News · OpenAI adds a 1GW Stargate campus in Saline Township, Michigan

Jul, 84 min to read
Automation

OpenAI adds a 1GW Stargate campus in Saline Township, Michigan

The Oracle-partnered site pushes Stargate past 8 gigawatts of planned capacity, but the jobs math and the infrastructure design tell a more specific story about what these campuses actually are.

What OpenAI committed to in Michigan

The announcement places a new Stargate campus in Saline Township, Michigan, as part of OpenAI's 4.5GW partnership with Oracle. OpenAI says the site brings Stargate to over 8 gigawatts of planned capacity and more than $450 billion in investment over the next three years, which it frames as running ahead of the $500 billion, 10-gigawatt commitment made in January.

The campus is being developed by Related Digital, with construction expected to begin in early 2026. It joins previously announced sites in Texas, New Mexico, Wisconsin, and Ohio. This is capacity accounting more than a product announcement: the news is another block of power and floorspace added to a running total, in a state OpenAI describes as central to American engineering and manufacturing.

The jobs figure is a construction figure

The only headcount OpenAI cites is 2,500 union construction jobs. That is a build-phase number, tied to the period beginning in early 2026, not a measure of how many people run the campus once it is live. The announcement is careful about this wording, and the distinction matters for anyone reading the Michigan news as a durable employment story.

A gigawatt-scale AI campus is, by design, a lightly staffed facility during operation. The economic activity comes from the buildout and the supply chains feeding it — energy systems, cooling, industrial equipment — which the announcement gestures at when it talks about 'spurring investment in modern energy and industrial systems.' The reindustrialization language describes the construction and supply pipeline, not a workforce that stays after the concrete cures.

How the campus was designed to avoid local friction

The specifics on water and power read as pre-emptive answers to the objections these projects usually draw. OpenAI says the site will use a closed-loop cooling system that significantly reduces water consumption, and that DTE Energy will serve the campus using existing excess transmission capacity, 'avoiding impacts on local energy supply.'

any upgrades required to support operations will be funded by the project and not local ratepayers.Montana Labs

That sentence is the clearest signal of what OpenAI has learned from data-center siting fights elsewhere. Water use and rate increases are the two levers communities push back on hardest, and the announcement addresses both directly. Whether existing excess transmission actually covers a 1GW-plus draw is the kind of claim worth watching once operations begin.

The automation reality behind the reindustrialization pitch

The specific implication of the Michigan campus is that AI 'reindustrialization' looks different from the industrial base it invokes. The value here is compute capacity that automates cognitive work at scale; the permanent local footprint is a power-hungry, sparsely-staffed building. OpenAI ties the site to benefits like 'better healthcare to improved public services,' but those are downstream of the models, not of the campus itself.

For teams building on this infrastructure, the takeaway is concrete: capacity is being added ahead of schedule, concentrated in a handful of states, and structured to keep the operating burden off local grids and water tables. The 2,500 construction jobs are real and time-bound. The lasting output is automated compute — which is precisely what makes the factory-town framing an imperfect fit for what Saline Township will host.

Find this story relevant to you?

Contact us to find a unique solution

Contact us

Need an AI engineering partner that can actually build?

We help businesses integrate AI, build AI-powered products, automate high-value workflows, and modernize the software systems behind them.

Get in touch

Related reading

More analysis around product delivery, operational AI, and the systems work that makes deployment hold up in reality.

Jul, 134 min to read
Automation

OpenAI reframes adoption as a 'capability overhang' problem

Jul, 134 min to read
Automation

Cisco built the majority of its AI Defense product with Codex writing the code

Jul, 134 min to read
Automation

Commonwealth Bank standardizes on ChatGPT Enterprise as the shared surface for 50,000 employees