A picture shows a server room at the Olympic Broadcasting Services (OBS) offices in Pyeongchang on February 17, 2018, during the Pyeongchang 2018 Winter Olympic Games.
Florian CHOBLET/AFP via Getty Images
Seoul’s colocation data center market has effectively run out of room. The South Korean capital’s vacancy rate plunged from 6.9% in the second half of 2025 to just 1.1% in the first half of 2026, according to Cushman & Wakefield’s H1 2026 report, making Seoul one of the most capacity-constrained server markets on the planet — and forcing the next wave of AI infrastructure into satellite cities where the power grid still has room. The crisis has a fresh technological answer: on August 10, LS Electric and GS E&C signed a memorandum of understanding to jointly develop direct-current power distribution technology specifically for AI data center environments — a response to the power conversion losses that are quietly consuming a portion of every megawatt Seoul generates.
Seoul at 1.1%: What That Number Actually Means
A 5.8 percentage-point vacancy drop in a single half-year means that for every 100 megawatts of Seoul colocation space, fewer than two megawatts are now sitting empty. Industry practitioners treat sub-5% vacancy as a seller’s market and sub-2% as effectively depleted. At 1.1%, Seoul is not a tight market — it is a queue. A company seeking colocation capacity in Seoul today is almost certainly waiting in line for space that does not yet exist, committing to a lease 12 to 18 months before delivery.
Total operational data center capacity in the greater Seoul metropolitan area reached 663 MW in H1 2026, a 10% year-on-year increase, according to the Cushman & Wakefield H1 2026 Update. That growth has done nothing to relieve pressure. Demand from AI workloads, cloud service providers, and hyperscalers absorbed additions roughly as fast as they came online — the same structural dynamic that reduced US colocation vacancy to 1.2% nationally by Q1 2026, per an Avison Young market overview.
Choi Yong-jun, Managing Director at Cushman & Wakefield Korea, noted that the Seoul metropolitan region “remains the premier data center market in Korea, continuing to attract sustained interest from both domestic and international operators on the back of strong customer demand,” according to the Cushman & Wakefield H1 2026 report.
Why Seoul Cannot Simply Build Its Way Out
The vacancy figure is a symptom, not the root problem. The Greater Seoul Area operates with a power self-sufficiency rate of approximately 66% — it imports more than a third of its electricity from other regions, according to Cushman & Wakefield’s Korea power report. Facilities exceeding a threshold megawatt level must undergo a Power Grid Impact Assessment before receiving permits, a regulatory requirement that extends approval timelines to two to three years.
The result: a data center operator who secures a Seoul core site today may not receive power authorization until 2028 or 2029. That timeline gap — between the urgency of AI compute demand and the pace of grid permitting — is the structural engine driving development out of Seoul proper.
Andrew Green, head of Cushman & Wakefield’s Asia Pacific data center group, described the regional dynamic precisely: “AI and cloud investment across Asia Pacific has entered a phase of rapid, power-constrained execution. The challenge increasingly lies in securing the power infrastructure needed to support the next generation of AI workloads.”
Power Is Now the Site-Selection Criterion
As power access in Seoul’s core grows harder to secure, development interest has shifted to Incheon and Ansan — satellite cities where industrial power infrastructure already exists. The shift is visible in project announcements.
In June, Digital Edge acquired a site in Ansan’s Sihwa National Industrial Complex — a government-designated industrial zone located approximately 40 km (25 miles) southwest of central Seoul — to develop SEL5, a 60 MW hyperscale, AI-ready facility backed by a 90 MVA power agreement, among the largest secured in Ansan. The facility’s power architecture features a dual-feed design supported by two independent 154 kV substations, and is designed to achieve a Power Usage Effectiveness (PUE) below 1.25. To put that number in practical terms: a PUE of 1.25 means that for every 1 MW powering AI servers, 1.25 MW enters the facility — the remaining 0.25 MW covering cooling and distribution overhead. At that efficiency level, a 60 MW IT load requires just 75 MW of total facility draw. That is genuinely tight engineering in a market where legacy facilities routinely run above 1.5 PUE.
Digital Edge CEO John Freeman was explicit about why the Ansan selection was driven by power, not location preference: “With SEL5, we have secured large-scale power in a highly constrained market and paired it with a resilient, scalable site capable of supporting next-generation hyperscale deployments.” Digital Edge’s South Korea portfolio now spans Seoul, Incheon, Ansan, and Busan.
On the same day Digital Edge announced SEL5, DCI Data Centers and Koramco Asset Management held a groundbreaking ceremony for a 40 MW facility in Ansan’s Sihwa National Industrial Complex, with power provided by Korea Electric Power Corporation (KEPCO) and operations targeted for 2028, per Data Center Dynamics. The simultaneous arrival of two separate data center projects at the same Ansan industrial zone on the same June day is not coincidence — it reflects the systematic exhaustion of Seoul power options pushing developers toward a single alternative that still has grid capacity.
Beyond the immediate Seoul periphery, demand is beginning to emerge in provincial regions including South Gyeongsang (Gyeongsangnam-do), where sites can offer access to large power connections. LG Uplus chose Paju — 31 km (19 miles) north of Seoul — for its flagship 200 MW AI data center, the largest AI data center project in the Seoul metropolitan area. On July 29, LG Uplus’s board approved an additional 1.3 trillion won (approximately $924 million) in investment for Phase 2 of that facility, after Computer Center 1 sold out all occupancy contracts before its June 2027 scheduled completion.
DC Power: Squeezing More Compute From the Same Grid Connection
Geography is one response to Seoul’s power constraint. Technology is another — and both are now being pursued in parallel.
The LS Electric and GS E&C MOU signed Monday targets the power conversion losses that accumulate at every stage of a traditional data center power chain. In conventional AC-distribution architecture, grid power passes through a transformer, then a UPS, then a power distribution unit, and finally into each server’s internal AC-to-DC converter. Each stage loses between 2% and 5% of power. Total conventional conversion losses across the full chain can reach 10–15% of facility power draw, according to Microchip Technology’s power efficiency analysis.
DC power distribution eliminates intermediate stages by distributing power at high-voltage DC — typically 400V at the rack level under the Open Compute Project ORv3 standard — to servers that accept DC directly. The result is a 5–15% efficiency gain on the same grid connection. Applied across Seoul’s 663 MW of operational colocation capacity, even a conservative 10% efficiency improvement would recover the equivalent of approximately 66 MW — roughly equivalent to adding one new medium-sized data center without a new grid connection or a new building permit.
That is not a hypothetical. LS Electric, one of Korea’s largest electrical equipment manufacturers, and GS Engineering & Construction, one of its largest builders, have decided the opportunity is large enough to pursue as a formal joint development program. The MOU, signed at LS Yongsan Tower on August 10, commits both companies to coordinating their response to AI data center power architecture evolution.
Pritesh Swamy, head of research and consulting for the Cushman & Wakefield data center group in Asia Pacific, has also flagged the broader shift: modular construction and advanced cooling technologies are helping operators accelerate deployment and maximise available infrastructure in power-constrained environments.
The APAC Context: A Record Pipeline Still Falling Short
Seoul’s squeeze is an acute local expression of a regional dynamic. Cushman & Wakefield’s APAC H1 2026 Update found the Asia Pacific development pipeline at a record 26.5 GW — comprising 4.8 GW under construction and 21.7 GW in the planning stage. The pipeline expanded by 7.1 GW in just the first six months of 2026, more than double the 2,751 MW added in the second half of 2025.
Approximately 1.4 GW of new operational capacity came online across the region during the period. Despite that delivery, the APAC-wide colocation vacancy rate fell further — from 10.9% in H2 2025 to 10.3% in H1 2026. Supply is arriving at scale; demand is outrunning it.
Southeast Asia is leading the regional construction surge. Malaysia’s Johor leads individual markets with a development pipeline exceeding 3,000 MW, followed by Sydney at 2,134 MW, Bangkok at 2,084 MW, Mumbai at 1,726 MW, and Jakarta at 1,699 MW, per Cushman & Wakefield’s H1 2026 Update. Bangkok’s pipeline grew 99% year-on-year, with capacity under construction surging 148%.
US Opposition Is Redirecting Capital — Here
Seoul’s vacancy collapse arrives alongside a structural constraint on its principal competitor market. In the first quarter of 2026, bipartisan community opposition blocked or delayed at least 75 data center projects worth approximately $130 billion across the United States, according to a June 2026 report from Data Center Watch, a tracker maintained by AI research firm 10a Labs — a single quarter that roughly matched the total for all of 2025. Active opposition groups more than doubled from 396 at end of 2025 to 833 across 49 states by March 2026.
That resistance constrains where new supply can be built in the world’s largest data center market, potentially rerouting hyperscaler capital toward Asia Pacific markets — including South Korea — that remain comparatively open to new development.
Globally, Cushman & Wakefield’s 2026 Global Data Center Market Comparison documented data center capacity under construction at record levels, with the Americas accounting for the majority of activity.
What the Next Phase Looks Like
Cushman & Wakefield projects that Australia, India, Japan, and Malaysia will each surpass 2 GW of operational data center capacity by 2028 as APAC’s center of gravity continues to broaden, per the H1 2026 Update. Andrew Green argued the next growth phase will be led by markets that successfully balance scale, power access, and long-term infrastructure readiness, per Real Estate Asia’s APAC report.
For South Korea specifically, the near-term outlook is a race between new supply arriving in satellite markets and demand that shows no sign of softening. At 1.1% vacancy, any company seeking Seoul colocation today is almost certainly in a queue — not a negotiation.
The two responses now underway — geographic migration to Ansan and Incheon, and technological mitigation through DC power distribution — are not mutually exclusive. South Korea is betting both approaches are necessary. The first rewrites the map of where AI infrastructure gets built. The second tries to get more capacity out of the infrastructure already in the ground.
Frequently Asked QuestionsWhy is Seoul’s data center vacancy rate so low, and what does 1.1% mean for companies seeking space?
Seoul’s colocation vacancy fell to 1.1% in H1 2026 because AI workload demand from hyperscalers and cloud providers absorbed new supply roughly as fast as it came online, according to Cushman & Wakefield’s APAC H1 2026 Update. For companies seeking capacity, 1.1% is effectively zero — industry practitioners treat anything below 5% as a seller’s market and below 2% as critically tight. At 1.1%, tenants cannot negotiate; they compete, and they wait. New colocation commitments typically require pre-leasing 12 to 18 months before delivery of a facility still under construction.
Why are data centers moving from Seoul to Ansan and Incheon rather than staying in the city?
Power, not land or location preference. The Greater Seoul Area operates at approximately 66% power self-sufficiency, importing more than a third of its electricity from other regions. Facilities above a threshold megawatt level must pass a Power Grid Impact Assessment before receiving permits — a process that currently extends timelines by two to three years. Ansan’s Sihwa National Industrial Complex was built with heavy manufacturing power infrastructure that predates the data center boom, which means it already has 154 kV substation access and independent grid feeds at a scale Seoul’s denser urban core cannot match. Two separate data center projects — Digital Edge’s SEL5 and DCI/Koramco’s SEL02 — both chose Sihwa National Industrial Complex on the same day in June 2026.
What is DC power distribution, and why does it matter for Seoul’s AI data center shortage?
Traditional data centers distribute power as alternating current (AC) from the grid to each building, then convert it to direct current (DC) inside individual servers. Each conversion step loses 2–5% of power. DC power distribution — sending high-voltage DC directly to racks at 400V under the Open Compute Project ORv3 standard — eliminates intermediate conversion stages, recovering 5–15% of total power draw. Applied across Seoul’s 663 MW of operational colocation capacity, even a modest 10% efficiency gain through DC distribution could recover approximately 66 MW of effective capacity without a new building or a new grid permit. LS Electric and GS E&C signed an MOU on August 10 to jointly develop this technology specifically for AI data center environments — the industry’s direct technological response to a grid that cannot expand fast enough to meet demand.
How does the US data center backlash affect South Korea’s infrastructure market?
When community opposition blocks or delays projects in the world’s largest data center market, capital has to go somewhere. In Q1 2026, at least 75 US projects worth approximately $130 billion were stalled or blocked by bipartisan opposition focused on electricity costs, water use, and noise, according to Data Center Watch. That constraint effectively narrows the field of viable hyperscale locations, increasing competitive pressure on markets that remain open to new development. South Korea — with its national AI infrastructure mandate, government-designated strategic facility status for data centers, and established hyperscale real estate market — is a direct beneficiary of that capital rerouting, even as its own Seoul market approaches saturation.