Mike MacMillan for Business North Carolina
Amazon, Apple, Google, OpenAI affiliate Stargate. Everybody is building a data center these days. Data center construction represents one of the largest building projects in world history.
It is certainly one of the most expensive. “Total investment in upcoming and under construction data centers has soared to around $750 billion,” estimates Alex Wang, an analyst at Nashville, Tennessee-based investment firm Alliance Bernstein. That seems understated, with Stargate alone committed to spending $500 billion for AI-related data center projects over the next four years.
In North Carolina, Amazon says it will spend $10 billion for data centers in Richmond County as part of a broader $100 billion plan over 15 years. Apple is planning a $175 million investment at its data center in Maiden in Catawba County. Major projects are underway or being considered in several other NC cities.
As a result, Duke Energy projects energy demand in North Carolina will grow eight times faster over the next 15 years than in the previous 15. Artificial intelligence requires computing that demands massive energy inputs.
Data center investment is having a measurable impact on the country’s overall economic growth. Information processing investment contributed more than half of the 1.2% growth in U.S. GDP during the first half of 2025, according to The Wall Street Journal. UBS, an investment bank, forecasts that companies will spend $375 billion on AI infrastructure in 2025 and another $500 billion next year.
It’s a big number, and North Carolina is hoping to get its fair share, or maybe a little bit more. The state ranks ninth nationally in the number of data centers, with 91. Virginia leads with more than 600, followed by Texas (379) and California (318).
In terms of megawatts of electricity consumed by data centers, Texas leads the way with about 8,000 MW, followed by Virginia with 7,000, according to CarbonCredits.com, which tracks carbon usage. In North Carolina, the comparable figure is around 770 megawatts, reports the Baxtel research group. In 2023, data centers consumed 4% of U.S. electricity generation, and that may double by 2030, according to the Electric Power Research Institute.
Data centers are also getting bigger, with many defined as “hyperscale,” a term that IBM characterizes as a physical site large enough for at least 5,000 servers and possibly miles of connection equipment. These centers can “easily encompass millions of square feet of space,” IBM says.
THE NEXT LOUDON COUNTY
In the old days, pioneers followed rivers and streams to open up new territory, and cities were founded along their banks. More recently, railroads and the interstate highway system have helped drive site selection for corporations and distribution centers. Now, for data centers, it’s power availability.
In discussing how to assess a potential data center site, site selection expert Ben Rojahn says he looks for high-voltage transmission lines first, followed by fiber capacity and “some reasonable access to water and sewer.” He’s a vice president for data center solutions with the CBRE real estate firm in Charlotte.
Loudon County, Virginia, which is directly west of Washington, D.C., has perhaps the world’s greatest concentration of data centers. Statewide, they consumed a whopping 23.6% of the Dominion State’s electricity in 2023, according to the Electric Power Research Institute. For North Carolina, the comparable number was 1.9%.
Partly as a result, data center development is on the move. “The whole I-85 corridor is seeing a lot of activity from data center developers and users,” says CBRE’s Rojahn, moving down from Virginia and, to a lesser extent, up from Atlanta.
North Carolina appears well positioned to capture some of that growth. The state is poised to meet demands for larger power loads over the next four or five years, Rojahn says, making it competitive with other states. It has regions served by high-voltage lines built originally for furniture and textile companies and other manufacturers that can be put to work bringing power to data centers. While those lines may serve smaller markets rather than large metros, that is not an issue with some AI and cloud applications where latency — the time it takes data to move from where it’s sourced to the destination – is less of a concern.
With billions of dollars at stake, states are competing aggressively to be the next Loudon County. Incentives often include a combination of sales and use tax exemptions, property tax abatements, investment tax credits, and regulatory relief like relaxed zoning requirements. Duke Energy’s power rates, which are lower than those of most surrounding states, is another factor for North Carolina.
“Our job is to work with local leaders and economic development officials to help power their economy in a manner that matches their local vision for the community,” says Danielle Buckins, Duke Energy’s director of North Carolina development.
The benefits are clear, the costs perhaps less so. Consider the potential impact on Richmond County, where median income was $46,600 in 2023, compared with about $69,000 statewide. The county’s value of assessed property stands at $4.4 billion, which is less than Amazon’s proposed long-term investment. As many as 1,500 construction workers are expected to participate over the next five years.
Amazon has promised to create 500 permanent jobs, but hasn’t estimated the expected average salaries. The county offered a 20-year incentive package of cash grants equal to 50% of the annual property tax and 65% of the annual personal property tax for each of the 20 buildings that are proposed for the project. The package is contingent on $1 billion of investments and 50 jobs by the end of 2030.
“Five hundred jobs is huge for Richmond County,” says Martie Butler, a county economic developer. “A lot of years of planning went into developing our infrastructure (to attract a company like Amazon),” including support for Duke Energy’s generating facility near Hamlet, she says. “It will enhance our tax base significantly, and it won’t be a heavy lift for our community.”
A FOOL’S RACE?
In site selection, “The gating issues are ones that drive material cost impact — the availability of power, the availability of fiber,” says Jim Grice, a partner in Dallas at the Akerman law firm and a specialist in data centers and digital infrastructure. “Beyond that, one of the other big inputs to costs is taxes. The data center asset class operates with a very high percent of assets that are personal property. Servers have a refresh cycle. They last three to five years. As a result, your tax framework will have a big impact on site selection.”
Not everyone thinks these incentives are a good idea. North Carolina’s lack of transparency around data center incentives suggests that the state is “hiding costs and exaggerating benefits,” thereby putting the state’s budget at risk, according to a September op-ed in the News & Observer of Raleigh penned by Goodjobsfirst executive director Greg LeRoy and Brian Balfour of the Locke Foundation. Both nonprofits oppose corporate subsidies.
“It’s a fool’s race to the bottom,” LeRoy says in an interview. “This is not a struggling industry. These are enormously profitable and valuable companies doing something that makes a ton of profit for them.”
The companies would build these data centers with or without tax subsidies and abatements, he says. Citing the loss of tax revenues and the potential strain on local infrastructure like schools and roads, “I think North Carolina would win by losing data centers.”
Some residents of potential sites for data centers share that view. Opposition to projects has blocked projects in various N.C. locations, including some economically struggling ones. In September, Tarboro voted to reject a proposal to build a $6.4 billion data center project in Edgecombe County, citing concerns over its impact on water, sewer, and quality of life issues. The county’s 6.1% unemployment rate in September was the state’s highest.
Akerman’s Grice offers a more positive view of the impact of giant server farms. “You have substantial upgrades (to infrastructure) that wouldn’t be done except for the data center. Typically, [the data centers] pay a higher than regional average salary. The investment is so material they’re not going to walk away overnight. They’ll stimulate the local economy. They don’t have smokestacks. They don’t have as much impact as a new manufacturing facility would. They tend to have a clustering effect — other companies show up.”
The data center industry contends there is a significant multiplier effect of as much as $7 of economic activity for every dollar invested. Good, bad, or indifferent, the tax incentive ship seems to have sailed among political and economic development leaders. As Rojahn puts it, “Tax incentives are a prerequisite for states to be competitive.”
Critics note other issues, most prominently the lack of significant long-term employment opportunities at data centers once they’re up and running, and the potential impact on ratepayers from all this demand for electricity.
“Hyperscale data centers actually don’t employ many staff — typically in the 50 to 150 range,” says Synergy Research Group’s Dinsdale. “For reasons of efficiency, security and cost, there is a lot of software and automation, rather than staff. The big employee numbers are in the construction phase, when there might easily be a thousand workers or many more.”
As for consumers, they are starting to notice the impact of all this construction on their electric bills. A recent study from Carnegie-Mellon and NC State universities reported that “data center and cryptocurrency mining growth through 2030 could increase average U.S. electricity costs by 8%” though the impact will vary by region and the sources of power generation. A Bloomberg analysis in September found that “electricity now costs as much as 267% more for a single month than it did five years ago in areas located near significant data center activity.”
One thing for sure is it will increase demand. Duke has more than 100 prospective customer projects of all types in the Carolinas with a base demand of 50 megawatts or greater. Historically, Duke says, 10 megawatts was considered a large load.
Dot-com reprise?
In tech land, any trend that persists over a multiyear period is almost inevitably deemed a “bubble.” AI and data centers are now entering that territory, prompting comparisons with the dot-com bubble in 1999-2001 that ended with the NASDAQ index crashing nearly 80% from its peak. Key differences are that the Internet was early in its adoption curve 25 years ago, while today’s AI data centers are being financed by massive companies such as Amazon, Apple, Google and Meta. There may be an excess of capital going into the sector, but the businesses are real.
Of course, AI could change everything and still leave behind hundreds of billions of dollars in stranded capital, which occurred during the early Internet era. San Jose, California-based Cisco Systems is a leading networking equipment operator, and operates a large site in Durham. It had the stock market’s largest valuation in March 2000, briefly topping $80 per share. It lost more than 80% of its value by late 2002. Twenty-three years later, it has rebounded to the $70 range.
Many following the tech industry remain resolutely bullish. “I don’t sense that we’re going to see a slowdown in the immediate future,” says Akerman’s Grice. “You have cloud [computing] growing at 18% compounded annually, and AI cloud is projected to grow
at 36%.”
The next big step in AI is expected to be “inference,” which refers to how technology is deployed once the robotic models have been completely trained.
“There are always going to be examples of data center developers getting it wrong in terms of scale, timing, demand or logistics, but the overriding picture is one of there not being enough data center capacity,” says Dinsdale from Synergy Research.
In August, Open AI CEO Sam Altman told a group of reporters, “When bubbles happen, s
mart people get overexcited about a kernel of truth. Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes.”
Venture capitalists put $192.7 billion into AI startups this year through late September, according to data provider Pitchbook. That was more than half of total VC investment during the period. Meanwhile, research from the real estate firm JLL found that the data center vacancy rate stood at a low 2.3% in North America.
In his Pulitzer Prize-winning book, Guns, Germs, and Steel author Jared Diamond describes how geography and environment shaped early civilization. If the hype is to be believed, artificial intelligence will soon supplant those forces in defining the future. That intelligence doesn’t exist in the physical world, but rather within a virtual geography of servers and fiber optic cables.
While AI may be agnostic as to location, it is matters very much to states, towns, and counties looking for a lifeline in a post-industrial world. Like Richmond County.