Data center operators face a complex conundrum in the US and elsewhere. In many regions, there are years-long delays when connecting to power; meanwhile grid transmission system operators (TSOs) fear they could destabilize infrastructure, and consumers blame them for rising electricity prices.

The US grid connection authority, the North American Electric Reliability Corporation (NERC), has responded, issuing an urgent alert to TSOs outlining a roadmap for managing large computational load connections based on detailed analysis of data center power profiles and demand. The definition of a large load is unclear, but NERC describes a computational load as exceeding 20 MW, connected at 60 kV and containing at least 1 MW of IT.

This should not come as a surprise: TSOs already require similar information from other major industrial loads, such as steel mills, as well as from generation sources, including natural gas turbines and solar farms. Data center operators were previously treated as passive consumers of electricity, but they now operate at a scale that requires them to become active participants in maintaining grid stability. Operators will need to navigate a connection process that assures TSOs they will not cause trouble for other service users.

Under such a process, data center connection approvals may remain slow but become more predictable. Even so, it will take time to establish a connection process capable of accommodating the complexity and evolving nature of data center power profiles, as well as the differing situations across TSO regions.

The NERC alert

In at least two documented cases, grid operators narrowly averted regional blackouts when data centers simultaneously switched to on-site power in response to grid fluctuations. In Northern Virginia, 1,500 MW of data center capacity across 60 facilities tripped in July 2024; in Ireland, 200 MW of capacity briefly switched to batteries and then back again in December 2022.

TSOs in multiple regions are now introducing new grid codes that require facilities to be capable of ride-through fluctuations (see Power companies act to stop data center-induced blackouts and EU power providers urged to protect grids from data centers).

Concerns have also emerged about the impact of the fluctuating load patterns associated with AI training facilities operating large GPU clusters.

NERC published a white paper on the characteristics and risks of large loads in July 2025 and launched a Large Loads Action Plan in April 2026 to develop initial reliability standards by the end of 2026. Previous NERC standards have subsequently been embedded in regulations issued by the US Federal Energy Regulatory Commission.

Given the urgency of the issue, NERC published a very rare Level 3 alert in May 2026. This was only the third such alert in the organization's 58-year history — the previous two were issued to address Winter Storm Elliott in December 2022 and a spate of reliability problems arising from renewable energy resources in May 2025.

The alert lists seven "essential actions." In theory, these actions are voluntary, but NERC is the membership organization of North American TSOs and its guidance carries significant weight. Collectively, these essential actions amount to a strong recommendation that TSOs should demand data from operators to build models of data center behavior, and then use those models to develop new procedures for when loads are commissioned or operators make major operational changes. TSOs are also encouraged to monitor data center behavior during grid faults and establish joint operational procedures with facility operations teams.

TSOs have until August 2026 to respond, either by explaining when they will implement the essential actions or, if they do not intend to implement them, why not. Table 1 lists the actions that TSOs are expected to carry out in developing these new procedures, together with the likely implications for data center operators.

Table 1 The seven essential actions for TSOs and their impact on data centers

Interim measures

Uptime Intelligence anticipates that NERC and its members will take two years or more to collect the necessary data and develop workable models. This work will run in parallel with the Large Loads Action Plan standards development, which is scheduled to publish the foundational standard for managing large data center loads on the grid by the end of 2026.

In the meantime, NERC has issued some interim advice to TSOs in the form of a preliminary non-binding draft intended to enable utilities, grid operators, equipment manufacturers and data center companies avoid problems and bridge the gap until the planned NERC standards are introduced.

The guidelines closely mirror the form of the program outlined in the essential actions with a few specific additions. They propose that TSO should develop new connection codes, including ride-through requirements and measures to mitigate the impact of the cyclical load profiles associated with AI training. They also instruct large-load operators to implement cybersecurity measures.

The need for co-operation

Data center operators already face increasing public scrutiny from planning authorities and from government environmental programs in the EU, California and elsewhere. NERC's program includes new data demands, backed by an overriding concern: grid stability.

To fulfil NERC's goals, TSOs must gather enough data from a representative sample of data centers to build reliable models of data center behavior, and then apply those models to all applications for new or modified data center connections.

Operators may be reluctant or unable to provide all the data required during the modeling stage, the connection phase, or later stages when operational changes are made. NERC will also face challenges in modeling emerging loads such as AI inferencing, whose characteristics are only partially understood.

Data center operators applying for grid connections will face a more complex commissioning process, with TSOs requiring more live testing. This will increase the time that should be allocated to commissioning, although it may not necessarily add to connection delays.

Once data centers are running, operators should expect more intrusive monitoring of their electrical activity, and may also need to conform to new joint operational procedures with the TSO during potential faults.

The new rules that emerge from this process should become more stable and predictable. While the details are up to the individual TSO, NERC's guidance could help avoid some of the pitfalls that have arisen in rules published so far. For instance, in Ireland, EirGrid's proposed codes include an impractical requirement that data centers be back online within 500 ms of grid stabilization; NERC proposes a more collaborative restart process.

Nevertheless, both the alert recommendations and the planned initial Large Loads Action Plan standard will be subject to further change. Data centers have much more varied profiles than traditional large loads or generation sources such as solar farms and gas power plants. Those profiles are also likely to change in the future as new AI training models arrive, new inference methods are adopted, and new types of data center loads emerge.

The following Uptime experts were consulted for this report:
Jay Dietrich, Research Director, Energy and Sustainability, Uptime Institute

Other related reports published by Uptime Institute include: 
Power companies act to stop data center-induced blackouts 
EU power providers urged to protect grids from data centers
Ireland's new grid rules signal shift in data center roles
EU power providers urged to protect grids from data centers