Compute Scarcity and Grid Pressure: The Microeconomics of AI Energy Externalities

Global data centre electricity consumption rose, reaching 17% (or nearly 6 times greater than overall electricity consumption increase of 3%) in 2025 per IEA; the problem lies in economics. The growth in AI technology has surpassed the ability of our grid to supply low-cost electricity; therefore, the cost of that economic gouging will fall on those people who have not requested to pay for it. AI uses two different types of resources as limiting factors: chip supply and electricity supply. Advanced chip supply, while it is an impediment to growth, is not the only impediment to growth in AI technology. Large-scale training and operations of large AI models require large-scale, concentrated sources of electrical supply at one location (as opposed to distributing that electrical supply evenly over larger areas). Due to that requirement, power grid operators must build additional substations and transmission lines much sooner than they would have previously. Who pays for those new infrastructure builds will determine who bears the cost of the shortage caused by the limitation of electricity supply. How we assess or bill the construction and maintenance costs of that new infrastructure will determine how the costs are borne by everyday ratepayers.

As of now, most states in the United States have charged all consumers for the electric lines that supply energy to large data centres. Historically, the costs to build electricity distribution lines to supply a small number of data centres (AI facilities) were allocated to all customers, regardless of whether they used any AI services or not. As a classic example of a negative externality, one party creates a cost through its activity, but another group (or groups) must pay for that cost even though they were never involved in the transaction itself. The social cost of the services provided to a data centre exceeds the price charged to that data centre; this gap represents another form of market failure.

Virginia has required data centres to pay for their own power lines

Virginia's State Corporation Commission ordered data centres to pay for transmission infrastructure built solely for the data centres, to prevent transmission costs from being passed along to the rest of the customer base. According to the office of Governor Abigail Spanberger, the change is expected to save Virginia households and small businesses hundreds of millions of dollars. In terms of economics, the ruling internalises an externality by placing the cost of the electricity necessary to run AI onto those operators instead of allowing households who have nothing to do with AI to pay.

Cryptocurrency mining is competing for the same limited available power

AI is not the only industry competing for low-cost and abundant electricity, with limited connections to the electricity grid. Cryptocurrency mining is also finding itself in competition under essentially the same constraints, as well as facing its own cost-shifting question. For example, in Texas, grid operators are incentivising miners to shut down their machines during peak periods of demand, with part of the cost being borne by the regional customer base. This presents a current political battle over who pays for the keeping of specialised computers in a low-cost electricity region. Companies such as SHR Miner and other cloud mining services allow retail customers to purchase computational power without needing to own or site the servers themselves; similarly, many AI users now rent cloud computing resources instead of constructing their own data centres. For both industries, the entity that can secure the most affordable and dependable source of electricity gains a significant long-term cost advantage, while grid capacity expands at a rate much slower than the demand created by either of these industries.

Regulators face the challenge of balancing two competing priorities: providing the investment and jobs that data centres and cryptocurrency mining operations generate, while simultaneously preventing the regular household from inadvertently subsidising another's computation costs. Virginia has demonstrated an avenue by which to bridge this gap by requiring the operators responsible for generating the cost to directly bear that cost. The rest of the world, including those states contemplating similar situations, remains undecided as to who will be responsible for the cost of their electricity supply.