Key Takeaways
- • Dentons analysis finds AI-focused data centres serving neocloud operators such as CoreWeave, Lambda, Crusoe, Nebius, Applied Digital and Fluidstack carry risk profiles that fall outside standard bank lending parameters, with contracts typically running two to five years and usage-based revenue.
- • Liquid cooling often accounts for about 10% of construction costs, and GPU hardware with three to four year useful lives against six-year accounting schedules raises re-leasing and residual value risk, so bespoke AI facilities could become stranded assets in a downside scenario.
- • Private credit now operates across three layers of AI-exposed data centre financing: equipment loans typically five-year and fully amortising against GPU hardware, development-stage lending, and holdco or mezzanine debt often deployed through DevCo/YieldCo structures.
- • Securitisation via rated SPV notes is best suited to stabilised assets with predictable cash flows, while assets exposed to emerging revenue models are less adaptable to this route.
- • Integrated platform-level capital stacks combining bank debt, private credit, institutional equity, ESG-linked financing and potentially securitisation are becoming the market standard, illustrated by Blackstone's approximately A$24 billion acquisition of AirTrunk and NEXTDC's A$1.3 billion equity raise in 2024.
Summary
Dentons, in the latest installment of a thought leadership series on Australian data centre financing published August 11, 2026, argues that AI-focused facilities are creating an asset class whose risk characteristics diverge from traditional infrastructure. The divergence stems not from AI demand itself but from business models in which operators sell GPU compute as a service to neocloud providers, rather than leasing space and power to creditworthy hyperscale tenants on 10 to 25 year fixed-capacity leases. Neoclouds are described as sub-investment-grade, venture-backed and highly leveraged, with revenue dependent on downstream AI demand and GPU pricing.
The physical characteristics of these facilities compound the counterparty risk. Purpose-built AI sites require specialised cooling, high-density power and custom layouts, and liquid cooling alone often accounts for around 10% of total construction costs according to the article. Converting such facilities to general-purpose use is costly, and the pool of replacement tenants for bespoke high-density sites is narrow, which may leave them as stranded assets with depressed liquidation value.
Dentons notes that this divergence is prompting greater selectivity among bank lenders and a significant role for private credit. Private capital is deployed across equipment financing, development-stage lending that bridges to bank or capital markets refinancing, and holdco or mezzanine debt ranking alongside senior facilities. Many platforms use DevCo/YieldCo structures, in which development assets are financed within a ring-fenced entity and transferred to an institutionally financed yield vehicle on stabilisation. The article adds that no single market-standard approach has yet emerged for AI-exposed assets.
Securitisation and institutional capital, including pension funds and insurers, are gaining traction for assets with infrastructure-like characteristics. Under a securitisation, revenue-generating assets are transferred to an SPV that issues tranched, rated notes. Rating analysis focuses on occupancy, contract duration and operator quality rather than individual tenant credit, so investment-grade senior tranches are achievable where these metrics are strong. Dentons cautions that the structure is less adaptable to assets with variable or emerging revenue models.
The article cites equity demand as evidence of institutional appetite. Blackstone's acquisition of AirTrunk for approximately A$24 billion in 2024 was described as the largest data centre transaction in Asia-Pacific history, and NEXTDC raised A$1.3 billion in equity in the same year. Joint ventures between operators and GPUaaS providers are also presented as a means of sharing development risk and improving bankability. Dentons concludes that financing is shifting from single-source, asset-level funding toward integrated platform capital structures, and says a forthcoming article will examine financing case studies involving AirTrunk, NEXTDC, CDC and Macquarie Data Centres.
The physical characteristics of these facilities compound the counterparty risk. Purpose-built AI sites require specialised cooling, high-density power and custom layouts, and liquid cooling alone often accounts for around 10% of total construction costs according to the article. Converting such facilities to general-purpose use is costly, and the pool of replacement tenants for bespoke high-density sites is narrow, which may leave them as stranded assets with depressed liquidation value.
Dentons notes that this divergence is prompting greater selectivity among bank lenders and a significant role for private credit. Private capital is deployed across equipment financing, development-stage lending that bridges to bank or capital markets refinancing, and holdco or mezzanine debt ranking alongside senior facilities. Many platforms use DevCo/YieldCo structures, in which development assets are financed within a ring-fenced entity and transferred to an institutionally financed yield vehicle on stabilisation. The article adds that no single market-standard approach has yet emerged for AI-exposed assets.
Securitisation and institutional capital, including pension funds and insurers, are gaining traction for assets with infrastructure-like characteristics. Under a securitisation, revenue-generating assets are transferred to an SPV that issues tranched, rated notes. Rating analysis focuses on occupancy, contract duration and operator quality rather than individual tenant credit, so investment-grade senior tranches are achievable where these metrics are strong. Dentons cautions that the structure is less adaptable to assets with variable or emerging revenue models.
The article cites equity demand as evidence of institutional appetite. Blackstone's acquisition of AirTrunk for approximately A$24 billion in 2024 was described as the largest data centre transaction in Asia-Pacific history, and NEXTDC raised A$1.3 billion in equity in the same year. Joint ventures between operators and GPUaaS providers are also presented as a means of sharing development risk and improving bankability. Dentons concludes that financing is shifting from single-source, asset-level funding toward integrated platform capital structures, and says a forthcoming article will examine financing case studies involving AirTrunk, NEXTDC, CDC and Macquarie Data Centres.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.