The U.S. Spends More Power on AI Than Every Hospital Combined

The topic of artificial intelligence has changed from interest to infrastructure in recent months. Algorithms and ethics were the main topics of discussion not long ago. Utility executives are talking about megawatts, transmission lines, and substations today. The transformation has happened quite quickly.

In the United States, artificial intelligence systems currently use more electricity than all hospitals put together. The underlying data is very consistent across different energy evaluations, despite the comparison’s initial overblown, almost dramatic, sound. Data centers powered by AI are growing at a rate that is noticeably quicker than that of conventional commercial expansion.

Hospitals run constantly, powering climate-controlled wards, imaging labs, ventilation systems, and operating rooms. As a result of decades of engineering discipline, their energy use is incredibly dependable and meticulously controlled. However, data centers used to train and run AI models have remarkably high electricity consumption, frequently using 10–50 times as much power per square foot as a typical hospital.

The adoption of renewable energy has significantly increased during the last ten years, primarily due to developments in wind and solar power. Meanwhile, AI workloads have increased even more sharply, resulting in a demand curve that utilities say is especially difficult to manage. According to projections, by 2028, electricity demand associated to AI may account for about 10% of all U.S. electricity consumption.

That figure is significant.

By the end of 2025, AI systems alone might be responsible for about 49% of all data center power consumption, or nearly 23 gigawatts of continuous demand. 23 gigawatts is equivalent to the combined output of dozens of big power plants, to put that into perspective. The scale is tangible, buzzing, and well-lit; it is not abstract.

Last fall, I took a tour of a regional data center and stood under rows of racks that were cooled by liquid and made a sound like a bee swarm. Each server was buzzing constantly, converting electricity into compute. With a proud smile, the engineer who was assisting me said that although their new GPUs were noticeably quicker and more efficient than the previous generation, the grid’s overall load had grown.

Expansion is not slowed by efficiency alone.

With the ability to process billions of computations per second and consume greater datasets, AI models are become more sophisticated. Clusters of specialized chips must run nonstop for weeks or months in order to train these systems. The demand is increased by inference workloads, which process millions of user requests every day.

In the meanwhile, hospitals are using AI into their operations in ways that are especially advantageous. Numerous health systems have reduced peak energy consumption by double-digit percentages by optimizing HVAC scheduling through the use of advanced analytics. By drastically cutting down on clinician charting time, ambient documentation solutions have freed up human talent and subsequently decreased computing waste within facilities.

It’s a subtly powerful contrast.

Although AI puts a load on the grid, it also makes hospitals more efficient. This contradiction illustrates a larger trend in technical advancement: innovation uses resources while streamlining other industries. Whether this increase can become particularly durable without overloading infrastructure is the crucial question.

Utilities are quickly changing. Grid operators are designing substations especially for hyperscale data centers through strategic alliances with tech companies. In an effort to develop very creative microgrid solutions, several suppliers are experimenting with placing battery storage and renewable plants right next to AI labs.

However, the tension can feel urgent in medium-sized towns. Local planners must assess if transmission upgrades for schools, factories, or healthcare expansions will be severely delayed when a new AI campus proposes a 150 megawatt load. These are engineering limitations, not philosophical conundrums.

However, the tone does not have to be negative.

AI is predicted to significantly improve medication discovery, healthcare diagnostics, and preventive care in the years to come. Compared to traditional review, AI-assisted imaging can identify small problems earlier. By automating processes that previously required hours of manual labor, predictive models are simplifying operations and revolutionizing patient triage systems.

Hospitals are evolving into increasingly digital environments. By 2022, about one in five hospitals in the United States had adopted AI in some capacity, and since then, adoption has increased. Many say that operational performance has significantly improved, especially in energy management, scheduling, and billing accuracy.

This momentum points to a promising future.

AI infrastructure can become much more sustainable by including grid-aware scheduling algorithms, more energy-efficient chips, and sophisticated cooling systems. Engineers are creating processors with exponentially higher throughput that use fewer watts per calculation and are shockingly inexpensive in comparison to their performance advantages.

Policy also has an impact.

Updated government grid modernization projects have led to a steady increase in transmission capacity investment. The need for AI can be satisfied without compromising necessary services if policymakers match energy expansion with digital growth. It will take especially creative cooperation between utilities, technologists, and public health experts to achieve this alignment.

The fundamental fact is still the same: the US today spends more money on electricity to power AI systems than it does to power its hospitals. That fact is a scale indicator, not a condemnation.

Infrastructure is frequently abruptly altered by technological revolutions that start out softly. The development of the internet, railroads, and telecommunications networks all had remarkably similar paths, requiring massive amounts of resources before being fully integrated and streamlined. It seems that AI is continuing that historical trend.

It is tempting to present the comparison as a contest between caring and computation at this point. It is seen as a coordination difficulty from a more positive perspective. AI is enhancing hospitals rather than replacing them by helping doctors, evaluating imaging, and accurately predicting consequences.

The same intelligence that is driving increased demand for power may also direct load balancing, grid optimization, and storage integration if it is handled carefully. AI can help utilities create incredibly resilient systems, incorporate renewables more seamlessly, and forecast peaks more precisely.

Seldom does progress come in a tidy bundle.

Making sure AI infrastructure develops in line with societal priorities is now a challenge. Hospitals will continue to run around the clock, offering incredibly dependable and deeply human care. AI facilities will keep growing, analyzing medical data, language, and photos at previously unthinkable speeds.

The challenge is to effectively harness innovation rather than to impede it. The country can turn this imbalance into an advantage by combining energy-efficient engineering with progressive legislation to create systems that are extraordinarily adaptable and ethically powered.

The silent foundation of every modern innovation is electricity. How we distribute it will influence caring as well as computation.