Healthcare's AI Boom Is Feeling the Heat

Or perhaps patients and communities will.

AI is rapidly becoming embedded throughout healthcare—from clinical documentation and decision support to revenue cycle management and population health analytics. Behind that transformation sits a physical infrastructure that receives far less attention: data centers.

As healthcare organizations make procurement decisions that may shape their technology ecosystems for years, few are considering the environmental and public health consequences of the infrastructure required to power these systems.

While maybe a third of the growth in data centers will be self-powered (“behind the meter”), the rest won’t. And while the explosion in renewables has helped support the global growth in grid demand from new data centers, a large part of the next chunk of capacity is going to come from expanded natural gas generation and, in some regions, continued reliance on coal.

In the meantime, we are moving into the El Niño phase of the ENSO cycle, with high probability of a strong or very strong (“Super”) El Niño. Heat waves will almost certainly intensify this year compared to last. Records for heat are likely to become memories as this El Nino takes global temperatures into unprecedented territory.

Data centers need a lot of electricity (and water) to cool themselves. And that’s not all. New studies are showing that data centers significantly heat the air and land surface around them, creating or worsening existing urban heat islands. (NOTE: the study showing major land surface warming is not yet peer reviewed, but the study showing downwind heating of the air in Phoenix has been.)

So this explosion of data centers is creating a cascade of unintended hazards including heat pollution, air pollution and also noise pollution because of the arrays of extremely loud fans employed.

People need electricity to cool themselves, but in the US, they are finding it increasingly unaffordable. Many are tragically (and needlessly) having to choose between buying food and keeping their house at a survivable temperature.

PJM, the biggest power grid in North America that powers the world’s largest cluster of data center, has clarified that it has the authority to cut off power to data centers in an emergency to avoid power blackouts. And last year Texas as the first state to pass a law that allowed utilities to cut off power to data centers during emergencies.

So at least the utilities are trying to avoid feeding data centers while people starve and succumb to overwhelming heat. (Although it looks like that disconnection itself can destabilize grids where data centers are highly concentrated, like in Northern Virginia!!)

Nonetheless, the grid could get stressed to and beyond capacity this summer. The New York Independent System Operator (NYISO) in their summer 2026 capacity assessment announced they will fail at the 99^th percentile estimate of peak demand, even with emergency measures. They fail at the 90^th percentile without them. The North American Electric Reliability Corporation warned in their 2025 Long Term Reliability Assessment that the grid could fail to meet demand in 2028. High risk areas cover much of the country, including parts of the Mid-Atlantic and Midwest, South Central and Northwest regions. But that’s baseline demand. If there are widespread extreme heat waves this summer, places beyond New York may see demand exceed capacity. As extreme heat conditions spike data centers’ demand for cooling, we could see utilities having to manage difficult tradeoffs between maintaining data center operations and meeting critical community needs.

What happens if electricity demand exceeds capacity and power shuts down during a heat wave? One study analyzed the health impacts of a grid failure during a heatwave in major US cities and found that not only did heat-related mortality double overall, but in Phoenix, a staggering 50% of the urban population would require medical attention. The city’s hospitals certainly don’t have that kind of surge capacity.

And so as healthcare makes this massive investment in implementation, who is paying attention to these alarms? Who is sounding the watch?

Organizations like the National Association to Advance Colored People (NAACP) and the Natural Resources Defense Council (NRDC) are mobilizing to protect communities, especially environmental justice communities. Virginia’s legislature commissioned a panel to re-examine state sales tax exemption for data centers and explore other policy options.

What are the regulatory options to address these interwoven risks? The policy space is complicated, dynamic, and undeveloped. Here is some of what I’m seeing and some additional thoughts:

· Public Utility Commissions control utility rates, interconnection standards, and the utility part of permitting (as opposed to land permitting). They have the power to demand financial commitments to avoid rate increases and assure data centers expand capacity when needed to avoid power shortages.

· Local governments control zoning, land use planning, and permitting. Many are already requiring environmental impact assessments and community benefits agreements. Consideration of heat-island impacts should become a routine component of these reviews.

· Communities and local governments should also incorporate projected data center growth into heat action plans, resilience strategies, and emergency preparedness planning.

· Federal agencies like the National Science Foundation and ARPA should promote research to speed discovery and development of technologies that are more energy efficient and don’t generate as much heat as present technologies.

· The current administration has sought to remove environmental guardrails from federal permitting processes for data centers. While this does not affect commercial installation on private land, it is intended to remove constraints from construction of data centers on federal lands. Congressional oversight might be able to affirm the need for less reckless approaches.

But what about the health sector? Hospitals and clinics depend on reliable supplies of electricity and water during extreme weather. They care for the patients most vulnerable to heat-related illness, respiratory disease, and power outages. And increasingly, they are among the largest purchasers of AI-enabled technologies. What can health systems, health care product and service companies, and health care professionals contribute to more intelligent build out of data centers and artificial intelligence?

· One critical concept for health to embrace is that bigger is not always better with AI-enhanced health IT tools. But bigger is always going to be hotter, dirtier, more consuming and more polluting. Scientist Sasha Luccioni offers a vision of how to right-size AI in this TED talk. Anu Ramachandran and co-authors have published a “Decision Framework to Mitigate the Energy, Emissions, and Cost of AI Implementation” in the NEJM Catalyst that guides healthcare providers and IT purchasers on considerations to right-size AI tools for healthcare. This kind of thinking needs to be rapidly expanded and implemented.

· Healthcare professionals need to get involved in procurement decisions and in creating criteria for procurement contracts. The National Academy of medicine is considering a process to create standard language for AI and EHR procurement.

The question is not whether healthcare should use artificial intelligence. It almost certainly will, and many applications may improve efficiency, quality, and patient outcomes. The question is whether healthcare can deploy AI thoughtfully enough to avoid creating new risks while solving existing ones. The signs are mixed. On the one hand, it seems very few leaders in healthcare are paying attention to the environmental damages associated with data centers as a consequence of their AI purchases. JAMA’s Summit Report on Artificial Intelligence failed to mention the harm to health from air pollution, water scarcity or noise pollution, much less the heat island effect. On the other hand, a new article suggests that health systems are slowing the breakneck rush to implementation of AI, doing more due diligence and starting fewer projects. This could be a unique opportunity to get ahead of the environmental harm curve and push for the integration of environmental considerations into healthcare’s use of AI-enhanced tools.

Healthcare has long understood that every intervention carries benefits, costs, and unintended consequences. AI should be treated no differently. As health systems, clinicians, technology companies, and policymakers shape the next generation of healthcare infrastructure, they should evaluate not only what AI can do for patients, but also what its physical footprint means for the communities those patients call home.