Stargate heads to Michigan with a gigawatt-scale AI campus

The $500 billion Stargate AI infrastructure project just planted a flag in the Midwest. A new campus in Saline Township, Michigan, will deliver more than one gigawatt of capacity, pushing the total planned Stargate footprint past 8 gigawatts across seven U.S. sites. That puts the OpenAI, SoftBank, and Oracle-backed venture ahead of schedule on its 10-gigawatt target, with over $450 billion in investment now teed up over the next three years.

The Michigan site, developed by Related Digital, is expected to break ground in early 2026 and will create upwards of 2,500 union construction jobs. Oracle is the named cloud partner on the 4.5-gigawatt deal that includes this campus. The project leans on DTE Energy, which happens to have excess transmission capacity in the area—meaning the buildout won’t cannibalize local power supply. Any grid upgrades needed down the line will be funded by the project itself, not Michigan ratepayers.

What’s notable here is the cooling approach. The campus will run on a closed-loop system, a design choice that dramatically cuts water consumption compared to the evaporative cooling that data centers typically guzzle. For a state surrounded by the Great Lakes, that’s more than symbolic. It’s also the kind of detail that matters when you’re trying to sell a hyperscale build to communities still wary of becoming utility subsidizers.

Michigan joins Texas, New Mexico, Wisconsin, and Ohio on the Stargate map. The pitch is straightforward: AI infrastructure as a vehicle for reindustrialization. The press release frames it as a chance to bring advanced manufacturing and engineering back to places that built the last century’s industrial backbone. Whether a data center campus in Saline Township actually delivers on that broader economic promise is an open question. But the jobs, the union labor, and the utility structure suggest the consortium has learned something from the backlash that’s hit other large-scale projects. This one’s designed to look like a neighbor, not a landlord.