A $16 billion AI data center in rural Michigan has become a test case for what happens when local opposition meets the compute demands of Big Tech.
The fight in Saline Township is not just about one construction site. It is about whether small communities can still say no when a project is tied to OpenAI, Oracle, state economic development goals, utility planning and the national race to build AI infrastructure.
The township, a farming community southwest of Ann Arbor with roughly 2,300 residents, rejected the proposed data center last September. The planning commission and township board both voted against the project, with the board reportedly voting 4-1 against the rezoning request. Residents packed meetings, posted signs and argued that a 21 million square foot data campus would change the character of the area beyond recognition.
Then the legal process took over. Related Digital, the developer behind the project, and landowners sued the township shortly after the rejection, arguing that the decision amounted to exclusionary zoning because the township had no land zoned for industrial use. The township settled within weeks. By November, construction activity was underway.
According to Fortune, the project is tied to OpenAI and Oracle's Stargate AI infrastructure initiative and is expected to become the largest construction project ever undertaken in Michigan. That is why the local vote matters far beyond Washtenaw County. For AI companies, this kind of campus is not a luxury. It is the physical foundation for model training, inference, enterprise AI services and the next wave of products that startups are already building around.
The numbers explain why the project has drawn so much attention. The Saline Township campus is expected to require about 1.4 gigawatts of electricity, a level often compared with the output of a large power plant. Earlier reporting has described the initial footprint at about 250 acres, with financing secured through Related Digital and backing from major financial players including Blackstone and PIMCO.
That scale changes the politics. A typical local zoning fight might be about traffic, farmland or tax revenue. This one is also about whether the United States can build enough AI capacity to support companies like OpenAI and Oracle as they compete with other countries and other cloud providers. Once a project is framed that way, local objections start to look smaller to state officials and utilities, even when those objections are grounded in real concerns.
Michigan Governor Gretchen Whitmer has called the development the largest investment in the state's history. DTE Energy has argued that the data center could produce hundreds of millions of dollars in utility benefits, potentially helping avoid future rate increases if the project performs as planned. Critics, including Michigan Attorney General Dana Nessel, have questioned whether ordinary ratepayers are being asked to accept too much risk on promises that remain hard to verify.
Water is the other pressure point. Developers have said the site will use a closed-loop cooling system intended to limit strain on local supplies, but nearby communities are already moving defensively. Saline and other Michigan municipalities have considered or adopted restrictions on water service for large data centers, while more than a dozen local governments have explored moratoriums as residents try to get ahead of the next proposal.
Backlash Could Reprice AI Growth
The practical lesson for the AI market is simple. Compute may be strategic, but land use is still political. Every large data center needs power lines, transformers, backup systems, road access, water planning, tax arrangements and neighbors willing to live with years of construction. When those pieces become contested, the cost of AI does not stay neatly inside a cloud invoice.
Startups feel this indirectly. Most young AI companies will never negotiate a 1.4 gigawatt power agreement, but they are exposed to the price of compute through cloud contracts, API costs and model access. If communities slow projects, utilities demand more protections or states impose stricter rules, the infrastructure bill behind AI growth can rise. That can make training harder, inference more expensive and margins thinner for companies already building on rented capacity.
There is also a trust problem. When residents see a board vote against a project and construction begins anyway after a lawsuit and settlement, they do not experience that as innovation. They experience it as power moving around them. That perception can spread quickly, especially as data centers arrive in rural areas that were not part of the original technology bargain.
The industry should pay attention to that signal. AI companies cannot build a national infrastructure footprint on the assumption that local resistance is just a procedural delay. If the public starts to see data centers as extractive, taking power, water and land while offering few permanent jobs, the permitting environment will become harder in more states.
Saline Township may still become a major node in the Stargate buildout. But the larger story is no longer only about OpenAI and Oracle needing more compute. It is about whether the next phase of AI growth can be built with communities, or whether it will be pushed through them until the backlash becomes too expensive to ignore.
Also read: Nvidia is turning AI funding into a supply chain weapon • Cloudflare's AI layoffs turn efficiency into a market test • Cheap Claude access is becoming a hidden risk for startups
