Note: This article was automatically translated from Korean by AI. All content refers to South Korea unless otherwise stated. See Korean original for the most accurate reading.
Seven out of every ten applications to build a data center in South Korea target the greater Seoul area — and more than half of those get turned away by the power grid. It’s been said many times that AI’s real bottleneck isn’t GPUs but the wires that carry electricity to them, and in Korea that bottleneck sits squarely on the transmission lines between Seoul and the rest of the country. That’s why Korea’s government has gone as far as passing special legislation to push data centers toward the regions. Here’s a numbers-first look at how far a single grid bottleneck is redrawing an industrial map.
Why data centers keep piling into greater Seoul
For a data center operator, greater Seoul is a hard location to walk away from. AI services are sensitive to latency, and when corporate headquarters and cloud users are concentrated in greater Seoul, physical proximity shaves off delay. Maintenance staff, telecom networks, and financial-sector clients are already built around the capital region, so new entrants gravitate there out of sheer inertia. As hyperscale AI data centers began demanding more than 100MW of power apiece, that inertia started colliding head-on with the physical limits of the grid. Korea’s private data center market is projected to grow from KRW 6.22 trillion in 2024 to KRW 10.19 trillion by 2028 — but the problem is that this growth wants to happen not where power is abundant, but precisely where power is tightest. With greater Seoul already holding all three of a developer’s key conditions — sites, workforce, and clients — new operators have had little reason to look elsewhere from the outset.
736 applications, 71% in greater Seoul — the concentration in numbers
From August 1, 2024 through March 27, 2026, a total of 736 first-stage technical review applications for data centers were filed. Of these, 522 — 71% — were concentrated in the greater Seoul area. Of those 522 Seoul-area applications, 279, or 53.4%, were rejected as “supply not available.” Making the concentration even clearer: 91.2% of all nationwide rejections came from the greater Seoul area alone. The table below breaks it down.
| Category | Count / share | Scale |
|---|---|---|
| Total first-stage technical review applications | 736 | |
| Of which, greater Seoul applications | 522 (71%) | |
| Rejected among greater Seoul applications | 279 (53.4%) | |
| Greater Seoul’s share of all nationwide rejections | 91.2% |
Based on first-stage technical review applications, Aug 1, 2024 – Mar 27, 2026 (bar width reflects share)
Break the numbers down further and the concentration doesn’t stop at the application stage — it gets amplified again at the outcome stage. If greater Seoul accounts for 91.2% of nationwide rejections, simple arithmetic means only 8.8% of rejections come from everywhere else combined. Applications skewed 71% toward greater Seoul, yet rejections skewed even harder at 91.2% — meaning that even under the same rules, pass rates diverge sharply by region. In other words, the grid’s actual spare capacity already differs enormously from one region to another.
The bottleneck’s real name: the grid impact assessment
The rule behind these rejections is the power grid impact assessment. Written into the Special Act on the Promotion of Distributed Energy — enacted June 13, 2023 and effective June 14, 2024 — the provision requires any operator seeking to contract for 10MW or more of electricity to survey, forecast, and evaluate its impact on the grid before starting a project, then submit a report to Korea’s Ministry of Trade, Industry and Energy for approval. The intent is reasonable enough: if large new consumers keep flowing into an already-saturated Seoul-area transmission grid, blackout risk rises, and speculative land purchases in specific areas become a real concern that the rule is meant to address. The problem is that the system has been running as a two-year pilot without even a formal implementing notice. Government and industry disagree over the qualification requirements for the “assessment agents” who draft these reports on operators’ behalf, and two rounds of legislative pre-announcement have both failed to produce a finalized notice. That leaves companies deciding on investments worth tens of billions of won with no clear standard to go by — and that very uncertainty produces a paradox: it deepens the concentration in greater Seoul, because the less clear the criteria, the more operators choose to line up where demand is already proven rather than take a chance elsewhere.
The real approval rate — it narrows further at the review stage
Move past the first-stage technical review to the actual review stage, and the bottleneck narrows further. Of 24 greater-Seoul-area applications that reached the review stage, only 10 were approved — and in Seoul proper, just a single application was approved. Set the 71% skew at the application stage next to this narrow gate at the review stage, and it’s fair to say greater Seoul’s data center market is currently stuck in a jam of demand piling up faster than approvals clear. For operators, getting blocked at the grid impact assessment stage after already securing a site and completing design work means sunk costs stay sunk. That uncertainty has become a new line item of risk in investment decisions, and lenders reviewing project financing for data centers increasingly treat power grid approval as a precondition to check first.
The government’s workaround — cutting a shortcut for non-capital regions
The government’s chosen fix isn’t to loosen the rule, but to open a new bypass around it. The Special Act on the Promotion of the AI Data Center Industry passed South Korea’s National Assembly on May 7, 2026 and was promulgated on June 9, 2026 (Act No. 21759). Its core measure exempts AI data centers built outside the greater Seoul area from the grid impact assessment — or streamlines the process — while introducing a one-stop permitting system and a “deemed-approved” timeout rule under which an application is treated as approved if not rejected within a set deadline. Implementation begins nine months after promulgation, on March 10, 2027. In short: rather than lowering the bar in greater Seoul, the government designed an incentive that removes the bar entirely for anyone willing to build in the regions. With time still to run before implementation, the greater Seoul bottleneck could actually get more pronounced in the interim — and operators may start pre-emptively staking out regional candidate sites ahead of the effective date.
That staking-out has, in fact, already begun. Saemangeum has pursued a KRW 2 trillion data center build-out since 2020, with KRW 5.8 trillion of its overall KRW 9 trillion AI project budget earmarked for data centers — though progress has been slow due to delays in building out transmission lines. In Ulsan, SK Telecom and the local government have signed an AI data center (AIDC) memorandum of understanding for a project spanning 100MW to 1GW and roughly KRW 7 trillion. In Donghae, Gangwon Province, the GS Group is in talks with the government over a 1.2GW-class mega data center targeting completion by 2030. All three share a common thread: siting conditions like power and water are relatively favorable, but the time to breaking ground is anything but short. As Saemangeum shows, even with policy will behind a project, plans stall on the drawing board when physical infrastructure like transmission lines can’t keep pace. Which means that even if the special act lowers the permitting bar, regional relocation could remain a solution on paper if actual grid buildout doesn’t keep up.
The electricity pricing system is changing too — evenings just got more expensive
Around the same time, electricity pricing structure got its first overhaul in 49 years. From April 16, 2026, a seasonal and time-of-use rate structure took effect for industrial (Category B) electricity rates. The 11am–3pm weekday window moved from the top rate tier down to a mid-tier rate, while the 6pm–9pm evening window moved from mid-tier up to the top rate tier. In summer, the evening peak-load rate rises by KRW 28.5 per kWh while the daytime rate falls by KRW 16.9 per kWh. From June 1, the change was extended to general and educational-use rates as well. Because data centers run around the clock, they absorb the full impact of this time-of-day gap. The government’s stated intent is to lower rates during the day, when renewable generation is abundant, to encourage consumption, and raise rates in the evening, when demand peaks, to encourage load-shifting — but for an operation like a data center that can’t easily shift its load across time slots, this structurally functions as a cost increase. Even after clearing the grid-approval hurdle, operators now face a second variable reshaping their cost structure: the new rate system.
| Time window | Before | After | Summer change |
|---|---|---|---|
| Daytime, 11am–3pm | Top rate | Mid rate | −KRW 16.9/kWh |
| Evening, 6pm–9pm | Mid rate | Top rate | +KRW 28.5/kWh |
The counter-argument — regional dispersal isn’t necessarily the answer
Looking at power alone, the regions look like the answer, but other reporting flags the opposite problem: the regions have power, but no customers. As long as financial-sector and large-company headquarters remain concentrated in greater Seoul, latency-sensitive services find it hard to route around a regional data center, leaving regional facilities with power but few tenants — a different kind of triple bind, as some reporting frames it. Exempting operators from the grid impact assessment doesn’t, by itself, relocate demand to the regions. That reveals the special act’s limitation: it lowers the permitting bar but can’t force clients to physically relocate or force telecom infrastructure to be redistributed. Ultimately, whether power policy and industrial siting policy actually point in the same direction is likely to be the next gate that determines whether this special act delivers real results.
The longer horizon — a KRW 10 trillion market by 2028, and the reshuffling after
Take at face value the forecast that Korea’s private data center market will grow to KRW 10.19 trillion by 2028, and today’s bottleneck looks less like a temporary adjustment and more like a structural problem likely to recur for years. Real restructuring is likely to begin once the special act’s implementation (March 2027) overlaps with the full rollout of time-of-use pricing (extended even to residential rates). One plausible scenario: greater Seoul specializes in small, high-density facilities for latency-sensitive real-time services, while the regions take on large-scale facilities focused on training and batch processing. But that scenario only holds if the law’s implementation timeline and the pace of grid buildout line up as planned — in practice, delay risk needs to be watched closely, since grid expansion typically takes years. Add cooling and water to the mix: large-scale data centers use enormous volumes of water to dissipate server heat, and in a region like greater Seoul where industrial and residential water demand is already high, securing water supply — independent of grid headroom — could emerge as yet another permitting variable. The search for a site where power, water, and customer proximity all line up simultaneously looks set to continue for years.
Frequently asked questions
Q1. Does the grid impact assessment apply to every data center?
It applies to operators seeking to contract for 10MW or more of electricity use. Small and mid-sized facilities are not subject to it.
Q2. Once the special act takes effect, does regulation in greater Seoul loosen too?
No. The exemption applies only to data centers outside the greater Seoul area; greater Seoul continues under the existing grid impact assessment process unchanged.
Q3. Does the industrial electricity rate overhaul affect households too?
Currently, industrial (Category B) rates and EV charging power are the priority targets, with general and educational-use rates extended from June 1. Residential rates have not yet taken effect.
Q4. Why has this system run for two years without even a formal notice?
Government and industry have been unable to close the gap over qualification requirements for the agents who draft assessment reports, so two rounds of legislative pre-announcement have both failed to result in a finalized notice.
Q5. Does moving to the regions solve the data center problem?
The power problem gets solved, but proximity to clients and telecom infrastructure issues remain. Finding a site that satisfies both conditions at once remains the next challenge.
The core of this issue is that building a single data center no longer ends with the power grid alone — rate structures, permitting procedures, and industrial siting strategy are all bound together in one tangle. Anyone evaluating investment in related industries should first check how and when these structural variables intersect. This article is not intended as investment advice to buy or sell, and readers should make their own decisions carefully based on their individual circumstances.
Primary sources: Press release on the Special Act on the Promotion of Distributed Energy (Korea’s Policy Briefing, korea.kr), Briefing on the seasonal/time-of-use electricity rate overhaul (Korea’s Policy Briefing, korea.kr)