The AI boom is not only a software story. Behind every larger model, video generator, and cloud service is a physical network of data centers that must draw electricity from somewhere.
New research led by a team at the Harvard T.H. Chan School of Public Health puts the scale of the problem in sharper focus: since 2018, carbon emissions from data centers in the US have tripled. That leaves the sector slightly below domestic commercial airlines as a source of this pollution.
Data centers are becoming an AI bottleneck
The United States has 2,990 data centers, and most of them do not look like symbols of a major environmental challenge. They are often plain, functional buildings. Yet they support much of the digital world, including the computing needed for artificial intelligence.
As AI companies compete to build bigger and more capable systems, the energy requirements keep rising. The source article points specifically to the trend toward more energy-intensive AI models, including video generators like OpenAI’s Sora, as a force that could push emissions higher.
That creates a direct conflict for leading AI companies. They face pressure to meet sustainability goals while also trying to keep pace in a market where larger AI models can require far more power. The technical race is therefore also an infrastructure race.
Why the electricity source matters
The emissions problem is not only about how much electricity data centers use. It is also about where that electricity comes from.
Many data centers are located in coal-producing regions, including Virginia. According to the research cited in the source article, the “carbon intensity” of the energy used by these facilities is 48% higher than the national average.
The same research found that 95% of data centers in the US are built in places with electricity sources that are dirtier than the national average. In practical terms, that means the AI energy problem is tied closely to the geography of power grids.
For companies trying to lower their climate impact, simply operating more efficiently may not be enough. If new AI data center capacity is built in regions with dirtier electricity, the emissions burden can remain high even when the buildings themselves are optimized.
Nuclear power is gaining attention
A growing group of technology companies is now looking at nuclear energy as one answer to AI’s power needs. The attraction is clear from the source: AI companies need large amounts of electricity, and nuclear power is being discussed as a way to supply it.
Several major companies have already moved in that direction:
- Meta announced on December 3 it was looking for nuclear partners.
- Microsoft is working to restart the Three Mile Island nuclear plant by 2028.
- Amazon signed nuclear agreements in October.
But nuclear power does not solve the short-term problem by itself. The source article notes that nuclear plants take ages to come online. Even where companies are interested, the timeline for new or restarted power sources may not match the immediate growth of AI computing demand.
Public opinion also remains a factor. Support has increased in recent years, and president-elect Donald Trump has signaled support, but only a slight majority of Americans say they favor more nuclear plants to generate electricity.
The buildout is moving beyond the US
The search for more AI infrastructure is not limited to the United States. OpenAI CEO Sam Altman pitched the White House in September on an unprecedented effort to build more data centers, but the industry is also looking abroad.
Countries in Southeast Asia, including Malaysia, Indonesia, Thailand, and Vietnam, are courting AI companies. They are hoping to become new data center hubs as demand for computing capacity grows.
This global push shows how quickly AI’s physical footprint is expanding. The software may be distributed instantly, but the machines that power it require buildings, land, electricity, and long-term energy planning.
What this means for AI’s next phase
The central issue is no longer whether AI needs more computing power. The source makes clear that the industry’s current direction is tied to larger models and more energy-intensive systems.
That puts AI companies in a difficult position. They can pursue larger systems, but doing so increases pressure on electricity supplies and sustainability commitments. They can seek cleaner energy, including nuclear power, but those plans may take time to become meaningful sources of supply.
For now, the research suggests that existing data centers are still drawing power from sources that are far from renewable. With emissions from US data centers already tripled since 2018, the next wave of AI development will be judged not only by what the models can do, but by how the industry powers them.