Edition 14 · Weekly · APAC · Infra

The MW That Doesn't Clear

APAC's data-centre pipeline is increasingly being repriced by bankability, not demand — and the premium is shifting from plans to proof.

By Sel Fang, Lim · Data centre and infrastructure finance, APAC
17 / 33
Approved Seoul-region projects delayed by community pressure — the site-deliverability gate, live REPORTED
~US$1.5B
Reported size of AirTrunk's prospective Singapore REIT — hyperscale risk moving onto a public tape REPORTED
~270 bps
Yield dispersion across three SGX data-centre REITs — the market already pricing clearing probability DIRECTIONAL

Edition 13 closed on a question: if offtake credit must now be financeable before an APAC project breaks ground, which planned regional builds get repriced before construction begins? This edition answers it directly.

The deal

The repricing will not begin with failed headline demand. Across APAC's leading hyperscale markets, demand remains substantial. The harder question is whether that demand can be converted into a site-specific, creditworthy and financeable commitment.

The market still counts pipeline in announced megawatts: the press release, the land parcel, the power application and the hyperscaler expression of interest. Capital increasingly prices something narrower: not announced megawatts, but the probability that those megawatts become bankable assets.

In Korea, a Savills assessment of the Seoul Metropolitan Area found that more than half of the 33 approved data-centre projects are facing challenges or delays — with 17 of those 33 specifically experiencing delays tied to community pressure, alongside difficulty securing operators and construction setbacks. REPORTED

In Singapore, AirTrunk is reported to be preparing a confidential filing for a Singapore-listed real estate investment trust (REIT) that could raise roughly US$1.5B. If completed, the transaction would place at least part of its operating data-centre exposure into a daily public-market valuation framework, although the proposed asset perimeter, leverage and capital structure are not yet public. REPORTED

Neither signal says that hyperscale demand has disappeared. Both say that the market is becoming more selective about which megawatts deserve capital. The question for a fiduciary is no longer how large the announced pipeline is. It is how much of that pipeline can clear into a bankable asset.

The Bankable MW Framework

A data-centre pipeline loses capacity at every stage between announcement and stable operation. This edition uses one frame to track that loss — an analytical construct of this newsletter, not an established industry formula.

Announced MWTechnical MWBankable MWOperating MW

Announced MW is what the developer intends to build. Technical MW is what secured power, cooling and IT design can physically support. Bankable MW is the portion of technically deliverable capacity with a credible pathway through the development, commercial and financing gates. Operating MW is what ultimately reaches energisation, tenant commencement and stable cash flow. The market usually reports the first number; capital should underwrite the third.

The engineering read

For a facility already designed, useful compute capacity is constrained by its weakest physical input.

Technical MW = min(power, cooling, design)

A campus advertised at 100MW does not necessarily deliver 100MW of useful compute. If the electrical infrastructure supports 100MW, the cooling system 80MW and the deployed design 90MW, technical capacity is 80MW. That is the engineering constraint. Bankability sits one layer above it.

Site status, tenant covenant, financing structure and execution risk are not measured in megawatts, so they should not sit inside the same minimum equation. Instead:

Bankable MW = Technical MW × Clearing Probability

One clarification the framework depends on: power sits in two places, and they are not the same thing. Committed, energised power — a firm connection the electrons already reach — sets the ceiling on Technical MW. The risk of getting from a queue position to that firm connection, on schedule, sits inside Clearing Probability as grid deliverability. Prior editions traced grid access as APAC's defining bottleneck; here it earns its own gate.

Clearing probability is not fixed. As multiple unresolved gates accumulate, the effective clearing probability generally declines more rapidly than any single risk viewed in isolation — a project four gates from clearing is worth far less than any one gate's odds suggest. This is why early-stage announced pipeline deserves a steeper discount than a naive read of its headline capacity.

The five bankability gates

1 · Site deliverability — the legal, physical and social ability to construct on schedule. In Korea, community opposition lowers site deliverability, lowers clearing probability, and yields fewer bankable MW. A permit is not a right to build.

2 · Grid deliverability — the risk between a queue position and a firm, energised connection. Where a tenant requires clean-energy procurement or carbon-free matching, compliant clean supply becomes part of this gate. Energised power sets Technical MW; the probability of reaching energisation on time sits here.

3 · Tenant bankability — whether the offtaker's covenant is financeable, not merely famous. A non-binding expression of interest is not a lease that lenders will advance against.

4 · Financing structure — capital and maturities matched to the cash-flow ramp. The risk is not the headline cost of debt; it is maturity mismatch that forces refinancing before stabilisation.

5 · Execution risk — where separate constraints become one delivery schedule: lead times, commissioning, sequencing. The framework recognises that each unresolved gate reduces the likelihood that announced capacity ultimately becomes bankable. The gates are not independent — a weak covenant raises financing cost, a grid delay drives execution overruns — so the effective haircut is usually deeper than treating each gate's odds as unrelated would imply.

The capital allocation read

The old pipeline shorthand is losing value

The old shorthand — land plus power application plus hyperscale interest equals pipeline — is giving way to a new one: technical MW times clearing probability equals bankable MW. The difference is the development haircut, and it should be applied before construction, not discovered after capital is committed.

Public markets can price the clearing probability

AirTrunk's reported Singapore REIT matters not because one listing can value every APAC project, but because of what a listed vehicle reveals. On the exchange, pure-play data-centre REITs already trade at very different required returns. As of mid-2026, published market estimates placed Keppel DC REIT's distribution yield in the mid-4% range, NTT DC REIT's near 5.5–6.0%, and Digital Core REIT's near 7.2%. DIRECTIONAL

These are not directly comparable property cap rates. The lower-yielding Keppel DC REIT and the higher-yielding Digital Core REIT illustrate that public markets assign materially different required returns to different combinations of portfolio characteristics — geography, tenant exposure, lease profile, leverage, sponsor, payout and asset maturity together, not any one alone. The spread runs to roughly 270 basis points across three pure-play data-centre REITs on the same exchange.

Public markets price operating assets, not announced capacity — and the gap between those two valuations reflects, among other things, the market's assessment of clearing probability. A listed vehicle makes that gap observable.

The premium moves to proof, not plans

The distinction can already be seen in Ansan. Digital Edge announced a 60MW AI-ready facility on powered land in the Sihwa National Industrial Complex, supported by a 90MVA power agreement and dual feeds from independent 154kV substations. Keppel separately acquired an effective 73% stake in the vehicle owning land for a planned 60MW Ansan data centre. REPORTED Powered industrial land with established substation access raises site deliverability, lifts the clearing probability, and converts more announced MW into bankable MW — before a spade is in the ground. That is capital paying for evidence.

Investment Lens
Constraint
Announced MW overstates what can become financeable infrastructure.
Impact
Bankable MW should be underwritten before construction, not after.
Capital response
Underwrite technical MW and clearing probability separately, then combine them.
Winning asset
An operating or near-operating campus with secured power, binding tenant commitments, deliverable expansion land and capital matched to stabilisation.
Losing asset
A speculative land bank whose value depends simultaneously on future grid access, future approvals, future tenant credit and short-dated refinancing.

What this means for the broader market

The through-line this newsletter has tracked is that scarcity migrates. First the bottleneck was land, then grid access, then equipment, cooling and density. Now the relevant scarcity is broader: the ability to assemble every requirement into an asset that capital can finance with confidence.

This is where the engineering read and the finance read meet. Engineering asks how many megawatts a design can physically support. Finance asks what the probability is that those megawatts clear the site, grid, tenant, capital and execution gates. Their product is bankable capacity. Not every technically possible site becomes a financeable infrastructure asset; not every expression of hyperscale demand becomes lender-underwritable cash flow.

Every announced megawatt carries an implied probability of becoming infrastructure. Investors who ignore that probability are valuing plans. Investors who underwrite it are valuing assets.

Key sources

Korea — site deliverability

Savills Korea (via DC Byte, May 2026) and Savills Prospects — of 33 approved Seoul Metropolitan Area projects, more than half face challenges or delays; 17 of 33 specifically delayed by community pressure.

Singapore — public price discovery

Bloomberg reporting carried by The Business Times / The Edge, June 2026 — AirTrunk near confidential Singapore REIT filing; proposed raise around US$1.5B. Keppel DC REIT, Digital Core REIT and NTT DC REIT disclosures and market data — directional distribution-yield reference points as at mid-2026.

Ansan — project deliverability

Digital Edge (8 June 2026) — 60MW Ansan facility, 90MVA power agreement, dual independent 154kV substation feeds. Keppel (2026) — effective 73% stake in the vehicle owning land for a planned 60MW Ansan data centre.

[REPORTED] Based on company announcements or reputable media reporting. AirTrunk's proposed filing, structure, timetable and asset perimeter remain subject to change. The Savills figures describe one dataset of 33 approved SMA projects.
[DIRECTIONAL] REIT distribution yields are point-in-time figures as at mid-2026, derived as annualised distribution per unit ÷ unit price, and move with market prices: Keppel DC REIT ~mid-4%; NTT DC REIT ~5.5–6.0%; Digital Core REIT ~7.2%. Reference dates and DPU conventions are not perfectly uniform across the three; they are not property capitalisation rates. No investment advice intended or implied.

Glossary

TermFull namePlain English
MWMegawattThe unit of power capacity used as the measure of data-centre scale.
Technical MWWhat secured power, cooling and IT design can physically support: min(power, cooling, design).
Bankable MWTechnically deliverable capacity with a credible pathway through the underwriting gates.
Clearing probabilityThe likelihood that supportable capacity clears its site, grid, tenant, financing and execution gates.
REITReal Estate Investment TrustA listed vehicle holding income-producing property, marked daily on the exchange.
DPUDistribution Per UnitThe payout per REIT unit; the numerator in a distribution yield.
WALEWeighted Average Lease ExpiryAverage remaining lease term, weighted by income.
MAS / SGXMonetary Authority of Singapore / Singapore ExchangeThe financial regulator and the listing venue.
CFECarbon-Free Energy24/7 clean-power matching that some hyperscale leases require.
PPAPower Purchase AgreementA contract to buy power, often used to procure clean supply.
MVA / kVMegavolt-ampere / KilovoltUnits of electrical capacity and voltage describing a grid connection.
SMASeoul Metropolitan AreaThe Seoul capital region.

The Uptime Brief · Where uptime meets capital allocation · Published weekly