ESG & Sustainability

Korea's Power Grid Impact Assessment, Step by Step: What Investors Should Ask at Each Stage

First in our data center series. Our opening piece made the case that on a Korean data center, the grid filing is the diligence — most capital-region applicants do not get through it. This article opens the mechanism: what triggers the assessment, how it is scored, where in the funnel projects actually stall, and the one question a buyer should be asking at each stage.

What triggers it, and what it replaced

The power grid impact assessment (전력계통영향평가) was introduced by Korea's distributed-energy legislation, which took effect in June 2024, to stop large electricity loads from piling onto parts of the grid that are already saturated — the capital region above all. The trigger is consumption, not sector: a party intending to use 10MW or more of electricity in Korea must go through it, which means data centers of any commercial scale, but also large factories and commercial complexes.

The system replaced the advance notice of intended power receipt (전력수전예정통지) that developers previously filed with the utility. That change is the whole point for diligence. The old notice was a queue position; the new assessment is a decision. A target that "applied to the utility" under the old regime and a target that "filed an assessment" under the new one hold different things, and neither holds confirmed power until the final stage described below.

The scorecard: technical items plus a large non-technical weight

As the implementing notice was published for consultation in August 2024, the assessment is scored out of 100, with 70 points needed to pass. The technical items are what an engineer would expect: available supply headroom on the relevant grid segment (reported at 25 points), the difficulty of securing that headroom (20 points), whether appropriate voltage can be maintained, the applicant's plan to minimize its grid impact, whether it applied at an appropriate voltage level, and whether it contributes land that helps the utility build supply capacity. The grid analysis data underlying the headroom items is made available to the applicant or its appointed preparer rather than published.

The non-technical items are the part investors tend not to expect. Roughly 40% of the total is allocated to factors such as local community acceptance, the business's stability, its contribution to local public finances, its industrial-activation effect, the underdevelopment of the region, the region's power self-sufficiency, and whether the project falls under regional or special-act support programs. Read as a diligence document, that structure says something specific: a capital-region site starts with a structural deficit on the regional-development and self-sufficiency items that no engineering submission can recover, while a well-sited non-capital-region project can score on those items before its technical case is even considered. That is the mechanism behind the geographic split our opening article described — and behind the exemption logic in the AI data center legislation we take up next in this series.

The funnel: four stages, and where projects actually stop

The published capital-region figures, cumulative to March 2026, describe the funnel better than any procedural summary:

Two features of that table matter more than the headline rate. First, the gap between 243 first-stage passes and 24 full-review entries: a target described as having "passed the technical review" belongs, statistically, to a group of which only about one in ten had progressed further. Second, the approved capacity of 1,010MW against 33,592MW requested — approval is not only rare but typically smaller than the filing, which is the corrective-measures mechanism at work.

Timeline and sequencing

The filing is prepared through an accredited assessment preparer and submitted to the Ministry of Trade, Industry and Energy roughly three months before the project's permit application, and the ministry is to respond within about three months of receipt with any corrective measures it requires. Consent of the local government is also required for projects that trigger the assessment. In practice, then, the grid review sits in front of the building permit, the environmental permits, and the construction schedule — a data center's entire permitting sequence is downstream of a decision that, on the published numbers, usually goes the other way.

"Corrective measures" is the term that should get a reader's attention. The categories the scorecard rewards — impact-minimization plans, land contribution for supply capacity, appropriate voltage — are also the levers a corrective notice can pull: a reduced or phased capacity, a requirement to provide substation land, a different connection voltage with different equipment and cost. An approval with corrective measures attached is a different project from the one in the investment memorandum, and the model should be rebuilt around the approved facility, not the filed one.

The question to ask at each stage

1. Before filing. Has a preparer been appointed, and has the grid analysis data for the relevant segment been obtained? A target that has neither has not started, whatever its site plan says.
2. Filed. What capacity was filed, at what voltage, and on what date? Which grid segment and substation? Are there other filings on the same segment that would compete for the same headroom?
3. First-stage technical review. Passed, or found unable to be supplied? If passed — on what date, and has the project moved toward full review since, or is it parked? Parked projects are the majority of the surviving pool.
4. Full review. What is the project's own estimate of its score on the non-technical items, and does it depend on any support-program designation that is not yet in hand? Have corrective measures been signalled?
5. Supply approval. What capacity was approved against what was requested, with what conditions, on what date — and does the approved capacity support the revenue case?
6. At every stage. As of what date is this status, and who verified it? A status carried forward from a teaser document is not a finding.

The provisional-rule problem

The scorecard and process above describe the notice as published for consultation and as subsequently reported. Press coverage in 2026 indicates the assessment has been administered for around two years without the implementing notice being finalized — operating, in effect, on a pilot basis. For diligence that has a plain consequence: item weights, procedural timing, and the treatment of competing applicants on the same segment can shift with the next revision, and a finding obtained under one version of the rule does not necessarily survive the next. Grid status belongs in the report with an explicit as-of date, a source, and a re-verification step before signing — and where the deal timeline runs across the March 2027 effective date of the AI data center legislation, across that date too.

Last reviewed: 9 September 2026 · Dime Works