Edition 14 identified five gates separating announced megawatts from bankable megawatts: site, grid, tenant, financing structure, and execution. This edition follows one of those gates, financing structure, past construction and into the years when leases renew and debt must be repaid or refinanced.
Getting financed is not the same as staying safe. A financed megawatt can still fail. Here is how.
Using datacenterHawk's commissioned-capacity definition, five APAC markets now exceed a gigawatt, and the region absorbed almost 1 GW of new take-up in the first quarter of 2026 — roughly half of everything absorbed in all of 2025, in a single quarter.
That looks like strong demand. But look at who is signing the leases. Most take-up still comes from hyperscalers. Neocloud providers are becoming a larger part of APAC demand, while sovereign-AI programmes are also expanding across the region. These are two different credit categories, and neither should automatically be underwritten as hyperscaler parent credit.
Johor shows why financing structure matters. It roughly doubled operational capacity in 2025, from 401 MW to 897 MW (up 124%), and ranked first in Cushman & Wakefield's APAC Maturity Index composite. Grid access is a live constraint there too: Wood Mackenzie now identifies transmission and distribution access as the primary bottleneck for new Johor developments. So project selection and tenant quality matter more — but the lease is not the asset's only source of value.
A financed data centre does not have one simple tenor ladder. It has several contractual clocks, and they reach their decision points at different times.
The debt clock — private placement
The refinancing clock — bank loan or securitisation
The compute clock — neocloud tenant
And if a tenant fails, the building may not re-let quickly — and how quickly is an engineering question as much as a market one. Re-leasing depends on how much of the electrical and cooling design a new tenant can reuse: power density, UPS topology, cooling architecture, and whether the facility can be divided into smaller halls. A single-tenant or high-density facility built to one specification can be materially slower and more expensive to backfill than a standardised multi-tenant hall — which is why two buildings in the same market can carry very different collateral and residual values.
Lease tail = lease expiry − the relevant repayment or refinancing date. Positive tail: contracted lease remains after the debt decision date. Negative tail: the financing extends beyond the initial lease expiry.
A longer lease is not automatically enough. What matters is how much lease remains at the repayment date, who provides the rent, and what the building is worth without that tenant — alongside leverage, coverage, escalators, and residual value.
A divergence is forming across APAC. Tokyo and Sydney may offer deeper replacement demand and more established financing markets; in Tokyo, industry reporting indicates grid-connection queues can extend up to a decade. But that resilience is often already reflected in land prices and yields, so "safer" is a hypothesis about risk versus price paid, not a guarantee. Johor is the faster-growing, less-seasoned case.
The US$5.3 trillion of expected AI- and data-centre-related capital expenditure by the major technology companies leading the build-out, 2025–2030 (Goldman Sachs Research), will be funded through internal cash generation alongside public bonds, private placements, and asset-backed structures. Across those structures, debt maturity, lease expiry, and the tenant's own contracted revenue rarely align perfectly — and that misalignment is where the risk sits.
So the thesis running through Editions 11 to 14 still holds, and gets sharper. The constraints have accumulated — from power and grid access, to bankable capacity, and now to the alignment between debt maturity, lease term, and tenant revenue. The megawatt clears. The debt gets issued. The mismatch often crystallises later — at tenant renewal, default, an anticipated repayment date, or refinancing — and becomes most damaging where replacement demand, asset adaptability, and sponsor support are weakest.
The tenant's customer contracts are often the shortest of the three timelines, and they form the operating cash-flow layer beneath the rent commitment. So what has to be true about the tenant, or the building, for the debt above it to survive a tenant that does not renew?
Key Sources
→ datacenterHawk — 1Q 2026 Asia-Pacific report (five >1 GW commissioned markets; ~1 GW quarterly absorption; Tokyo grid queues)
→ Cushman & Wakefield — APAC Data Centre update 2026 (Johor 401→897 MW, +124%; 0.7% vacancy; Maturity Index composite)
→ Wood Mackenzie — Powering Johor's Data Centre Boom, Jun 2026 (transmission/distribution as primary constraint)
→ Introl / industry reporting — Tokyo grid-connection queues up to a decade, Jan 2026
→ A&O Shearman — data-centre financing tenors, Jul 2026 (private placement 20+yr vs ~10yr lease; securitisation ~5yr ARD; step-ups and accelerated amortisation)
→ Norton Rose Fulbright — data-centre financing, European perspective (20–35yr placements; neocloud contracts ≤5yr; multiple financing drivers)
→ Bisnow (citing JLL and Mintz) — neocloud lease-vs-contract mismatch (10–15yr leases vs 2–5yr GPU contracts), Sep 2025
→ Ropes & Gray — Data Center Investment in 2026 (financeable leases; credit wrappers; fixed rent-start dates)
→ Goldman Sachs Research — "Private Markets Are Expected to Have a Growing Role in Data Center Financing," Jun 2026 (US$5.3T AI/data-centre capex 2025–2030)
→ Morgan Stanley — "Bridging a $1.5tr Data Center Financing Gap," 2025 (~US$800B private-credit opportunity through 2028, led by asset-based finance)
[DIRECTIONAL] Morgan Stanley ~US$800B private-credit opportunity through 2028 — projected financing opportunity, not capital already deployed
[INFERENCE] Johor's replacement-demand depth relative to Tokyo, and the relative backfill difficulty of customised vs standardised facilities, are underwriting judgments, not directly reported comparative datasets
[REPORTED] Tenor structures, absorption, Johor capacity, and the Tokyo connection wait sourced to the named primary, legal, market-research, and industry publications above
No investment advice intended or implied.
Glossary — Terms used in this edition
| Term | Full name | Plain English |
|---|---|---|
| ABS | Asset-backed security | Debt secured on a pool of assets — here, data-centre lease cash flows |
| APAC | Asia-Pacific | The Asia-Pacific region |
| ARD | Anticipated repayment date | A securitisation's soft maturity — the date the market expects refinancing, before legal final maturity |
| Credit wrapper | — | A stronger party (e.g. a hyperscaler parent) guaranteeing a weaker tenant's obligations |
| First loss | — | The capital layer (sponsor equity, then junior debt) that absorbs losses before the senior lender |
| GPU | Graphics processing unit | The core AI-compute chip |
| IG | Investment grade | A credit rating denoting relatively low default risk |
| Lease tail | — | The lease term still remaining at the debt's repayment or refinancing date |
| Neocloud | — | A specialised AI/GPU cloud provider, often on short customer contracts under a long lease |
| Private placement | — | Long-dated debt (often 20–35 years) sold directly to insurers and pensions |
| Tenor | — | The length of time until a debt or contract matures |