Scarcity Doesn't Equal Return — power-secured capacity is scarce in Greater Seoul, but a scarcity premium only becomes a return if it survives to investor IRR

Last edition ended on a distinction: a premium can help a sponsor raise money — it tells you almost nothing about whether the underlying asset will return what an investor is underwriting. This edition follows that thread from the raise to the realisation, and Korea is an unusually clear place to watch it, because the constraint, capital structure and emerging exit pathways are unusually visible at the same time.

Under Korea's grid-impact regime, large loads must clear a Power System Impact Assessment before they can connect. As of March 2026, the capital region had accumulated 522 initial applications; of those, only 10 had passed final supply approval — 1.9% of the original pool — for roughly 1,010 MW against the 33,592 MW applied for. Seoul proper accounts for exactly one of the ten.

One number, two denominators. 10 / 522 = 1.9% is whole-process attrition from the original pool. 10 / 24 = 41.7% is the conditional rate among the 24 projects that reached full review. The 1.9% is a story about how few projects advance — not a claim that the rest were formally rejected.

The scarcity case is supported by the capital entering the market. When Keppel entered Korea in mid-2026, its investment rationale explicitly cited a power-constrained Seoul Metropolitan Area and limited new supply, pointing to JLL data showing 1.4% vacancy with the 2025–27 pipeline fully pre-leased. CBRE's case data — its own estimates, not an official index — show Greater Seoul rents up over 70% since 2019.

Market rent is not the same thing as the rent an asset actually collects. Korean development leans on short-horizon capital — 3–7 year PFV exits — against five-plus-five leases and legacy escalators near 2% a year, with ramp-up discounts early. Market rent can climb double digits while a legacy lease moves only around 2%. The rise is real; it just doesn't arrive inside the fund's holding period.

This is not universal. CBRE notes newer Greater Seoul leases are improving — less ramp-up discounting, some 10-year minimums, 3% or CPI-linked escalators — so the leakage is asset-specific. Which is exactly what makes lease structure, not the market rent, the thing to underwrite.

The scarce asset can quietly become the obsolete one — and in Korea, the owner is often the one who pays to stop it. AI hardware refreshes on a two-to-three-year cycle against a thirty-to-forty-year building. CBRE, drawing on Vertiv, puts current industry-wide installed density near 15–25 kW per rack; higher-density AI configurations can require materially more, forcing recurring reinvestment in power delivery and cooling.

Who carries that cost turns on structure. Under fully-fitted arrangements, more of that retrofit exposure can sit with the landlord — a risk-allocation point, not a claim that landlords universally fund it. It is worth being precise: configuration decides who funds reinvestment, tenor decides when income reprices, and obsolescence decides whether the kit stays economically useful. Three distinct risks that converge only at the exit.

An asset that works for today's owner does not automatically work for tomorrow's buyer — and in a market whose logic is short-horizon capital selling to longer-horizon capital, the next buyer's underwriting becomes one of the things that ultimately determines the exit price. As the holding horizon lengthens, income durability, remaining CapEx, lease terms and technical fit weigh more heavily. A different buyer pool may underwrite the same asset differently — a narrower pool, pricing pressure, and a softer exit than the scarcity implied.

The permanent-capital bar isn't a fixed number you can design to. Keppel DC REIT — a reasonable reference for what that pool currently holds — runs a portfolio around 6.7-year WALE by area, with occupancy that sat at 95.6% at the end of March and slipped to 92.5% by the end of June, a single Cardiff contract expiry accounting for most of the drop (95.3% excluding Cardiff). That detail is the useful one: long WALE doesn't make income permanent; it tells you when the renewal risk arrives. Even perpetual capital feels the expiry.

Exit value, roughly, is forward NOI divided by an exit cap rate; an investor's return is a function of entry price, interim cash flows, CapEx and that exit. Each leak can pull down the same eventual IRR.

The exit channel is a pricing-pressure mechanism; a buyer's required total return and the property's exit cap rate are different metrics and are not equated here.

None of that says Korean data centres are bad investments. It says the scarcity premium is not the return. Scarcity creates optionality — the potential for the premium to be captured. Whether it becomes IRR depends on whether the lease lets the rent through, whether reinvestment doesn't eat it, and whether the finished asset is something the next owner actually wants.

Who carries the cost turns on distinctions that are easy to run together. Physical configuration — fully-fitted or shell & core — sets how much of the equipment and fit-out the landlord provides. Within fully-fitted, the operating model can be colocation or single-tenant. And separately, contractual cost allocation — triple-net, double-net or other — sets who pays operating expenses and property costs. These overlap but are not interchangeable: an asset can be shell & core on a triple-net lease, or fully-fitted on a double-net one. The underwriting question is who carries the refresh CapEx and the operating obligations, not which label sits opposite which.

When Keppel entered Korea via the Ansan Global Meta Data Centre — an approximately 73% effective stake via KDCF III, 60MW incoming power and 40MW IT load, greenfield with power and construction permits secured, service targeted for 2030 — it bought into a power-constrained market on purpose. Keppel has used private-fund-to-listed-REIT recycling for selected stabilised assets elsewhere, and CBRE identifies Ansan as a potential future candidate for that kind of pathway. No transfer to Keppel DC REIT has been committed. That uncertainty is the point: before permanent capital would take an asset like Ansan, the asset has to become something — tenanted with the right credit, stabilised at the right NOI, on lease terms and a technical configuration a perpetual holder will accept. The exit isn't a thing you arrange after development. It's a set of requirements you underwrite from the first day.

That is the discipline the scarcity narrative tends to bury. Secured power gets you pricing power. Pricing power only becomes return if the lease carries it into NOI, reinvestment doesn't consume it, and the finished asset still clears the next buyer's bar. Scarcity creates optionality. Structure decides how much of it becomes IRR. And even when the lease economics work, another question remains: whose credit actually sits behind the rent?