Article

The grid queue is part of AI strategy

Grid access belongs inside AI strategy because the capacity and terms a power system can offer may set a data centre's usable scale before computing equipment is installed.

FERC has opened the tariff question

On 18 June 2026, the US Federal Energy Regulatory Commission opened six proceedings under section 206 of the Federal Power Act. It directed every regional grid operator under its jurisdiction to justify or reform the tariff rules used when data centres and other very large electricity users seek to connect.1

These are show-cause orders. They start a process; they do not grant a connection or establish one final national tariff. FERC has made preliminary findings about existing rules and asked each grid operator to answer them. A later order may require changes after the records have developed.

That distinction changes strategy because a proposed data centre cannot treat regulatory attention as delivered electricity. It still needs a route through the applicable tariff and a connection that the power system can operate safely.

The project exists at the power it can receive

An AI data-centre plan may begin with a target number of specialised AI processors. The building can run at its intended scale only if the site can supply their electricity and remove the heat they produce.

The International Energy Agency examined announced data-centre projects against local grid conditions and connection timing. It estimated that grid constraints could delay around 20 per cent of the global capacity planned for construction by 2030.2

The global percentage should not be applied mechanically to a site. It shows that grid timing is large enough to change the build-out of the sector, while the answer for one project remains local to the network that must serve it.

A queue position is not delivered capacity

A connection request enters a process that tests what the existing network can carry and what must be built. The requested load may require new lines or a substation change, making the schedule for that reinforcement part of the data centre's schedule.

A place in the study process shows only that the operator is considering the request; it does not prove delivery by the preferred date. For investment, use the capacity the operator is prepared to make available under stated conditions.

This changes site selection because land that can take a building is not necessarily a site that can host the planned compute. A cheaper plot may be the more expensive choice if connection work postpones revenue or leaves the first phase below the scale assumed in the business case.

The tariff allocates failure

FERC's orders test the entry requirements for a large-load study and the allocation of network costs when expected demand does not appear. A separate part of the inquiry concerns operating requirements for connected loads.3

Those questions allocate risk before a server is installed. A financial commitment can protect existing customers if a project disappears, but requiring it early moves developer capital forward before the final compute design is settled.

Because the connection tariff decides which committed project can be studied, it forms part of the commercial model. Strategy has to account for that gate before delivery planning.

A contract can be the point of control

A developer may control the site yet still lack the right to draw the required electricity. The connection agreement identifies the capacity the site may take and its conditions, allowing a grid document to set the ceiling for the compute project.

A power-purchase contract can allocate the commercial value of electricity from a generator. It does not create transmission capacity at the data-centre site, where the physical route still has to work.

On-site generation changes the route rather than removing the energy problem. Its feasibility depends on the plant's own supply arrangement, and some designs still use the public grid.

FERC has not set the final tariff

The six proceedings may produce different reforms because the regional tariffs are not identical, while FERC's June orders only frame issues and possible responses. They do not establish that every current provision is unlawful or guarantee one connection timetable for all large loads.

FERC's orders cover regional operators within its jurisdiction, so a project outside organised markets may face a different legal route. Local planning rules remain separate from any federal transmission tariff.

For readers outside the United States, FERC is a current example of a regulator putting large-load connection rules at the centre of AI build-out rather than a universal connection code.

The decision moves upstream

A board considering an AI data-centre investment needs evidence of deliverable capacity under the governing connection process before it relies on the proposed compute scale. The decision record should give the supply date and identify any condition that could reduce usable capacity.

Source notes

Sources

Sources reviewed

  1. 01
    FERC Launches Aggressive Targeted Action to Speed Large Load Integration

    Federal Energy Regulatory Commission, 18 June 2026

    The release links the six show-cause orders and supports their Section 206 basis and scope.

    Checked 20 August 2026

  2. 02
    Connecting data centres to electricity grids

    International Energy Agency, 10 April 2025

    The IEA's location-specific analysis supports the estimate of planned data-centre capacity exposed to grid-connection delay.

    Checked 20 August 2026