhere without European-style district heating networks, and I don’t yet know whether the economics work in U.S. contexts. But as a test case for whether community benefit frameworks can integrate climate resilience, it feels worth investigating.

As I document emerging patterns, the window for community voice narrows. State legislatures are debating bills now. I keep coming back to one question: What would change the calculus on these deals regardless of whether the boom delivers on its promises?

Discussion about this post

Ready for more?