Alfred Kahn once spoke of the way in which a regulatory assumption can hold us captive. His assumption was that of the natural monopoly: we imagine a power grid serving all customers equally, as if every megawatt is interchangeable and every load is the same. This assumption comes to mind when we have to give an account of how electricity should be priced and allocated. It is, to borrow a phrase, 'ready-to-hand'.Alfred Kahn, often called the father of airline deregulation, was equally influential in utility rate design theory. His core insight: regulatory assumptions about natural monopoly shape infrastructure policy long after the underlying economics have changed.
From Kahn, and from the century of rate design that followed, we learn that the assumption misleads, rather than clarifies. Consider Century Aluminum's Hawesville plant in western Kentucky: 628 workers producing military-grade aluminum for the U.S. defense supply chain, consuming 482 megawatts of baseload power. For decades, the plant ran on the assumption that industrial load and grid capacity were partners — that a smelter's appetite for cheap, steady power was the kind of customer the grid was built to serve. Then power costs tripled past the $40 per megawatt-hour threshold that makes smelting viable. Closed in 2022. Idle for four years. Substations still humming, transmission lines still live, 750 acres sitting empty.Century Aluminum's Hawesville plant: 628 workers, 482 MW baseload, military-grade aluminum for the U.S. defense supply chain. Closed 2022 when power costs tripled past the $40/MWh viability threshold.
Then TeraWulf, an AI data center developer, bought the whole thing for $200 million cash. Projected new workforce: 100. Same grid footprint. An 84 percent reduction in employment for the same megawatts. Nobody protested. The grid just found a higher-value customer.
To use the terms that I will shortly explain, the grid is not a project of assigning megawatts; it is rather a system in which the whole and the elements are both present from the beginning. Or perhaps it is a project. This is the question.
The Duck and the Rabbit §
Consider: the same grid infrastructure can appear as a market or as a strategic asset. It cannot, however, be seen as both at once; our policy moves back and forth between the two. We cannot decide to see just one.
That seeing-as will determine a set of expectations and connections that will sustain different forms of regulation responsive to different questions and moving in different directions. Infrastructure does not change, but the world within which it has meaning changes.
America's power grid is like that duck/rabbit image. We are not just captured by a picture; we are captured by two pictures of the same phenomenon.
These two pictures are that of project and system.
The Grid as Project §
'Project' imagines the grid as the product of regulators who are free agents capable of acting with intention after some sort of deliberation.
For the 'project' imagination, the tariff is the paradigm of grid governance. Ohio figured this out first. In July 2025, the Public Utilities Commission ordered AEP Ohio to create the first standalone data center tariff in the country. Centerpiece: an 85 percent minimum load payment floor. A data center must pay for at least 85 percent of its contracted capacity regardless of actual consumption. Underperform the commitment, and the shortfall stays with the operator, not other ratepayers. This is a tariff designed with intention.Ohio PUCO: first standalone data center tariff in the country. 85% minimum load payment floor — if the data center underperforms, the shortfall stays with the operator, not ratepayers.
A tariff has a beginning in the action of a free subject. That subject explains his tariff by referring to his intentions: protect existing ratepayers, allocate costs fairly, preserve grid stability for industrial users who cannot match hyperscale willingness to pay. An 85 percent floor is a deliberate choice about what kind of grid Ohio wants — one where data centers bear the infrastructure costs they impose, not one where those costs drift to the homeowner's bill.
Michigan took a different path to the same principle. The Public Service Commission approved 1,383 megawatts of hyperscale AI data center load for DTE Energy, with an explicit condition that DTE bears full financial responsibility for any unrecovered costs. Risk stays with the utility that sought the customer, not the residential ratepayer who didn't. A different project. Same core value.Michigan PSC: approved 1,383 MW of hyperscale AI load for DTE Energy, with the condition that DTE bears full financial responsibility for unrecovered costs.
That 'might have been' is critical to projects and thus to strategic control. In a world of projects, we are always thinking of what we might do, what we might have done, and what we might do better. If 'project' delineates a space of strategic control, then 'system' delineates a space of market efficiency.
The Grid as System §
If 'project' is our duck, then 'system' is our rabbit. To explain Hawesville on this account is to give a narrative of how the particular fits into the whole.
Instead of looking backwards to a point of origin in a regulatory decision, we offer a synchronic account of reciprocal support among elements: generators, transmission, load, price signals. A grid is always greater than the aggregate of its elements. The whole gives meaning to the parts.
A market has the curious temporal quality of preceding the elements that constitute its parts. Power's price existed before any particular data center or smelter connected to it. TeraWulf buying Hawesville for $200 million was not an act of disruption; it was the realization of a systemic whole. It grew into itself spontaneously, without plan or intention.
Systems have the capacity for maintenance and some ability of repair. PJM capacity auctions repricing from $29 per megawatt-day to the $333 FERC regulatory ceiling is the market healing itself — finding the price that balances supply and demand. Data center growth driving $7.3 billion in added capacity costs in a single PJM auction cycle is the system doing what systems do: absorbing new demand, repricing resources, reallocating infrastructure to the bidder willing to pay the most.PJM capacity auctions: $29/MW-day to the $333 FERC regulatory ceiling. Data centers added $7.3B in capacity costs in a single auction cycle. The market reprices itself.
Projects, though, ordinarily have no such capacities. When a tariff breaks, we take it to the regulator. When the grid loses a smelter, the system shrugs. The project mourns.
The Historical Pivot §
In a curious coincidence of history, 2022 marks the collision of both project and system in American grid policy.
On the system side: Hawesville closes. The $40/MWh threshold that sustained aluminum smelting for decades becomes uncompetitive against data center load willing to pay multiples of that price. TeraWulf's acquisition follows — $200 million for a site whose value lies not in its buildings or its equipment but in its grid connection. Substations, transmission, 482 megawatts of interconnection capacity. The market speaks. It says: these megawatts are worth more running servers than smelting aluminum.
On the project side: state commissions begin studying data center tariffs. Ohio, Michigan, and a growing number of regulators start designing purpose-built rate structures. They say: the market's answer is not the only answer. Strategic value cannot be captured by price signals alone.
This collision was decades in the making. Flat industrial tariffs gave way to time-of-use pricing. Bilateral contracts gave way to capacity auctions. Manual metering gave way to real-time demand response. Each step moved the grid closer to a pure system — and further from the project-frame that once kept baseload power affordable for heavy industry.
One category error accelerated the collision past the point of no return: tariff structures that treat a 10-megawatt factory and a 500-megawatt hyperscale campus identically. No mechanism existed to distinguish them. Why would it? A megawatt is a megawatt. Now the project frame, arriving late, has to undo a decade of system-frame rate design in a matter of years.
Deep Roots §
These two master narratives are as old as the American grid itself. Consider the Rural Electrification Act of 1936, which describes the government's project: free action of regulators capable of acting on final causes, deliberating, and judging their product. Cheap baseload power for industrial employment and rural development. A bargain.
Like all projects, the grid's original compact took time — decades of rate cases, infrastructure buildouts, and political bargains — and was subject to an external, normative evaluation: did it deliver industrialization and employment? It did. Twenty aluminum smelters across the country. Twelve million manufacturing jobs dependent on affordable baseload power. Steel mills, chemical plants, paper manufacturers, all built on the assumption that the grid was a partner, not an auction house.
Some things did not quite work out as planned — deregulation, for instance — and required new interventions, such as capacity markets.
But the systemic view can also be found in that same history. Picture the wholesale electricity markets of the 1990s and 2000s, humming along, pricing power. Participants do not know the regulator's plan. For them, the grid has the order of a market in which the parts seem naturally to support each other. Primary aluminum smelting capacity fell 80 percent since 2000, from 20 smelters to four. In a system, things cannot be other than they are.USGS data: U.S. primary aluminum smelting capacity fell 80% since 2000 — from 20 smelters to four. The systemic view treats this as the market working as intended.
Modern Oscillation §
Today's battleground is grid capacity allocation, and the numbers are staggering. ERCOT's interconnection queue alone has pending requests totaling 233 gigawatts against an all-time peak demand of 85.5 — more than two and a half times the system's historical maximum, most of it data center load. PJM, ERCOT, and the Southeast are all feeling the same pressure. Every defense-tech conversation this year circles the same bottleneck. Not talent. Not capital. Power. Every gigawatt that goes to an AI training cluster is a gigawatt that doesn't go to the chip fab, battery plant, or smelter the defense industrial base depends on.ERCOT interconnection queue: 233 GW pending against 85.5 GW peak demand — more than 2.5x the system's historical maximum, mostly data center load.
Every choice of tariff structure depends upon a prior choice: system or project?
Neither frame wins outright. We might think that system is closer to the 'facts of the matter', but that is because we live in a market-dominant era. Yet the evidence is seductive. Duke University researchers found that 100 gigawatts of new data center load can be accommodated with 0.5 percent annual curtailment, saving $40 to $150 billion over a decade. E3 calculated that a properly structured large-load tariff generates $3.4 million in surplus revenue per 100-megawatt facility. Under this math, the data center becomes a net contributor to the grid, not a cost-shifter. System thinking makes data centers look like the answer.Duke University: 100 GW of data center load accommodated with 0.5% curtailment, saving $40–$150B over a decade. E3: a structured large-load tariff generates $3.4M surplus per 100 MW facility.
But Ohio's 85 percent minimum load floor is the project-response to the systemic view. Michigan's conditional approvals are the project-response to DTE Energy's market-driven data center ambitions. Wherever there is a dominant discourse of system, a critical response will be framed as project, and vice versa.
Can it be long before every shuttered factory with grid infrastructure is in a position to host a hyperscale data center? How about the defense supply chain — can it survive indefinitely on foreign-sourced aluminum while domestic smelting capacity is converted to server farms? The system says yes — that's efficiency. The project says wait — that's strategic capacity we cannot rebuild once it's gone.
No Final Resting Point §
To which picture — project or system — should we pledge our allegiance? Neither.
Rate design theorists thought that we are held captive by inherited tariff assumptions, and they wanted to liberate us. But they understood that we cannot do without them. We think and allocate through them.
However, there is no transcendental standpoint from which to decide which picture applies. A properly designed tariff can fix the incentives, but it cannot fix the timing. We work through these pictures, never escaping them entirely.
This does not mean that we cannot become aware of their limits and their temptations. We can work through them to a recognition that they do not exhaust the possibilities of meaning. Ohio's minimum load floors and Michigan's conditional approvals are acts of that awareness — regulators recognizing that the system frame alone leaves strategic value on the table, but also that the project frame alone cannot move fast enough to keep up with the market.
In the grid, and perhaps in the larger infrastructure of American industry, there is no final resting point. There is only the continual movement between project and system — a movement that is necessary if we are to maintain our strategic control and our market efficiency.