Published on: 2026-08-21
Updated on: 2026-08-21
Electricity markets match power supply with demand, deciding which generators run and what electricity costs at different locations. Prices can change sharply when demand rises, or transmission becomes congested, limiting how much electricity the grid can deliver where it is needed. AI data centres are putting those mechanics under greater pressure, with a single facility capable of adding hundreds of megawatts of demand at one location and affecting both today's generator dispatch and future investment in generation and grid infrastructure.
Day-ahead and real-time electricity markets help grid operators match generation with expected and actual demand.
Wholesale electricity prices differ by location when transmission constraints make some areas more expensive to serve.
AI data centres can change generator dispatch and congestion because they add substantial demand at specific points on the grid.
Future data-centre demand can also increase the need for generation, transmission and dependable capacity before a facility begins operating.
In many organised U.S. markets, the process has two main stages. The day-ahead market schedules electricity for the following day based on expected demand and generator offers. The real-time market adjusts those schedules as actual conditions change, including unexpected demand, generator outages, weather changes and transmission constraints.
Grid operators update real-time dispatch every few minutes because supply and demand must remain closely balanced. These are wholesale markets, so prices differ from the retail electricity rates households and businesses ultimately pay.
Price formation starts with the cost of available generation. Operators generally dispatch lower-cost resources before more expensive ones, subject to the grid's physical limits.
For example, if two generators can supply power at $25 and $40 per megawatt-hour but rising demand requires another plant offering at $70, the higher-cost generator becomes necessary to meet demand. That additional unit helps set the marginal cost of supplying electricity.
The same logic explains why wholesale power prices can rise quickly during periods of high demand: the system may have to call on increasingly expensive generation.
Electricity cannot always move freely from the cheapest generator to where demand is highest. Transmission lines, transformers and substations all have operating limits.
In many organised markets, these physical conditions are reflected through a Locational Marginal Price, or LMP. FERC describes LMP as reflecting three components: the cost of supplying the next unit of energy, transmission losses and congestion.
Congestion becomes important when part of the network cannot carry all the electricity the market would otherwise send through it.
A low-cost generation is available 200 miles from a major demand centre. Normally, transmission lines can carry that electricity into the area. Now add a new 500 MW load.
If the transmission corridor is already close to its limit, the grid operator may be unable to send enough additional low-cost power through it. A more expensive generator closer to the demand may have to increase production instead.
As a result, two locations in the same regional market can have different wholesale electricity prices during the same hour.
A large new load can therefore affect prices very differently depending on where it connects. An area with spare generation and transmission capacity can absorb new demand more easily than one where the network is already constrained.
What makes the current buildout difficult for grid planners is the combination of project size, location and development speed.
Lawrence Berkeley National Laboratory estimates that U.S. data centres could consume 649 TWh of electricity in 2030 in its reference case, equal to 11.8% of total U.S. electricity use. Its sensitivity scenarios range from 9.5% to 15.3%.
Those figures cover all data centres, not AI alone. However, Berkeley Lab estimates that AI servers could account for 55% of total data-centre electricity consumption by 2030 in the reference case and identifies them as the main driver of projected demand growth. Where that demand lands matters as much as the national total.
Adding 500 MW across a broad region is different from placing the same load behind a limited number of substations and transmission connections. A major data-centre project can therefore create a local grid challenge even when the wider region still has substantial generation capacity.
Timing adds another constraint. Data centres can sometimes be developed faster than major power plants and transmission projects, which may require years of planning, permitting and construction. FERC has highlighted both the growing size and concentration of large loads and their demand for faster grid connections.
Once a data centre starts consuming electricity, the grid has to serve the additional load. If enough low-cost generation and transmission capacity are available, the immediate price effect may be limited. If the network is already tight, the additional demand can require more expensive generators to run or make an existing transmission constraint more severe.
Transmission networks are interconnected. Adding demand at one point changes generation schedules and power flows across the wider system, which can tighten constraints elsewhere.
Large new load → more generation → changing power flows → tighter congestion → different dispatch → changing local prices
A data centre hundreds of miles away is not literally taking electricity from another customer. Its demand changes how the interconnected network has to be operated, which can affect which generators run and where congestion appears.
Some U.S. regions use capacity markets, which pay eligible resources for committing to be available for future electricity needs rather than paying only for electricity actually produced. FERC identifies PJM, MISO, NYISO and ISO New England as the four organised U.S. regions with centralised capacity markets. Other regions use different resource-adequacy systems.
If expected data-centre demand raises future peak needs, the system may require more dependable generation or other resources to maintain reliability. A proposed facility can therefore influence planning and capacity needs before it consumes its first megawatt-hour.
The same forecast can affect transmission investment. If existing lines and substations cannot serve the expected load, upgrades or entirely new infrastructure may be required.
There is no single solution. Additional demand can be met through some combination of new generation, transmission upgrades, storage and demand response. Large users may also agree to reduce consumption during stressed periods or pay for network upgrades required to connect them.
Cost allocation is an important part of that process. Building infrastructure around a proposed data centre creates a problem if the project is later delayed, downsized or cancelled: someone still has to pay for the equipment that was built.
Regulators are therefore reviewing how very large loads should connect and how those costs should be assigned. In June 2026, FERC required the six regional grid operators under its jurisdiction to justify their existing large-load tariffs or propose reforms covering issues including reliability, cost allocation and connection rules.
The underlying problem will remain even as individual rules change. Electricity systems must connect growing demand without creating avoidable congestion, reliability problems, or infrastructure costs elsewhere on the grid.
Electricity markets must balance generation and demand while staying within the grid's physical limits. Wholesale prices reflect both the cost of producing electricity and whether the network can deliver it where it is needed, while capacity arrangements in some regions help prepare for future demand.
AI has not changed those basic mechanics. It has increased the scale, concentration and speed of the new demand those markets increasingly have to accommodate.
