Seoul Data Center Vacancy Hits 1.1%, but Power Constraints Threaten the Next Wave

Seoul data center vacancy fell to 1.1% in the first half of 2026, leaving the market almost full despite a 10% capacity increase. Only Johor, Malaysia, recorded a lower rate among the primary Asian markets tracked by Cushman & Wakefield. The comparison signals strong demand, but it also exposes Seoul’s central problem. Customers want more computing capacity near the Korean capital, while power and suitable development sites remain difficult to secure.

Greater Seoul’s operational capacity reached 663 megawatts during the period. Colocation vacancy, meaning completed capacity available for lease to multiple customers, dropped from 6.9% at the end of 2025. Those figures suggest that cloud companies, enterprises, and AI operators absorbed new supply faster than developers delivered it.

Johor provides the critical contrast. It combined a lower 0.7% vacancy rate with 1,110 megawatts already operating and a development pipeline exceeding 3,000 megawatts. Seoul is therefore tight because demand is strong and new supply faces constraints. Johor is tight while developers are building at a much larger scale.

That difference matters more than the ranking. Low vacancy rewards existing Seoul operators, but it can restrict customer choice and raise the value of projects with secured electricity. It also places pressure on South Korea to reconcile its AI ambitions with a power system already concentrated around the capital region.

Seoul Data Center Vacancy Fell as New Capacity Arrived

The most important change is not Seoul’s ranking. It is the speed at which customers absorbed newly available capacity.

The latest APAC market update from Cushman & Wakefield covers the first half of 2026. It places Greater Seoul’s operational capacity at 663 megawatts, up 10% from the previous half.

Vacancy simultaneously tightened from 6.9% to 1.1%. A falling vacancy rate during an expansion means leasing demand exceeded the amount of newly completed, uncommitted space.

That result follows continued growth during 2025. Greater Seoul had 533 megawatts operating in the first half of that year, according to Cushman & Wakefield. Capacity reached 601 megawatts by year-end, with vacancy at 6.9%.

The 2026 figures indicate another 62 megawatts entered operation within six months. Yet the proportion available to customers fell by 5.8 percentage points.

A simple capacity estimate illustrates the change without replacing the consultancy’s project-level data. Applying the reported vacancy rates suggests roughly 41 megawatts were vacant at year-end 2025. The same calculation leaves about seven megawatts vacant in the larger 2026 market.

These estimates are directional because available inventory can differ by building, power density, commissioning stage, and contract status. However, the scale of the movement remains clear. Seoul added capacity while its immediately leasable inventory contracted.

The Seoul data center market is also larger than the city’s administrative boundaries. Greater Seoul includes important clusters in Incheon and Gyeonggi Province, where operators can serve capital-region customers while pursuing larger sites.

That wider geography has become essential. Seoul itself has expensive land, limited suitable parcels, and competition for grid capacity. Developers increasingly distribute facilities across the metropolitan area instead of relying on one dominant cluster.

This pattern already appeared in Cushman & Wakefield’s first-half 2025 assessment. Areas beyond the established clusters accounted for 32% of operational capacity and 15% of the development pipeline at that time.

Decentralization within Greater Seoul can create more development options. It does not remove the need for transmission capacity, grid approvals, local permits, and community acceptance.

The current vacancy figure also needs careful interpretation. It measures colocation availability, not every server room or privately owned computing facility in South Korea. Hyperscalers often build or contract dedicated capacity that never enters the general leasing market.

A 1.1% colocation vacancy rate therefore does not mean 98.9% of every data center in Seoul is occupied. It means available capacity within the measured commercial market has become exceptionally scarce.

That distinction matters for enterprise buyers. A company may find space in one facility but still struggle to obtain the required power density, network connections, expansion rights, or delivery schedule.

The headline number nevertheless captures a real shift. Greater Seoul moved from a tight market to one where suitable capacity can become unavailable long before a facility formally opens.

AI and Cloud Demand Are Pressuring Seoul’s Power Advantage

Seoul’s scarcity reflects a collision between concentrated digital demand and a grid that policymakers want to decentralize.

Cushman & Wakefield attributes rising site-selection and leasing activity primarily to the spread of AI and cloud services. Both workloads benefit from large, reliable power connections, but they do not create identical location requirements.

Cloud regions often need proximity to business customers, telecom networks, and other facilities. These connections support low latency and allow data to move between availability zones.

AI training can be more flexible about location, especially when a workload runs for hours or days without constant user interaction. However, large GPU clusters demand substantial electricity and specialized cooling.

AI inference, which runs a trained model when users submit requests, creates another location tradeoff. Some services benefit from being near major population and enterprise centers. Others can operate farther away when small delays are acceptable.

Greater Seoul concentrates corporate headquarters, financial institutions, technology companies, public agencies, and a large share of South Korea’s population. That concentration makes the region commercially attractive even when construction becomes difficult.

The same concentration creates grid pressure. Korea Electric Power Corporation says electricity demand remains concentrated in areas such as the Seoul metropolitan region. Its decentralization strategy includes regional tariffs, distributed energy, demand response, and alternatives to conventional network expansion.

Data centers intensify the challenge because they require dependable power throughout the day. Operators also need backup systems and enough network capacity to support equipment that customers expect to run continuously.

Connecting one facility involves more than producing enough electricity nationally. Power must reach the specific site through substations and transmission infrastructure with sufficient remaining capacity.

A proposed development can therefore face a long connection delay even when South Korea’s total generation appears adequate. Local bottlenecks, network stability, and competing connection requests shape the actual schedule.

Land adds another constraint. Large campuses need sizeable parcels, access roads, utility corridors, and adequate separation from sensitive neighboring uses. Those requirements are harder to satisfy in a dense metropolitan area.

High land costs also affect facility design. Developers may build vertically or pursue smaller sites, while large AI deployments often favor campuses that support phased expansion.

These conditions help explain why a low Seoul vacancy rate is not an automatic signal for unlimited construction. Demand can justify more capacity without making every proposed site practical.

The government has been trying to redirect large electricity users toward regions with greater grid availability. Earlier measures allowed electricity supply to be delayed or refused when requests above five megawatts threatened network reliability.

South Korea later introduced grid impact assessments under its distributed energy framework. These reviews consider how a large project affects the power network and surrounding region.

The system has attracted criticism from developers who view approvals as another obstacle to capital-region projects. The government disputes claims that the process effectively prohibits new Seoul-area data centers.

In an August 2025 clarification, the Ministry of Trade, Industry and Energy said 27 projects had passed assessments by that point. Capital-region developments represented 33% of all approvals and 42% of approved data center cases.

Those figures show that approvals remained possible. They do not guarantee that every approved project can secure land, permits, financing, equipment, and a timely connection.

The International Energy Agency identified transmission development delays and regional imbalances as significant Korean energy challenges. Its Korea energy review notes that Seoul has the country’s largest concentration of electricity consumption and transmission capacity.

This creates the defining pressure on the Seoul data center market. Customers prefer the capital region for commercial and network reasons. Energy policy increasingly rewards projects that move closer to power generation elsewhere.

Seoul Data Center Vacancy Looks Different Beside Johor

Johor shows that low vacancy can accompany enormous construction, while Seoul’s low vacancy reflects a far more constrained supply path.

Johor’s colocation vacancy reached 0.7% in the first half of 2026. That placed it slightly below Greater Seoul’s 1.1%, but the difference between the markets extends far beyond six-tenths of a percentage point.

Johor had 1,110 megawatts operating, 602 megawatts under construction, and 2,486 megawatts planned. Its combined operational and future capacity reached 4,198 megawatts.

Greater Seoul’s 663 megawatts of operational capacity was substantial, but it sat within a more mature and geographically constrained urban economy. Johor offered developers larger sites and access to demand spilling across the Singapore border.

Johor initially gained attention as an alternative to land-constrained Singapore. It has since become a regional destination for hyperscale cloud and AI campuses in its own right.

Its scale expanded quickly. Operational capacity increased by more than 212 megawatts during the first half of 2026, according to the Cushman & Wakefield figures.

Vacancy still tightened to 0.7%. That suggests customers absorbed new facilities almost immediately or committed to them before completion.

Preleasing is central to understanding modern data center vacancy. A building can join operational inventory with most or all capacity already contracted, leaving little space visible in market statistics.

Johor’s development model supports large campuses with multiple phases. Operators can reserve land, construct an initial building, and add capacity as power connections and customer commitments mature.

Seoul offers a different value proposition. Its appeal rests on proximity to Korean enterprises, dense connectivity, established operators, and workloads that benefit from a capital-region location.

The two markets are therefore not substitutes for every customer. Moving a Korean financial workload to Malaysia involves latency, data-governance, resilience, and operational considerations beyond real estate costs.

However, regional cloud companies allocate infrastructure across portfolios. A hyperscaler can place latency-sensitive services near Seoul while directing some training or regional processing toward larger campuses elsewhere.

This makes Johor a strategic benchmark rather than a direct replacement. It demonstrates what becomes possible when available land, power planning, and large customer commitments align.

The comparison also exposes a risk for Seoul. A market can remain attractive while losing incremental investment that does not require a Korean location.

Developers will naturally compare connection timelines and construction certainty. If one market offers an approved power path for a multi-building campus, that certainty can outweigh proximity to Seoul.

Johor faces its own limits. Rapid development increases electricity and water demand, while a large planned pipeline can create future oversupply if customer commitments weaken.

Planned capacity is not completed capacity. Projects can be delayed, resized, or canceled because of financing, grid infrastructure, equipment availability, or changing demand.

Its 0.7% vacancy rate also describes current commercial availability, not the eventual balance after 2,486 megawatts of planned projects enter the market.

Seoul’s smaller pipeline can protect occupancy if demand remains strong. It can also leave customers with too few options and push deployments toward other Korean regions.

The broader Asia-Pacific market sits between these models. Cushman & Wakefield reported 4.8 gigawatts under construction and 21.7 gigawatts planned across the region in mid-2026.

Regional vacancy declined from 10.9% to 10.3% despite new deliveries. Demand was still absorbing capacity overall, but conditions varied widely among cities.

JLL expects Asia-Pacific vacancy to remain near 6.5% to 7% over the next several years. Its regional outlook says 78% of the 4.8 gigawatts expected by 2027 was already preleased.

That forecast reinforces the significance of Seoul’s 1.1%. Greater Seoul is considerably tighter than the expected regional range, even before every AI-related demand projection becomes reality.

What the 1.1% Seoul Vacancy Rate Does Not Prove

A near-zero vacancy rate confirms scarcity, but it does not prove that every proposed facility is viable or every AI forecast will materialize.

Vacancy is a snapshot shaped by the report’s market boundary, inventory definitions, and observation date. It cannot reveal every contract’s duration or each tenant’s future utilization.

Customers sometimes reserve capacity before their computing demand fully arrives. They may value certainty because suitable data center space takes years to develop.

That behavior can reduce reported vacancy without producing immediate, full equipment utilization. Contracted electrical capacity and actual electricity consumption are related, but they are not identical.

AI forecasts add further uncertainty. Training and inference demand has grown, yet hardware efficiency, model architecture, and workload scheduling continue to change.

A more efficient accelerator can complete the same task with less energy. Conversely, lower computing costs can encourage customers to run more workloads and offset those savings.

Developers must make long-term investment decisions while those variables remain unsettled. A data center built today may operate for decades, while server designs change within a few years.

Power density creates another measurement problem. One available megawatt in an older facility may not meet the cooling or rack-density requirements of a modern AI cluster.

Headline vacancy can therefore understate scarcity for AI-ready capacity. It can also overstate scarcity for customers with conventional enterprise workloads and flexible location requirements.

Project pipelines deserve similar caution. Announced megawatts are not interchangeable with operating megawatts. A plan needs land control, power, permits, financing, construction capacity, and customer demand.

Greater Seoul also faces community concerns. Residents have challenged some proposed facilities over noise, cooling systems, power infrastructure, and perceived environmental effects.

Not every objection has the same technical basis. Still, community resistance can delay permits, increase design requirements, or stop projects regardless of regional demand.

Transmission projects face their own opposition. A data center can secure a site while the substation or power line serving it remains delayed.

South Korea’s policy response acknowledges this conflict. The country wants additional AI infrastructure, but it also wants to distribute electricity demand beyond the capital region.

The AI Data Center Industry Promotion Act makes that direction more explicit. South Korea’s Ministry of Science and ICT says the law passed the National Assembly in May 2026.

The AI data center law introduces coordinated permitting and a deadline mechanism for administrative reviews. It also provides grid-assessment exemptions for qualifying AI data centers outside the capital region.

The law is scheduled to take effect in February 2027. Its regional incentives could make non-capital locations more competitive, especially for projects that do not require the lowest possible latency to Seoul.

However, faster permits cannot create transmission capacity instantly. Developers still need practical electricity delivery, skilled workers, fiber connectivity, and viable customer contracts.

The legislation also does not erase Seoul’s commercial pull. Enterprises may continue to demand facilities near their users, offices, and network partners.

A distributed model is the more likely outcome. Greater Seoul can retain high-value, latency-sensitive workloads while other regions attract larger training campuses or projects needing abundant land.

That division is not guaranteed. It depends on whether alternative Korean locations can offer reliable power and network connectivity on schedules that match customer demand.

The Seoul vacancy rate should therefore be read as evidence of present tightness, not as a promise of permanent scarcity. It identifies where pressure exists, while leaving the next supply response unresolved.

Three Signals Will Show Whether Seoul Can Add Supply

The next phase will be decided by delivered capacity, regional migration, and whether Johor absorbs its much larger pipeline.

The first signal is Greater Seoul’s operational capacity and vacancy in the next semiannual market update. Both figures must be considered together.

If capacity rises materially while vacancy remains near 1%, the case for durable demand becomes stronger. That outcome would show that customers continue absorbing new supply despite delivery growth.

If vacancy rises sharply after several projects open, the current shortage may prove partly cyclical. It could also reveal that tenants reserved more capacity than they immediately needed.

A flat capacity figure would point toward supply constraints. In that case, low vacancy would say less about accelerating demand and more about the difficulty of completing projects.

The second signal is how South Korea implements its AI data center law before February 2027. The critical evidence will be actual non-capital projects gaining approvals, grid connections, financing, and anchor customers.

Announcements alone will not establish a successful regional shift. Developers need credible energization dates, which identify when a facility can receive operating power.

If projects outside Greater Seoul move from planning into construction, the government’s decentralization strategy gains credibility. It would reduce pressure on the capital-region grid without abandoning national AI capacity growth.

If customers continue insisting on Seoul-area locations, incentives may be insufficient. Policymakers would then face a harder choice between reinforcing the metropolitan grid and accepting slower capacity growth.

The distribution of workload types will matter too. Large AI training campuses offer a stronger test of regional development than small facilities serving local enterprise applications.

The third signal is Johor’s ability to absorb its 3,088-megawatt development pipeline. Johor enters this test with just 0.7% vacancy and significant hyperscale interest.

Continued preleasing would strengthen the argument that regional AI and cloud demand can support extraordinary expansion. It would also increase pressure on Seoul to improve development certainty.

Delays, cancellations, or rising vacancy would weaken that conclusion. They would suggest that even a market with favorable land and large campuses can build ahead of realized demand.

Johor’s trajectory matters to Korean planners because infrastructure capital is mobile. Operators compare Asian markets when workloads allow geographic flexibility.

Yet data sovereignty, latency, network design, and customer preferences keep these markets from becoming one interchangeable pool. Seoul retains advantages that Johor cannot replicate for every Korean workload.

The central question is whether those advantages justify waiting for scarce power and difficult sites. Developers and enterprise buyers will answer through contracts, not rankings.

For buyers, the practical response is to evaluate capacity earlier and examine what a vacancy figure hides. A facility’s power density, delivery commitment, expansion rights, and network routes can matter more than nominal availability.

For policymakers, the challenge is broader. South Korea must add the computing infrastructure needed for its AI strategy without deepening the capital region’s grid imbalance.

For operators, low Seoul data center vacancy creates pricing and leasing leverage today. It also creates an obligation to prove that contracted projects can receive power and open on schedule.

Seoul’s 1.1% rate is therefore both an achievement and a warning. Demand has validated the market, but the next stage cannot rely on scarcity alone.

Watch the next capacity delivery, the first projects advanced under regional AI incentives, and Johor’s pipeline absorption. Together, those signals will show whether Seoul’s shortage becomes a durable advantage or a constraint on Korea’s AI ambitions.