Can Duke University’s small AI data center ever be small enough for critics?
AI-generated summary reviewed by our newsroom.
- Duke is building a two‑story GPU center on its Central Campus in West Durham.
- The center will initially draw 1.5 megawatts with capacity to expand to 3 megawatts.
- Duke will cool the data center with its water plant and send the hot water to campus.
“Data center” has become a pejorative term to many across North Carolina. However, the projects compelling people to pack town halls and demand construction moratoriums have largely been the so-called hyperscale AI campuses, which draw 100 megawatts or more at peak and are filled with graphics processing units, or GPUs.
Neighbors near the town of Apex recently rallied against a local 300-megawatt proposal until the developer withdrew. Edgecombe County residents are still fighting an even bigger data center plan in Eastern North Carolina. And while Amazon won’t share the projected energy use of its incoming Richmond County campus, it’s safe to say this promised $10 billion data center will be hyperscale.
But other AI data centers are much smaller — if still controversial. Last month, Duke University began building a “Graphics Processing Unit center” on its Central Campus in Durham, a project first reported by Inside Climate News. In a statement to The News & Observer, Duke wrote it plans to open this two-story structure next year to facilitate research. Work crews were on the site Thursday.
The center is being constructed off Anderson Street, Duke’s building permit shows, near the university’s main campus and adjacent to a water plant and high-voltage power station. It will initially draw 1.5 megawatts, Duke said, with space to double. The estimated price tag is nearly $23 million.
“Consistent with Duke’s Climate Commitment, the facility is designed with a focus on environmental responsibility and sustainability,” the university wrote in a statement Wednesday to The News & Observer. “With this project, Duke aims to set an example for how to build energy-efficient, carbon emission-aware infrastructure that meets the computing needs of the modern research university.”
Data centers have existed for decades, but artificial intelligence has spiked interest in facilities with graphics processing units, or GPUs. AI data centers consume more electricity and water than general computing facilities because GPUs generate significant heat and need to be cooled.
Duke University says it will chill its data center with its own water plant and then redirect hot water to campus buildings. “The center will be built and operated without cost to the public,” the school wrote.
In a February report, a Duke University committee on AI policy recommended the school stop competing with large tech companies on general-purpose computing, and “instead architect an infrastructure that is uniquely suited to our interdisciplinary mission.”
Duke is not unique among other universities in looking toward AI computing. The University of Texas at Austin last fall acquired servers run on more than 4,000 Nvidia GPUs. The University of Florida has also developed a massive computing center, called the HiPerGator, with support from its state government and Nvidia. Harvard University and Rice University are among other institutions with GPU processing clusters.
The Durham data center plan has divided people on social media. Some have argued there’s nothing worrisome about a top research university opening a multi-megawatt computing facility. Others have questioned Duke’s framing of an environmentally responsible data center — and emphasized that an AI facility full of GPUs is different from what came before.
“We should be demanding that every data center, regardless of the size, tell us how and where they intend to get their water from,” said Rania Masri, director of organizing and policy at the North Carolina Environmental Justice Network. “What kind of chemicals they will use in the cooling process, exactly how they get their energy. And every single data center in North Carolina, regardless of the size, regardless of its location, should be applying for a conditional use permit.”
Masri added there is no legal assurance that Duke University will stop at one three-megawatt site.
Duke began site work on May 18, two weeks after the Durham City Council adopted a two-month moratorium on new data center projects.
According to the city, the university obtained its building permit in early April before the temporary pause. City spokesperson Amy Blalock told The N&O that the moratorium also excludes data processing sites “that are secondary to a main use” — which includes those linked to hospitals and educational institutions — “so long as they are used only for on-site needs.”
This story was originally published June 4, 2026 at 2:55 PM.
