The FERC Data Centers order issued on June 18, 2026, was not an energy-efficiency rule in the narrow sense. It did not prescribe server utilization targets, cooling designs, or power usage effectiveness thresholds. Its impact is likely to come through grid interconnection, transmission cost assignment, co-location rules, and the treatment of flexible large loads that include AI data centers.
That distinction matters. AI operators often describe efficiency in terms of chips, cooling, model architecture, and facility design. FERC is addressing a different layer: whether grid tariffs and interconnection processes can handle much larger, faster-moving electricity requests without shifting costs onto other customers or weakening reliability.
What The FERC Data Centers Order Changed
Why FERC Data Centers Rules Matter
On June 18, 2026, FERC issued show-cause orders under Section 206 of the Federal Power Act to all six Regional Transmission Organizations and Independent System Operators: PJM, MISO, SPP, CAISO, ISO-NE, and NYISO. Data Center Dynamics reported that the grid operators were directed to justify their current rules for large-load interconnection and tariffs or propose reforms under the FERC orders.
The order set two short response windows. Grid operators had 30 days from June 18, 2026, to submit reports on how they would ensure enough generation to serve existing and new large-load demand. They also had 60 days to justify their tariffs or submit changes. By September 3, 2026, both windows had elapsed; the research record provided here does not establish the content or adequacy of each operator’s response.
The FERC Data Centers proceeding sits at the intersection of AI infrastructure growth and power system planning. The order asked grid operators to address five categories of reform:
- Transmission service applications and study processes, including the use of alternative technologies.
- Cost assignment transparency and protections against shifting data center-related upgrade costs to other customers.
- Rules for co-location agreements and behind-the-meter generation.
- Transmission services for flexible large loads.
- Study processes for generation serving electrically proximate or co-located large loads.
The Order Is About Grid Process, Not Facility Design
The order does not make a less efficient AI cluster illegal, and it does not certify a more efficient one. Instead, it changes the questions that data center projects may face before they receive service. A proposed facility may need clearer evidence on when it will take power, how firm that load will be, whether it can reduce demand under defined conditions, and who pays for network upgrades if the project changes or fails to appear.
That can still influence efficiency. If interconnection rules attach more cost to a project’s peak demand or require more credible operating assumptions, developers have a reason to reduce stranded capacity, improve power scheduling, and avoid over-ordering grid service that is not used. These are indirect effects rather than direct engineering mandates.
Why AI Loads Put Pressure On Planning Models
Project Size Has Increased
FERC’s March 2026 State of the Markets report said U.S. data centers doubled electricity usage between 2018 and 2023 and accounted for about 4.4% of total U.S. electricity consumption in 2023. The same report said more than 50 GW of data center capacity was in service nationwide by the end of 2025, with a 24% compound annual growth rate for in-service capacity since 2020. Average project size entering service rose from about 25 MW in 2020 to nearly 80 MW in 2025, according to the FERC market report.
Those figures explain why interconnection rules have become a technical issue for AI infrastructure. A single 80 MW project can resemble an industrial load. Multiple projects in the same transmission zone can change local congestion, reserve planning, and upgrade timing. For related context on U.S. data center demand estimates, Abacus News has also examined DOE and Berkeley Lab ranges.
Regional growth also matters. The research notes identify MISO as the fastest-growing region for in-service data center capacity, with about 43% compound annual growth since 2020, followed by ERCOT, SPP, and the Southeast at roughly 28% to 30%. ERCOT is not one of the six grid operators covered by the June 18 FERC order, but its growth helps explain why the data center power question extends beyond one market design.
AI Demand Is Not A Single Load Type
AI demand is often discussed as if all compute behaves the same way. In grid planning, that is too broad. Some facilities may have workloads that can be shifted in time or throttled under contract. Others may serve latency-sensitive applications, enterprise commitments, or colocated customers that expect firmer service. FERC’s request for new transmission services for flexible large loads recognizes that distinction without assuming every data center can behave like a dispatchable resource.
This is where energy efficiency and grid efficiency separate. A facility can use efficient chips and cooling while still creating a difficult peak demand profile. A less optimized facility might offer interruptibility that reduces grid stress during constrained hours. The tariff question is how to value those operating differences in a way that is measurable and enforceable.
For those interested in exploring how power-system policies intersect with infrastructure technologies, Camp TechWise offers related insights and discussions that might be beneficial.
Efficiency Effects Will Be Indirect
Cost Assignment Can Change Design Incentives
The most immediate efficiency effect may come from cost assignment. FERC’s order sought to prevent transmission upgrade costs linked to data centers from being shifted to residential customers or other industrial users when projects do not materialize or do not pay their share. If a large-load customer faces clearer responsibility for upgrades, it has a stronger reason to request a realistic capacity amount, phase construction, and coordinate with generation supply.
That does not guarantee lower electricity use. A cost-reflective tariff can make some projects more disciplined, but it can also make power procurement more expensive or slow deployment in constrained zones. The likely result is not a single national efficiency outcome. It is a set of regional tariff changes that may reward better load forecasting and penalize speculative queue positions.
Co-Location Rules Could Reduce Or Increase Grid Stress
Co-location with generation is another technical pressure point. In February 2025, FERC launched a show-cause proceeding focused on large loads, including AI-enabled data centers, co-located with generating facilities in PJM. In December 2025, FERC directed PJM to create transparent tariff rules for AI-driven data centers and other large loads co-located with generation.
Co-location can be attractive because a generator and data center can be planned together. Yet it raises questions about backup service, transmission use, reliability obligations, and whether the facility draws from the grid when the colocated generator is unavailable. FERC’s June 2026 order widened that issue beyond PJM by asking all six covered grid operators to address co-location and behind-the-meter generation in their tariffs.
For AI operators, the technical lesson is that power architecture is no longer separate from model deployment strategy. A cluster sized for peak training runs, inference growth, or customer commitments may face different grid costs depending on how flexible, verifiable, and local its power supply is.
Operational Limits And Risks For Grid Operators
Study Processes Need Better Load Evidence
Interconnection studies were not originally built around waves of very large loads arriving with short commercial timelines. The research notes describe FERC’s interest in efficient application and study processes, including alternative technologies. Faster studies can reduce uncertainty, but only if the inputs are credible.
Grid operators will need enough evidence to distinguish firm projects from speculative requests. They also need tariff language that assigns costs when projects are delayed, resized, or canceled. Without that discipline, queues can fill with capacity reservations that block other users and distort transmission planning.
Flexible Load Programs Need Verification
Flexible service for large loads is attractive in theory. It could allow some data centers to connect under terms that reduce consumption during grid stress or accept a lower service level in exchange for different charges. The challenge is verification. A tariff has to define what flexibility means, how performance is measured, and what happens if a customer does not curtail as agreed.
This is a technical and governance problem, not only a pricing issue. If curtailment obligations are vague, grid operators may overestimate available flexibility. If obligations are too strict, data center operators may avoid the product and seek firm service instead. The order did not settle that balance; it required grid operators to justify or reform their rules.
FERC Data Centers Order And AI Efficiency
The FERC Data Centers order should be read as a grid integration measure with efficiency consequences, not as a direct AI efficiency standard. It may push data center developers toward more accurate load requests, clearer power sourcing, phased interconnection plans, and better alignment between facility design and grid constraints.
Its effect will depend on the tariff reforms accepted for each region and on how strictly grid operators apply cost assignment, co-location, and flexible-load rules. The evidence supports a cautious interpretation: FERC has recognized that AI-related data center growth is large enough to affect transmission planning, but the order does not by itself prove that AI facilities will use less electricity per unit of compute. It changes the rules around how that electricity is requested, studied, priced, and integrated into the grid.
