The Harwood data center vote delivered a rare win for the industry at a moment when local opposition is rising across the United States. In the North Dakota town of roughly 940 residents, about 75% of voters rejected a charter amendment that would have added new procedural hurdles for large industrial projects.
The unofficial result matters well beyond one small town. Harwood was the only place out of 45 data center-related ballot measures nationwide where voters sided with the industry, underscoring how permitting and community pushback are becoming material risks for AI infrastructure buildouts.
For Applied Digital, the outcome preserves momentum for Polaris Forge 2, a $3 billion, 280-megawatt AI campus on the edge of Harwood. For investors, it offers a case study in how local politics can influence timelines, tax revenue assumptions, and the economics of large-scale data center development.
Key Facts
- About 325 of roughly 430 ballots cast in Harwood went against Measure 2, equal to around 75% of the vote.
- Harwood was the only pro-data center result among 45 data center ballot measures tracked across the United States.
- Applied Digital’s Polaris Forge 2 project carries a planned investment of $3 billion and 280 megawatts of capacity.
- Applied Digital disclosed a 15-year lease worth about $5 billion for 200 megawatts with an unnamed hyperscaler in October 2025.
- The company had closed $2.15 billion in financing for the campus by March 2026.
Harwood data center vote
At the center of the dispute was Measure 2, a proposed amendment to Harwood’s Home Rule Charter. The measure would have required independent reviews and public hearings for large industrial projects. Supporters argued the change was about process and community oversight, especially after residents said they learned of the AI campus only weeks before local approval.
Opponents, including town leaders and regional business advocates, framed the amendment as a threat to investment and future tax revenue. Harwood’s mayor had said the data center could generate $1.65 million in taxes by 2028, while critics of the measure argued added requirements could deter development or encourage the project to seek a different jurisdictional structure.
The result is significant because it highlights the diverging pressures shaping the AI infrastructure trade. On one side are hyperscalers, cloud providers, and data center operators racing to secure power, land, and financing for compute capacity. On the other are communities increasingly focused on land use, electricity demand, water needs, noise, and the speed at which large projects move through local approval processes.
Harwood’s vote shows that local permitting has become one of the most important variables in the economics of AI data center expansion.
Why Harwood mattered to Applied Digital
Applied Digital chose Harwood in part for practical and financial reasons: a cooler climate, access to energy, and a business-friendly environment. North Dakota also exempts key equipment such as computers, cooling systems, and servers from the state’s 5% sales tax, improving project economics for capital-intensive facilities.
Those advantages help explain why the company moved quickly. The city council approved Polaris Forge 2 in September 2025, and construction began the same month. The subsequent lease announcement and multibillion-dollar financing package suggest the project is not a speculative land play but a core part of the company’s push into AI-oriented infrastructure with long-duration revenue potential.
Implications for Investors
The Harwood result is constructive for Applied Digital and for investors tracking AI infrastructure names, but it does not erase the broader warning from the other 44 ballot measures. The main lesson is that demand for data center capacity is not the only driver of valuation. Local approval risk, power availability, and political resistance can all affect project timing, capex schedules, and customer onboarding.
For companies tied to hyperscale and AI buildouts, investors should watch whether local communities seek moratoriums, stricter zoning standards, or new review requirements. Even where projects ultimately move ahead, delays can change financing costs and reduce the value of signed leases if capacity comes online later than customers expect. That risk extends beyond developers to utilities, construction suppliers, chip-linked infrastructure plays, and landlords exposed to data center pipelines.
There is also a policy angle. States that combine lower taxes, supportive permitting, and reliable power access may attract a larger share of AI infrastructure investment, while restrictive jurisdictions could see projects migrate. In that sense, Harwood may be less an isolated vote than a signal that geography will matter more in separating winners from laggards across the data center ecosystem.
The next phase for investors is to track execution: construction progress, customer disclosures, financing terms, and any new local policy efforts that could alter the operating landscape. Demand for AI compute remains strong, but Harwood shows that the road from announced project to revenue-producing asset increasingly runs through town halls as much as corporate boardrooms.