U.S. states are starting to pull back the tax breaks that made data centers cheap to build, and the AI companies racing to add compute are about to feel it. According to The Information, several states are moving to repeal or scale back the incentives that have subsidized data center construction for years. That shift, as detailed in The Information’s exclusive report, points to higher costs for the exact infrastructure powering the current AI boom.
This matters because data centers are where AI actually lives. Every model you use runs on racks of GPUs sitting inside these buildings, and the economics of building them just got worse.
What’s happening
For over a decade, states competed to lure data centers with generous tax packages. Sales tax exemptions on servers and equipment. Property tax abatements. Discounted power. The pitch was simple: land a big tech tenant, get jobs and prestige.
Now the mood has turned. The Information reports that states are reconsidering those deals as the downsides pile up:
- Data centers use enormous amounts of electricity and water, straining local grids.
- They create few permanent jobs relative to their physical footprint and tax cost.
- Residents in some areas are seeing power bills climb as utilities build out capacity to serve them.
When a state repeals an exemption, the operator suddenly owes sales tax on hundreds of millions in hardware, plus ongoing property taxes it had been avoiding. Those numbers add up fast at hyperscale.
Why this hits AI hard
The timing is the story. This pullback lands right as OpenAI, Anthropic, Google, Meta, Microsoft, and a wave of startups are committing to unprecedented buildouts. We’re talking about tens of billions in planned capital spending on new compute.
Here’s what stands out to me: the industry has been modeling its expansion on a cost structure that assumed these subsidies would hold. Strip them away and the math changes. Higher build costs flow downstream into:
- Higher training costs for frontier models.
- Higher inference costs, which is what you pay every time an AI answers a query.
- Pressure on the already-thin margins of AI startups that rent capacity from cloud providers.
The status quo was cheap, subsidized expansion. That era is ending in a growing number of states.
The bigger squeeze
Tax breaks are only one piece. Data center operators are already fighting rising costs on several fronts at once:
- Power. Grid capacity is scarce and utilities are raising rates to fund expansion.
- Chips. High-end GPUs remain expensive and supply-constrained.
- Land and construction. Prime sites near power and fiber are getting picked over.
Remove the tax incentive and you add a fourth cost line to a bill that’s already growing. Operators can’t easily absorb all of it, which means some of it gets passed to customers.
What to expect next
A few things worth watching in the coming months:
- Expect operators to shift new projects toward states that still offer strong incentives. Site selection becomes a tax-arbitrage game.
- Expect louder lobbying. Tech companies will push hard to keep the breaks they still have, arguing jobs and investment.
- Expect the cost pressure to surface in pricing. If your company runs on API calls or cloud GPUs, factor in that the underlying economics are trending up, not down.
For practitioners, the practical takeaway is to stop assuming compute gets cheaper on a smooth curve. Falling model prices have been the trend, driven by efficiency gains and competition. Rising infrastructure costs work against that. The two forces are now pulling in opposite directions.
None of this stops the AI buildout. Demand is too strong and the strategic stakes too high. But it does make the buildout more expensive, and it hands states more leverage over where the next wave of data centers actually lands.
The full breakdown, including which states are leading the repeal push, is available in The Information’s original reporting.