Compute Is the New Car
What the data center fights in Michigan tell us about the next industrial economy
In December, Michigan’s utility regulators approved a contract to power a 1.4-gigawatt data center campus south of Ann Arbor. The project, code-named “Stargate,” will sprawl across 575 acres of former farmland in Saline Township. Oracle and OpenAI are the names attached. The price tag is seven billion dollars.
Most of the coverage treated this as a tech story: AI comes to the Midwest, Silicon Valley meets the Rust Belt, the usual tropes. But if you look at the actual numbers—the megawatts, the concrete, the water, the electricians—the story is different. It’s not a tech story at all.
It’s a heavy industry story. And that’s what makes it interesting.
The Math
A data center running at 1.4 gigawatts isn’t a server farm in someone’s basement. It’s a factory. Let’s do the arithmetic.
Data centers don’t shut off at night. They run continuously, typically at 85 to 95 percent of their rated capacity. Assume 90 percent for Saline. That gives you:
1.4 GW × 0.90 × 8,760 hours/year ≈ 11 terawatt-hours per year
To make that number legible: the average American household uses about 10,500 kilowatt-hours per year. Divide 11 trillion watt-hours by 10,500 and you get roughly one million households. One data center, consuming electricity equivalent to a million homes.
Now compare it to Michigan’s total retail electricity sales, which run about 99 terawatt-hours annually. This single project represents approximately 11 percent of the state’s entire electricity consumption.
That’s not software. That’s the scale of a steel mill. That’s the kind of load that reorganizes a regional grid.
The auto industry didn’t transform Michigan because people liked cars. It transformed Michigan because building cars required factories, which required power plants, which required transmission lines, which required thousands of people who knew how to weld and wire and pour concrete. The demand created the infrastructure. The infrastructure created the workforce. The workforce created the capability to build other things.
Data centers work the same way. The “product” is different—computation instead of Buicks—but the physical economy underneath it is eerily similar: megawatts, cooling systems, high-voltage distribution, skilled trades, construction crews, long-term contracts that justify capital investment.
The question is whether Michigan understands what it’s actually being offered.
The Auto Analogy, Taken Seriously
Here’s the parallel, stated plainly:
Automotive era:
The scarce resource was mass production capacity.
The infrastructure buildout was highways, refineries, factories, suburbs.
The workforce need was machinists, line workers, mechanics, electricians.
The region that won was the one that assembled land, labor, logistics, and energy into a coherent package faster than anyone else.
AI era:
The scarce resource is compute capacity.
The infrastructure buildout is transmission lines, substations, cooling systems, fiber routes.
The workforce need is electricians, pipefitters, HVAC technicians, power systems engineers.
The region that wins will be the one that assembles land, labor, logistics, and energy into a coherent package faster than anyone else.
The U.S. Department of Energy estimates that American data centers consumed about 176 terawatt-hours in 2023—roughly 4.4 percent of total U.S. electricity. By 2028, that figure could reach 325 to 580 terawatt-hours, or 6.7 to 12 percent of national consumption. That’s not a linear growth curve. That’s an industrial revolution compressed into half a decade.
Converting those terawatt-hours to average load: we’re talking about adding something like 17 to 46 gigawatts of continuous demand to the American grid in three to five years. That’s dozens of new power plants. Thousands of miles of transmission. Hundreds of thousands of skilled workers to build and maintain it.
The parallel to automotive isn’t poetic license. It’s structural. Both represent demand shocks large enough to reorganize physical infrastructure around a new core machine.
Why Michigan
The Midwest has a lot of cheap land and desperate municipalities. What makes Michigan different?
Water.
High-density AI compute is a thermal problem as much as a computational one. Modern GPU clusters generate enormous heat. You have to move that heat somewhere, and water is the most efficient medium for doing it.
The Great Lakes hold about 20 percent of the world’s surface freshwater. More importantly, the Great Lakes Compact—an interstate agreement with the force of law—bans large-scale diversions of that water outside the basin. You can’t pipe Lake Michigan to a data center in Arizona. If you want Great Lakes cooling, you bring the load to the water.
For the first time in decades, the region’s defining asset isn’t a liability (too cold, too gray, too far from the coasts). It’s a structural advantage that can’t be replicated.
Energy.
Michigan is simultaneously rebuilding its grid and restarting its nuclear capacity. The Palisades plant—the first commercial reactor in U.S. history to come back online after being shut down—is expected to return to service in 2026, backed by a $1.52 billion federal loan guarantee. Holtec, the owner, has also received $400 million in DOE funding to deploy two small modular reactors at the same site, targeting 2030.
Do the math again: Palisades at 800 megawatts plus two 300-megawatt SMRs gives you 1.4 gigawatts of firm, carbon-free generation. That’s almost exactly what the Saline data center needs. The alignment isn’t coincidence. The state is betting that large industrial loads can anchor the business case for the kind of generation that residential demand alone can’t justify.
Workforce.
A data center isn’t a coder’s playground. It’s an electrician’s jobsite. The permanent technical staff is modest—maybe a few hundred people—but the construction crews, the substation installers, the fiber technicians, the HVAC specialists, the ongoing maintenance contracts: those jobs are numerous and durable.
Michigan already has a deep base of skilled trades workers, a legacy of the auto industry. The state’s infrastructure workforce plan aims to train 5,000 new workers by 2030 for jobs in construction, energy, and broadband. The data center boom provides the demand signal that keeps those training pipelines full.
This is the bridge people miss when they complain that data centers “don’t create enough jobs.” They’re correct about the permanent headcount inside the fence line. They’re wrong about the secondary effects. Data centers are anchor tenants for a much larger ecosystem of construction, energy, and maintenance work—the same ecosystem you need to build anything else heavy.
The Conflicts
None of this is automatic, and plenty of people are rightly skeptical.
Who pays?
The central tension in every data center deal is cost allocation. If a massive new load requires billions in grid upgrades, who bears that expense—the data center, or existing ratepayers?
Michigan’s regulators attached conditions to the Saline contract: the developers must fund major upgrades, including over 1,300 megawatts of battery storage. The commission cited a net benefit to other customers of approximately $300 million, based on the logic that spreading fixed grid costs over more kilowatt-hours reduces the per-unit burden on everyone else.
That math is real, but it’s not guaranteed. If data centers arrive faster than new generation, or if contracts are structured badly, rates can spike. Other states have already hit this wall. Ohio’s grid became so congested from data center demand that regulators imposed a special tariff in July 2025 requiring new facilities to post collateral and commit to long-term purchases. Indiana’s attempt to build a tech mega-site near Lebanon ran into furious opposition over water sourcing—proposals to tap aquifers along the Wabash River sparked lawsuits and protests.
Michigan isn’t immune to these risks. It’s just managing them earlier in the cycle.
Water usage.
Data centers drink. Evaporative cooling systems can consume millions of gallons per day. Even in a water-rich region, that matters at the local level—aquifers can be stressed, municipal systems can be strained.
The answer isn’t to reject data centers. It’s to require efficient cooling technologies, closed-loop systems, and real-time monitoring. The Great Lakes offer abundance, but abundance isn’t infinity.
Speed versus deliberation.
The Saline project was approved on an expedited schedule, without a full contested hearing. Consumer advocates and environmental groups objected. They weren’t wrong to want more scrutiny. They also weren’t wrong that delay has its own costs: projects go elsewhere, and Michigan loses the chance to shape how this industry develops in the region.
There’s no clean answer here. Governance is always a tradeoff between speed and caution. What matters is whether the protections in the contract actually hold—whether the exit fees are enforceable, whether the cost allocation sticks, whether the promised benefits materialize.
We’ll find out.
The Broader Case
Here’s the part that doesn’t fit neatly into a regulatory docket or a news cycle:
The Great Lakes region spent fifty years being told that its industrial past was a handicap. Heavy industry was dirty, old economy, left behind. The future was in software and finance and places with better weather.
That story was never entirely true, and it’s becoming less true by the year.
The world is relearning that physical infrastructure matters. Supply chains are shortening because companies discovered that saving five percent on unit cost doesn’t help when your shipment is stuck in a port for three months. Climate change is repricing geography—reliable freshwater and moderate temperatures are becoming premium assets, not afterthoughts. And AI, for all its abstraction, turns out to require the most concrete things: electricity, cooling, construction, skilled hands.
Michigan isn’t going to out-compete Silicon Valley on software. It shouldn’t try. But it can compete on the physical economy that software depends on. It can offer what the coasts increasingly can’t: water, power, land, trades workers, and a regulatory environment that takes heavy industry seriously.
Data centers are just the first wave. Behind them are advanced manufacturing facilities, battery plants, hydrogen production, industries that need the same ingredients. The question is whether the region uses this moment to rebuild the capacity to build—not just data centers, but the next generation of things we haven’t named yet.
The Box on the Workbench
In “The Iron and the Mesh,” I wrote about Maria’s shop in Toledo, and the small box sitting on her workbench—a sensor that could connect her lathe to a regional network, help her find work, help other shops find her. She hadn’t decided whether to plug it in.
The data center story is the same choice at a larger scale. The box is bigger. The stakes are higher. The skepticism is justified.
But the question is the same: do you trust that this time, the tools are real?
The lakes will still be here in fifty years. The water will still flow. The seasons will still turn. The question is whether the people who kept the knowledge alive—who know how to weld and wire and build—whether they get infrastructure that actually makes their work easier, or just another round of promises that evaporate when the money moves on.
The data center boom is a bet that this region can anchor the next industrial economy. It’s not a guarantee. It’s a chance.
And chances have to be taken.
If you’re in Michigan and you’ve followed these proceedings more closely than I have, I’d welcome corrections and additions. The MPSC docket is public. The debate is ongoing. This is my read on the trajectory, not the last word.