Who Should Pay for AI Infrastructure?

— AI and Public Governance Series (Part 3)

When people talk about artificial intelligence (AI), the most visible benefits usually come to mind first: faster search, more capable writing tools, smarter customer service, and a growing range of applications that automate tasks once performed by humans.

Yet behind these seemingly effortless digital services lies an enormous physical infrastructure. AI depends on land, data centers, transmission lines, substations, power generation, cooling systems, and the people and capital required to build and maintain them over decades.

The first two articles in this series examined two related questions. The first explored why data centers have become a new object of local governance. The second explained how AI’s rapidly growing demand for electricity is pushing the electric grid back to the center of public policy.

Those discussions naturally lead to a third—and perhaps more difficult—question.

If AI infrastructure requires massive new investment and continuous expansion, who should ultimately pay for it?

This is not simply a financial question. It is fundamentally a question of public governance. How costs are allocated often determines whether a community’s pursuit of “economic transformation” ultimately serves the public interest—or merely enables private expansion at public expense.

The Cost Is More Than a Building

When a company proposes a new data center, the costs appear straightforward.

The developer purchases land, constructs buildings, installs equipment, hires employees, and assumes operating expenses. If those were the only costs involved, public controversy would likely be limited.

The reality, however, is considerably more complicated.

A modern AI data center requires much more than the buildings inside its own campus.

It may require additional transmission capacity, new substations, expanded generation resources, long-term utility planning, and adjustments to roads, drainage systems, emergency services, land-use planning, and other public infrastructure.

In other words, the costs associated with a data center exist on two levels.

The first consists of costs that appear on the company’s balance sheet.

The second consists of costs that arise because the broader public system must adapt to accommodate the project.

It is the second category that is often overlooked.

Economic development announcements frequently emphasize investment totals, construction activity, and projected tax revenue. Much less attention is given to infrastructure expansion, pressure on energy systems, or the long-term responsibility for maintaining public assets.

As a result, public discussions may present only one side of the ledger while leaving significant costs embedded within public systems.

Those Who Benefit Should Bear the Costs

Any discussion of AI infrastructure should begin with a simple principle:

those who directly benefit should bear an appropriate share of the costs.

If a data center requires dedicated transmission upgrades, specialized substations, or unique electrical capacity, those incremental investments should not automatically be transferred to the general public. Ordinary households are neither the primary users of these facilities nor the direct beneficiaries of the profits they generate.

This is not an argument against data centers, nor is it an argument against AI-driven economic development.

It is a basic principle of public finance.

Local governments should welcome investment, encourage economic diversification, and create reasonable conditions for emerging industries. But welcoming investment does not mean shifting private operating costs onto taxpayers or utility customers.

A sound industrial policy distinguishes between two different kinds of infrastructure spending.

The first consists of project-specific investments required for a company’s own operations—specialized electrical connections, dedicated infrastructure, and facilities that primarily serve a single project. These costs should largely remain the responsibility of the developer.

The second consists of investments that strengthen the long-term capacity of the entire community: a more resilient electrical grid, cleaner energy systems, stronger regional planning, and improved emergency infrastructure.

When infrastructure benefits society broadly rather than serving a single company, public investment becomes easier to justify.

The central issue, therefore, is not whether governments should invest.

It is whether the boundaries between public investment and private responsibility remain clearly defined.

The Most Difficult Subsidies Are the Ones Nobody Sees

The most controversial forms of economic development are often not the subsidies that appear openly in government budgets.

They are the ones that remain largely invisible.

Consider an electric utility that expands transmission capacity to serve major industrial customers. If those costs are gradually recovered through general electricity rates, many residents may never realize that they are indirectly financing part of a private company’s expansion.

Or consider a local government that adjusts land-use priorities, infrastructure schedules, or public resource allocation to accommodate a major development project. Without sufficient transparency, the public may find it difficult to evaluate whether those decisions are justified.

The greatest problem with invisible subsidies is not merely their existence.

It is that they weaken meaningful public debate.

If governments openly explain how much infrastructure support they intend to provide, what tax incentives are being offered, and what public benefits are expected in return, residents can evaluate whether those trade-offs are worthwhile.

But when costs are quietly distributed across electricity bills, infrastructure budgets, land-use decisions, or long-term maintenance obligations, they become much harder to identify—and therefore much harder to question.

This represents one of the defining governance challenges of the AI infrastructure era.

Unlike traditional manufacturing plants, data centers often generate relatively limited long-term employment. Unlike commercial districts, they may have little direct interaction with surrounding neighborhoods.

Their economic value may be substantial, but their public visibility is relatively low.

If governments emphasize only technological investment while failing to explain the accompanying public costs, communities may discover years later that they have been paying for commitments they never fully understood.

Public Investment Is Not the Problem—The Conditions Matter

None of this suggests that governments should avoid investing in infrastructure related to AI.

Such a conclusion would be far too simplistic.

Every major stage of economic development has required public investment. Railroads, highways, airports, ports, and internet backbone networks all depended, to varying degrees, on government participation. Without public infrastructure, modern economies cannot grow.

The real question is therefore not whether governments should invest.

It is whether those investments genuinely advance the public interest.

If upgrading the electric grid improves reliability for residents, supports cleaner sources of energy, reduces future outage risks, and creates capacity for multiple industries rather than a single project, such investment should not automatically be viewed as a subsidy to one company.

It may instead represent an investment in the city’s long-term competitiveness.

On the other hand, if infrastructure primarily serves one company, one industrial campus, or one business model while distributing costs across the broader public, then those decisions deserve much closer scrutiny.

That requires local governments and public utility regulators to ask difficult questions.

Who ultimately benefits from this infrastructure?

Who pays?

How will costs be recovered?

Will electricity prices be affected?

What happens if technological assumptions change or corporate investment declines?

Could today’s expansion become tomorrow’s stranded public asset?

The earlier these questions are asked, the lower the public risks are likely to be.

Taxes, Jobs, and Public Costs Cannot Be Measured Separately

Supporters of large-scale data center projects often emphasize tax revenue, private investment, and economic diversification. These arguments are neither unreasonable nor insignificant. For regions that have long depended on tourism, gaming, or real estate, attracting digital infrastructure can broaden the economic base and create new opportunities for future growth.

Yet evaluating these projects requires more than calculating potential revenue.

Public officials must also account for public costs.

Employment generated by data centers is often concentrated during the construction phase, while long-term operational staffing tends to be comparatively modest. Tax revenue should therefore be assessed alongside public expenditures, infrastructure demands, land use, and long-term maintenance obligations.

Otherwise, an impressive investment announcement may not necessarily translate into meaningful public benefit.

This does not diminish the value of data centers.

Rather, it suggests that public discussion should move beyond the language of economic recruitment.

A mature local government should be able to explain clearly:

How much long-term tax revenue is expected?

How many permanent jobs will actually be created?

What supporting infrastructure will be required?

Will electricity prices or utility capacity be affected?

How much public investment will accompany private investment?

If tax incentives are offered, what return is expected—and over what period?

Conversely, what opportunities might the community lose if projects like these are rejected?

Only when both benefits and costs appear on the same public ledger can residents reasonably judge whether a development project truly serves the community.

The AI Era Requires a New Public Ledger

One reason AI infrastructure presents such unusual governance challenges is that it connects policy areas that have traditionally been treated separately.

Technology policy, electricity systems, land-use planning, environmental resources, public finance, and regional economic development are increasingly becoming parts of the same conversation.

That means traditional approval processes may no longer be sufficient.

In the past, a project could often be approved if it complied with zoning regulations, satisfied transportation requirements, and met building standards.

AI infrastructure demands broader questions.

Can this region support the long-term electrical demand?

Will public infrastructure need to expand alongside private investment?

How should cumulative impacts be evaluated?

Who bears financial responsibility if future assumptions prove incorrect?

Most importantly, are the costs being distributed fairly?

These are questions that conventional development review was never designed to answer.

For that reason, data centers should not be viewed simply as real estate projects or technology investments.

They are infrastructure projects.

And infrastructure projects inevitably raise questions about how costs, benefits, and risks should be shared across society.

The Real Position: Transparency, Fairness, and Accountability

Debates over AI infrastructure are often framed as a choice between supporting development and opposing it.

That is a false choice.

Unconditional support risks overlooking public costs.

Automatic opposition risks preventing communities from benefiting from technological change.

A more responsible position begins elsewhere—with three principles.

The first is transparency.

Governments and developers should explain not only how much private investment a project will generate, but also what public infrastructure it requires, what long-term obligations it creates, and what uncertainties remain.

The second is fairness.

Those who directly benefit from infrastructure should bear an appropriate share of its costs. Private expansion should not quietly become a public financial obligation without clear public justification.

The third is accountability.

Every tax incentive, every infrastructure commitment, and every cost-sharing arrangement should be understandable, open to public scrutiny, and subject to meaningful evaluation.

These principles matter because competition among cities in the AI era will not be determined solely by who builds data centers the fastest.

It will also depend on who governs technological growth most responsibly.

Ultimately, the future of a city will not be defined simply by its ability to attract AI infrastructure.

It will be defined by whether it can welcome innovation while protecting the public interest, maintaining fairness in the distribution of costs, and ensuring that residents understand the future they are collectively helping to build.

By Voice in Between