A new analysis finds that data centers impose billions in environmental and public health costs, with AI operations accounting for a smaller but still significant share. The findings sharpen the debate over who pays for the power, water, and pollution tied to digital expansion.
taxesAIdata centerspublic healtheconomyexternalitieselectricitywater usepollutionclimate
The rapid expansion of data centers is creating a growing economic burden that is often left off the balance sheet. A new analysis estimates that the environmental damage from operational data centers cost the economy about $25 billion last year, with $3.7 billion of that directly tied to AI activity. The estimate includes the effects of electricity demand, air pollution, greenhouse gas emissions, and the health consequences linked to those emissions, including premature deaths.
The findings point to a familiar problem in modern industry: profits are concentrated while many of the costs are spread across the public. Data centers consume large amounts of electricity to power servers and cooling systems, and that demand can increase pollution from the grids supplying them. The study treats those harms as an externality, meaning a cost imposed on people and communities that are not the direct beneficiaries of the activity.
The scale of the buildout is striking. In North America, investment in server farms used to train and run artificial intelligence models surged by $47 billion last year, with companies also borrowing heavily to finance the expansion. Much of that spending went into cooling systems, plumbing, and other infrastructure needed to keep the facilities running around the clock. Supporters say the facilities are essential to the digital economy. Critics argue that the pace of growth is outstripping the public's ability to absorb the consequences.
One of the biggest concerns is electricity use. Data centers require high and steady power loads, which can force utilities to generate more energy from fossil fuels when cleaner supply is not available. That translates into more pollution and more climate damage, especially in regions where the grid is already under strain. The analysis converts those effects into dollar terms using standard economic methods, including the social cost of carbon and estimates of mortality risk tied to dirty power generation.
Water use is another pressure point. Many facilities rely on large volumes of water for cooling, and in drought-prone areas that can add stress to local supplies. Residents in some places are already seeing higher utility bills and worrying about long-term access to water. The problem is not limited to one kind of facility. While AI systems are a major driver of new demand, many ordinary data centers also add to the load, meaning the issue extends beyond the newest wave of machine learning investment.
The debate over possible fixes is increasingly practical rather than theoretical. Some argue that data centers should be required to use more renewable energy, install solar panels, recycle water, and adopt more efficient cooling systems. Others point out that the basic physics are unforgiving: solar panels can help offset some demand, but their power density is far lower than the power density of a data center. That means rooftop solar alone cannot replace the large amount of electricity these facilities consume. Batteries can smooth short interruptions, but they are not a substitute for continuous power.
Nuclear power is often raised as another option, especially for large, steady industrial loads. But that path comes with its own political and safety questions, and not everyone trusts companies or regulators to handle the risks responsibly. Some observers argue that the real issue is not the technology itself but the incentives surrounding it: businesses want cheap power, fast permitting, and minimal oversight, while communities are left to deal with the consequences if the system is not carefully managed.
That tension is especially visible in places where data centers have been built quickly and with generous tax treatment. In some cases, local governments have offered long-term tax breaks to attract facilities, reducing the immediate public revenue while leaving residents with the environmental costs. The result is a poor trade for many communities: a lot of electricity use, relatively few permanent jobs, and a larger burden on the grid, the water system, and public health.
The broader argument is not that all data centers are identical or that the internet can function without them. They are critical infrastructure for communications, commerce, and storage. But the latest estimates suggest that the current model of expansion is too often treated as an economic win even when the full costs are pushed outward. That is especially true for AI, where the pace of construction has been driven by competition, speculation, and the promise of future profits that may not fully materialize.
The central question is no longer whether digital infrastructure matters. It does. The question is who pays for it, how much damage is acceptable, and whether the public should continue subsidizing a model that leaves pollution, water stress, and health costs behind. If the economy is absorbing $25 billion in damage from data centers in a single year, then the price of convenience and computation is higher than many people have been told.
That matters because these costs do not stay abstract for long. They show up in higher utility bills, strained water systems, dirtier air, and communities that have little say over the facilities arriving in their neighborhoods. The more AI and cloud services expand, the more urgent it becomes to decide whether the current incentives make sense. Without stronger guardrails, the digital economy may continue to grow in ways that are profitable for a few while expensive for everyone else.
