Who Pays for the AI Buildout?

The federal government’s plans to expand AI infrastructure are coming to light. In my experience, a budget lives as a moral document and tells a story.

The FY2027 federal budget, released this month, makes two critical moves. It invests $4.7 billion through the Department of Energy (DOE) to accelerate AI computing infrastructure. It also eliminates roughly $15.7 billion in federal programs that serve the communities where AI data centers are being built.

The $4.7 billion includes $1.2 billion for seven AI supercomputers at two national laboratories in Illinois and Tennessee, and $3.5 billion to build new power plants to run them. These systems are part of the Genesis Mission, a public-private partnership between the federal government and companies including NVIDIA, Oracle, AMD, and Google, aimed at making the United States the global leader in AI-powered scientific research.¹

The DOE has also designated four national laboratory and legacy nuclear sites — in Idaho, Tennessee, Kentucky, and South Carolina — to host AI data centers powered by nuclear energy, and solicited proposals from private companies late last year.2 A low-carbon energy source, nuclear power tends to be more reliable and stable, and it is well-suited to support the constant load AI systems require.

Nuclear-powered AI infrastructure is also being developed in other communities — Three Mile Island, for example – that have absorbed decades of environmental costs from nuclear waste storage and radioactive releases from accidents. That project in Pennsylvania, for Microsoft, is now seeking accelerated approval from regulators to go online by the end of 2027, instead of 2031.

Increasingly, policymakers and the public need institutional capacity to negotiate terms for a new generation of infrastructure they do not fully understand. The pattern of siting high-impact federal projects in communities with limited negotiating power looks familiar. This is what we are seeing happen with AI data centers funded by hyperscalers and developers, often lacking transparency.

The $15.7 billion in federal programs, proposed to be cut, includes programs that millions of Americans depend on: energy bill assistance for low-income households (LIHEAP, $4 billion), community development grants that fund local planning and infrastructure (CDBG, $3.3 billion), affordable housing investments (HOME, $1.3 billion), economic development assistance for distressed communities (EDA, $449 million), digital access programs (NTIA, $2.2 billion), and state water infrastructure funding (EPA SRFs, $2.5 billion), among others.3

There appear to be no line items — in any federal agency — that fund community infrastructure, technical assistance, ratepayer protection, or water use and disclosure in the context of data center siting or any environmental program. These kinds of investments have been characterized in the budget as wasteful. The administration's stated priorities — energy dominance, AI leadership, and faster permitting — receive funding. The proposed budget funds faster permitting for developers, not community capacity to engage with permitting.

The ratio of cuts to spending is roughly 3 to 1. For every dollar the federal government invests in AI energy infrastructure, three dollars in community support are proposed for elimination. The budget investments and cuts not only indicate that their priorities have changed, but the eliminations tell us that when states and local communities seek these funds, the money will not be there. The federal government has national security interests in mind, has positioned itself at the forefront of the AI race, and deprioritizes community resilience programs that help lower-income households adapt to climate and infrastructure impacts.

Share

The Convergence Imperative framework launched in January began with a hypothesis — the issues of climate, community and capital are colliding. Not only is this coming to light in more visible ways now, but the data show us that communities at the highest risk of climate change impacts will also bear the cost and burden of infrastructure buildout, if we do not address it head on and take more action to correct this imbalance.

Many of the federal programs on the chopping block are the tools communities use to adapt to climate change. For example, CDBG funds municipal stormwater upgrades and flood mitigation. EPA’s State Revolving Funds finance water infrastructure in communities facing drought and contamination. The Economic Development Administration supports climate resilience planning in distressed regions. The IIJA (Infrastructure Investment and Jobs Act) clean energy funding proposed to be cancelled — $15.2 billion — eliminates the largest federal investment in distributed energy and grid modernization in a generation.

Philadelphia still carries $1.2 billion in unmet costs from Hurricane Ida. The communities absorbing new AI infrastructure demand are, in many cases, the same communities lacking the investment required to be more climate resilient ahead of the next storm. The FY2027 federal budget proposes to withdraw the federal programs that address both — simultaneously.

With federal support receding, states and municipalities are taking action. As of March 2026, more than 54 local governments have passed moratoriums or restrictions on data center development. Twelve states have filed bills this legislative session. These are bipartisan and geographically diverse, spanning Georgia, Michigan, Virginia, Oklahoma, Vermont, and Wisconsin. Maine just passed a moratorium through November 2027, now awaiting the governor to sign it into law.4

Moratoriums are a blunt instrument. They stop everything while communities figure out what they want. The more instructive cases are the ones where communities are building accountability tools during the pause — while imperfect, under pressure, and with limited resources.

Pennsylvania is developing what may be the most layered model in the country. Beyond the only Community Benefit Agreement (or CBA) documented in Lancaster, a recent rate case settlement between PPL Electric and consumer advocates creates a new electricity rate category specifically for large data centers.5 It is the first time a Pennsylvania utility has agreed to prevent data center infrastructure costs from being passed to residential customers.6 A new municipal ordinance guide published by Chester and Montgomery Counties gives local governments model language for zoning, water, and energy standards.7 Meanwhile, state legislation (HB 1834) requiring ratepayer protections has passed the House and is awaiting Senate action. All of these efforts are needed, and most actions are reactive.

Michigan offers a different lesson. Google’s proposed 1-gigawatt data center in Van Buren Township — which would use as much electricity as a major city and up to 3.6 million gallons of water per day — advanced over the opposition of more than 1,500 petition signers.8 But Google committed to a contested regulatory review of its energy contracts (a process a prior developer had tried to skip), pledged to cover the full cost of energy infrastructure, and announced a $10 million fund for energy affordability. These are not community benefit agreements. But they are concessions no other developer in our dataset has offered voluntarily. The question is whether they can be formalized and implemented, with clear accountability, before the next project arrives.

In contrast, Richland Parish, Louisiana — one of the poorest parishes in the state, with 25% living under the poverty line — faces a proposed Meta campus, named Hyperion, amongst the largest of its peers estimated to require between 5GW to 10GW. The parish lacks the planning staff, legal capacity, or technical expertise to evaluate a proposal of this scale, let alone negotiate protections. Under the FY2027 budget, the federal program most likely to have funded that capacity (the Economic Development Administration) is eliminated entirely. The intermediary gap is not a metaphor here. It is the literal absence of institutions.

Share

This brief is not a set of policy recommendations, but it is a call to action. First principles have been emerging throughout this series. For foundations, family offices, and impact investors positioning capital in or around AI infrastructure, the FY2027 budget clarifies several conditions that were previously ambiguous. Now that we know what the data shows us, what we do next must consider the following principles:

The regulatory environment is being written right now, state by state, and investors who engage before frameworks harden will shape better terms than those who arrive after.

The federal floor is dropping. The community development programs that historically provided baseline institutional capacity in lower-income communities (e.g., energy assistance, housing investment, planning support, digital equity) are proposed for significant reduction or elimination. Whether Congress enacts every cut is uncertain. If left uncontested, communities across the country will experience a substantial loss of funding, risking state budgets. Governors and mayors would need to backfill eliminated programs in their state budgets, or lose capacity entirely.

State and local regulatory responses will intensify. Communities that have lost federal support will use the tools they still control: zoning, permitting, rate cases, and ballot initiatives. Investors with data center exposure should expect the regulatory environment to shift — not at the federal level, but at the state and municipal level, where community resistance is organized and accelerating. No matter where you live in the United States, your community will face a decision: whether to allow continued underinvestment in local infrastructure while private actors gain access to land and resources without federal accountability. Position capital in front of the frameworks — not after.

Lower-income communities need capacity to negotiate. Compound Impact tracks 32 AI data center projects across 20 states.9 Each project is scored on three measures: the share of low-to-moderate income residents in the surrounding county, access to high-voltage power lines, and water stress. When we mapped these sites against the programs the FY2027 budget proposes to cut, a pattern emerged – three out of four sites are in lower-income communities. These are the same communities that rely disproportionately on the federal programs being eliminated. $16.2 billion in public subsidies have flowed to these projects with no community benefit agreements in place. Capital spending on AI infrastructure outpaces verified enterprise AI revenue by more than 10 to 1, creating risk if projected demand evaporates. A CBA would create a binding contract that specifies what residents receive in exchange for hosting the facility, whether or not the project succeeds.10

Philanthropic capital that arrives after permits are approved is funding advocacy. AI infrastructure financing operates on short cycles: GPU hardware depreciates in 1–3 years, and private equity funds run 5–7 years. But the grid infrastructure built to serve these facilities locks communities into 30–40 year commitments. The federal budget is accelerating the short-cycle side while doing nothing to prepare communities for the long-cycle consequences. Once power plants are built and grid connections are signed, the terms are fixed for decades.

Intermediary capacity is the binding constraint — and the investment opportunity. Invest in the intermediaries already doing the work, and organize at the state level. The gap between communities and developers is not principally about money. It is about the technical assistance, legal support, and institutional knowledge that allow communities to participate in infrastructure decisions before the terms are set. Funding that capacity — through intermediary organizations, community programs, environmental impact assessors, and legal clinics — is where philanthropic and impact capital can do work that public markets cannot. Throughout this year we have come across a number of new and existing organizations where capacity needs are high, including the Better Data Center Project, largely a volunteer effort of former federal workers and concerned citizens.

The window for pre-deal negotiation is closing in the jurisdictions where these projects are planned. Capital that arrives before is funding architecture.

Share Compound Impact

This brief is part of Compound Impact’s quarterly convergence analysis, which examines what the data show at the intersection of AI infrastructure, climate finance, and community development.

The pattern documented here — capital accelerating in one direction while community capacity is withdrawn in the other — is not unique to AI. It is the same structural condition that climate finance has been working to address for decades, now replicating at unprecedented speed.