Digital Infrastructure Is Now an Energy Infrastructure Business
For most of the history of the internet, the electricity system sat quietly in the background of digital infrastructure.
Data centers required power, of course, but the scale of demand rarely forced difficult decisions. Utilities built generation gradually. Transmission networks expanded over time. Costs were absorbed into the broader rate base and recovered across millions of customers.
The system assumed something simple. Electricity demand would grow steadily and predictably.
Artificial intelligence is beginning to break that assumption.
Across the United States and Europe a new generation of data centers is appearing that requires electricity at a scale utilities rarely planned for. Individual facilities can require power equal to a small city. Entire campuses can approach the demand of a metropolitan region.
The technical challenge is obvious. Electricity systems must build more infrastructure.
The deeper issue is structural. Power systems were designed to allocate costs socially. Hyperscale demand forces regulators to decide whether that model should continue.
Understanding that structural rule is essential for anyone evaluating digital infrastructure.
The Structural Rule
Electricity systems operate according to a cost allocation framework that has been remarkably stable for decades.
Utilities recover the cost of generation, transmission, and distribution infrastructure through regulated tariffs. Those costs are placed into the rate base and spread across the customers who use the system.
The result is a form of collective financing.
A new power plant, a transmission line, or a substation upgrade is rarely paid for by a single customer. Instead the investment is absorbed into the overall electricity system and recovered gradually through rates charged to households and businesses.
This model works well when demand grows slowly.
A new factory opens. A suburb expands. A commercial district adds new buildings. The system adapts over time and infrastructure costs are distributed across a broad population of users.
Electricity networks were built for that kind of growth.
What they were not designed for was the sudden arrival of industrial scale loads that rival the electricity demand of entire cities.
Hyperscale Demand Breaks the Model
Artificial intelligence infrastructure has introduced a new type of electricity demand.
Traditional enterprise data centers often operated in the range of ten to twenty megawatts. Hyperscale AI facilities routinely exceed one hundred megawatts. Some proposed campuses are several times larger.
A single hyperscale facility drawing one hundred megawatts can consume as much electricity as roughly one hundred thousand households. Proposed multi building campuses can push into the hundreds of megawatts or even approach gigawatt scale.
Electricity systems experience three immediate stresses when loads reach this scale.
The first is generation capacity. If enough new load appears, the system must build new power plants simply to maintain reliability.
The second is transmission infrastructure. Massive facilities require substations, high voltage connections, and reinforcement of regional transmission lines.
The third challenge is operational stability. Electricity systems must maintain constant balance between supply and demand. When large industrial loads suddenly disconnect from the grid or rapidly ramp their demand, that balance becomes more fragile.
Grid operators in the PJM region have already experienced incidents where dozens of data centers switched to backup generation at the same time after a transmission disturbance. The sudden drop in demand forced the operator to scramble to rebalance the system.
These operational concerns are not theoretical.
They illustrate how hyperscale computing is beginning to interact with the electricity system in ways the system was not designed to accommodate.
The Three Regulatory Responses
Faced with this new demand profile, regulators are beginning to experiment with several approaches.
Each approach attempts to answer the same question. Who pays for the infrastructure required to support hyperscale demand.
One response is the creation of separate electricity tariffs for very large loads.
Delaware is currently considering such a structure through a proposal by Delmarva Power to create a new tariff category for customers drawing more than twenty five megawatts of electricity. The goal is to define how transmission upgrades, generation investments, and other system costs will be assigned when extremely large users connect to the grid.
The debate in that proceeding focuses on whether residential customers should bear the cost of infrastructure built primarily to serve hyperscale data centers.
A second response is requiring developers to finance dedicated infrastructure.
In several jurisdictions utilities are asking large customers to fund generation capacity or transmission upgrades directly. Some projects now require long term contracts, financial guarantees, or minimum demand commitments before utilities will begin construction of new infrastructure.
These mechanisms shift financial risk away from the broader rate base and onto the customer creating the demand.
The third approach is encouraging or requiring large customers to secure their own electricity supply.
Regional grid operator PJM recently proposed revisions to its rules governing behind the meter generation, a structure that allows facilities with onsite power plants to offset their demand from the public grid. The proposal would limit how large new data center loads can use that structure while establishing new transmission service arrangements for colocated generation.
The policy debate surrounding these rules reflects a larger shift. Data center developers are increasingly exploring onsite generation, microgrids, and direct power purchase agreements in order to avoid delays or cost disputes associated with traditional utility interconnection.
Each of these regulatory approaches reflects the same underlying reality.
The traditional cost allocation model for electricity systems is being reconsidered.
Why Geography Will Change
As regulators experiment with these new frameworks, the geography of digital infrastructure will begin to shift.
Some regions will adapt quickly.
These markets will create clear tariff structures for large loads. They will establish predictable interconnection processes and transparent cost allocation frameworks. Developers will be able to evaluate electricity pricing and infrastructure obligations with reasonable certainty before committing capital.
Other regions will move more slowly.
In those markets the debate over who pays for infrastructure will continue to unfold through regulatory proceedings, legislative proposals, and public opposition to data center development. Interconnection queues will lengthen and infrastructure planning will become more uncertain.
The result will be a widening gap between power secure regions and regulatory friction regions.
Developers will follow the path of least uncertainty.
Electricity markets that provide clear rules for hyperscale demand will attract investment. Markets where cost allocation remains contested will struggle to move projects forward.
The New Foundation of the AI Economy
The rapid expansion of artificial intelligence has produced extraordinary attention on chips, models, and computing architectures.
Yet the deeper constraint on the AI economy is far more physical.
Electricity infrastructure determines where computation can occur.
Transmission planning determines how quickly new facilities can connect. Tariff design determines who pays for the infrastructure required to support those facilities. Regulatory frameworks determine whether hyperscale demand is treated as a shared system investment or a private industrial load.
These decisions will shape the geography of digital infrastructure far more than software architecture.
The first rule of AI infrastructure is simple.
Power must be contracted before it exists.
The next decade of digital infrastructure will be defined not only by advances in computing technology but by the design of electricity markets capable of supporting it.
Artificial intelligence may be built from code and silicon.
But its physical foundation is energy.