- Tension: Ohio became a leading AI infrastructure destination on a tax break it suspended when the bill arrived: $1.6 billion instead of the projected $142 million
- Noise: The story has been covered as a political dispute, not as the first data point in a national reckoning about what AI infrastructure genuinely costs local communities.
- The Direct Message: A projection of $142 million that became $1.6 billion in a year doesn’t change what AI infrastructure costs — it changes who finally asks the question.
To learn more about our editorial approach, explore The Direct Message methodology.
The number that changed everything was not a small one. Ohio’s data center tax exemption was projected to cost the state $136 million in fiscal year 2025 and $142 million in fiscal year 2026. Those projections had been reviewed, debated, and accepted by a legislature that decided the economic returns — billions in investment, construction jobs, the reputational currency of being an AI infrastructure hub — justified the cost. Then the actual cost arrived. In 2024, the exemption ran to $554 million. In 2025, it reached nearly $1.6 billion. Governor Mike DeWine suspended the tax break.
The gap between what was agreed to and what materialized is not a rounding error. It is a factor of more than ten. Whatever Ohio’s legislators had weighed when they approved the incentive, they had not weighed a $1.6 billion annual cost, because no one had told them that was the number on the table. When that number became clear, the calculation changed.
How Ohio became the AI infrastructure state
Ohio did not arrive at this position by accident. Over the past several years, the state has attracted approximately $37 billion in data center investment, with Amazon, Google, and Meta building or expanding facilities in New Albany, Dublin, Hilliard, and surrounding communities in central Ohio. The tax exemption — which removed sales tax on equipment purchases for qualifying data centers — was the instrument used to secure that investment. It worked, in the sense that the data centers arrived. The question the suspension now forces is what “working” actually cost.
Data centers require two things in large quantities: land and power. Ohio offered both, along with the tax exemption and a location well-positioned on fiber routes and power grids. The facilities that moved in are not modest. Hyperscale data centers — the kind that support AI training and inference at the scale that major technology companies now require — consume hundreds of megawatts of electricity and draw enormous volumes of water for cooling. The energy demand alone reshapes the infrastructure requirements of any region that hosts significant concentrations of them.
The tax exemption that drew them was, in effect, a commitment by Ohio’s public to absorb the fiscal cost of that hosting. What was not fully reckoned in the original projections was how quickly the demand for AI infrastructure would grow, and how directly that growth would translate into the tax exemption’s cost. The exemption is structured to scale with investment — more data center equipment purchased, more tax forgiven — and investment in AI infrastructure has not followed a modest growth curve.
What happened at the community level
The fiscal story playing out in Columbus was, simultaneously, a different story in the towns and townships absorbing the infrastructure itself. Jerome Township passed a moratorium on new data center construction as residents began organizing a referendum campaign. Lordstown and Washington Township followed with similar measures. In Preble County, residents temporarily blocked a proposed data center project through the threat of a referendum effort alone.
The concerns animating these campaigns are specific: electricity prices rising as grid demand increases; water usage at a scale that affects local supply; the visual and acoustic character of communities being transformed by server farms that employ relatively few people per acre of land they occupy. Data centers are not factories. They don’t generate the kind of employment density that traditionally justifies the land use and infrastructure disruption they require. They generate tax revenue — except, under the exemption, substantially less of it — and they generate the kind of digital infrastructure that makes them valuable to the technology companies running them, a benefit that accrues primarily elsewhere.
The statewide response has escalated beyond local moratoriums. A group called Ohio Residents for Responsible Development received approval from the Ohio Ballot Board to begin collecting signatures for a proposed constitutional amendment that would permanently prohibit new hyperscale data center construction in the state. The campaign needed 413,000 signatures by July 1 to make the 2026 ballot. Organizers fell well short — collecting roughly 77,000 signatures — and have announced they will target the 2027 ballot instead. The fact that a statewide campaign to ban new data center construction above 25 megawatts cleared legal approval and gathered tens of thousands of signatures in a state that has attracted $37 billion of the relevant investment remains a significant signal of public sentiment, regardless of whether it reaches voters this cycle.
The question other states now have to ask
Ohio’s suspension is consequential, not because it resolves anything — DeWine has been explicit that he supports data center investment and intends to replace the suspended exemption with something more carefully structured — but because it establishes, publicly and with specific numbers, what the cost of hosting AI infrastructure actually looks like when it runs at scale. Before Ohio’s suspension, the national conversation about data center incentives was largely conducted in the language of economic development: investment figures, job projections, tax revenue. The actual cost of the tax forgiveness involved was less visible and less often stated as a single number against which the benefits could be weighed.
Ohio has now stated that number. $1.6 billion in a single year, for a program projected to cost $142 million. Those figures will be in the room in every state legislature that has its own data center incentive under discussion or renewal, and there are many. Virginia, Texas, Georgia, and Arizona have all attracted significant data center investment under comparable incentive structures. None of them have published fiscal costs at the scale Ohio just disclosed, but Ohio’s experience suggests the costs may not have been fully modeled.
The broader context matters here. The four largest technology companies are together spending approximately $725 billion on AI capital expenditure in 2026, up 77% from the prior year. That investment goes somewhere physical — into land, power, cooling, and equipment — in communities that host it. The tax structures that have subsidized that hosting were designed before the scale of investment was known. Ohio found out what the scale implied when the exemption cost arrived. Other states will find out too, and Ohio’s experience gives them a reference point they didn’t have before.
What “first” actually means
The characterization of Ohio’s action as the first major state to pull back from AI infrastructure investment on cost grounds is accurate but requires some precision. Ohio has not banned data centers. It has not expelled the companies already invested. It has suspended a specific tax instrument while a legislative committee examines its design. DeWine’s stated intention is to rebuild the incentive in a form that doesn’t produce $1.6 billion annual costs — a more targeted exemption, presumably with caps or scaling limits that prevent the kind of unconstrained growth the current structure allowed.
What Ohio has done is become the first state where the actual fiscal cost of hosting AI infrastructure was large enough, and the gap between projected and actual cost was dramatic enough, that a sitting governor concluded the existing arrangement could not continue unchanged. That is a different and more durable kind of “first” than a political statement or a symbolic vote. It is a state government looking at the real number and deciding it changes the terms of the deal.
The question for other states is not whether to attract data center investment. The question is whether the deals they have made, or are considering making, have been modeled honestly — whether the projected costs reflect what AI infrastructure build-out at 2026 scale actually requires, rather than what it required when the incentive was designed. Ohio answered that question by discovering the answer in its fiscal accounts. The more prudent approach, which Ohio’s experience has now made possible, is to ask the question before the bill arrives.
