Digital infrastructure is entering a new phase. While data centres are now firmly established as critical infrastructure, investment is increasingly shifting deeper into the high-performance GPU clusters powering artificial intelligence. As global capital flows into AI accelerate, the GPU layer is emerging as a critical focal point for investors, lenders and developers alike.
This builds on the broader shift already underway, where AI infrastructure is emerging as a distinct infrastructure asset class (see AI infrastructure and financing: a new infrastructure asset class).
Key takeaways
- AI infrastructure investment is shifting beyond data centres to the Graphics Processing Unit (GPU) layer powering high performance compute.
- GPU clusters are emerging as a distinct infrastructure asset class, underpinned by take-or-pay customer contracts with hyperscalers and major AI customers.
- Bankability is primarily driven by take-or-pay customer contracts, with the GPU hardware providing collateral support rather than the primary credit driver.
- Energy is becoming a critical constraint shaping project viability.
- In Australia, the key challenge is coordination across planning, energy and transmission not access to capital.
From data centres to compute platforms
Digital infrastructure is no longer defined solely by ‘bricks and mortar’, data hosting/storage and connectivity. It is increasingly centred on high-performance compute. GPU clusters and the systems that support them.
GPU infrastructure now represents a substantial proportion of overall AI data centre capital expenditure and is increasingly being structured and financed on a standalone basis.
At the same time, the sponsor mix is evolving. Alongside the major hyperscalers, specialist AI infrastructure platforms (such as CoreWeave and Firmus Technologies) are emerging, building GPU capacity at scale, typically supported by take-or-pay customer contracts with anchor customers.
Private credit has led early transactions in this sector, with commercial banks now showing increasing appetite to participate as structures mature.
Bankability sits in the customer contract
Despite the focus on GPU hardware, financing outcomes are ultimately driven by contracted revenue. Long-dated, take-or-pay arrangements with hyperscalers and AI customers underpin bankability, with the GPUs themselves providing collateral support rather than the primary source of credit strength.
This is aligning GPU financings more closely with traditional infrastructure models, where predictable cashflows underpin lending decisions.
However, new risks are emerging. Lenders are focused on counterparty concentration. Particularly where a small number of US hyperscalers underpin a large share of GPU capacity, as well as asset lifecycle, technology obsolescence and refresh cycle considerations. The rapid pace of GPU generational change means residual value assumptions require careful stress-testing.
As the market develops, transactions are increasingly incorporating infrastructure-style features, including availability-based revenue models and more sophisticated covenant frameworks.
Energy as a financing constraint
This reflects a broader shift already playing out across digital infrastructure where energy strategy is emerging as the primary constraint on development, rather than capital or connectivity (see Power first: why Australia’s data centre boom will be won in the energy market).
AI workloads amplify demand for power, cooling and grid access. As a result, the quality of a project’s energy strategy is increasingly influencing whether it proceeds and on what terms it can be financed.
More sophisticated projects are moving beyond simple consumption models. With co-located storage, flexible demand and integrated energy solutions, data centres and AI infrastructure providers can play a more active role in the energy system by absorbing surplus renewable generation and supporting grid stability.
At the same time, social licence is increasingly a financing consideration. In practice, projects that cannot demonstrate credible renewable integration and alignment with broader system needs are finding it harder to secure offtake, grid connection and ultimately finance.
These dynamics are becoming particularly relevant in Australia, where energy availability and infrastructure coordination (rather than capital access) are emerging as the decisive constraints on large-scale AI compute development.
The Australian opportunity and coordination challenge
Australia is well positioned in the global AI supply chain with abundant land, growing digital infrastructure momentum and, critically, some of the world’s best renewable energy resources. The fundamentals for large-scale AI compute are strong.
However, the key constraint is not capital, but coordination.
Planning frameworks, energy policy, transmission infrastructure and approval processes continue to operate on different timelines (nb: often under different governments) and across different jurisdictions. State-level energy planning does not yet integrate digital infrastructure demand at scale, and the transmission pipeline is not being built with hyperscaler demand in mind. This fragmentation is already shaping which projects proceed and how they are structured.
The Commonwealth’s recently released Expectations of data centres and AI infrastructure developers, while not legally binding, are functioning as a de facto gatekeeper for major projects, particularly in relation to energy strategy, system impact and broader national interest considerations. For foreign sponsors and investors, this sits alongside existing FIRB and critical infrastructure screening requirements, adding a further layer to the approvals landscape.
Projects lacking credible, integrated approaches to energy and infrastructure are increasingly unlikely to progress at pace.
Beyond infrastructure investment, access to large-scale AI compute is increasingly being viewed as a strategic economic capability with implications for productivity, industrial policy and sovereign resilience.
There is also a competitive dimension. AI infrastructure capital is highly mobile. Hyperscalers and specialist compute platforms can deploy capacity across multiple jurisdictions. The Middle East, Southeast Asia and parts of Northern Europe are all actively courting hyperscaler capacity, often with faster permitting timelines and sovereign-backed incentives. Delays in planning, transmission and approvals directly affect Australia’s competitiveness.
What this means
- High performance compute is becoming the core infrastructure asset - with value shifting from data centre real estate to GPU-enabled compute platforms.
- Take-or-pay contracted revenues remain central to bankability - take-or-pay arrangements with creditworthy counterparties continue to underpin financing structures.
- AI compute infrastructure is increasingly inseparable from energy infrastructure - creating new interdependencies across capital markets, energy systems and technology platforms.
- Energy strategy is now fundamental to project viability - shaping risk allocation, financing terms and development timelines.
- Australia’s competitiveness depends on faster coordination – alignment across planning, transmission and energy policy will determine which projects proceed.
- The opportunity is significant – if those coordination challenges are resolved, Australia has the potential to become a globally significant destination for renewable-backed AI compute capacity.
Watch: AI infrastructure, GPUs and financing trends
Watch Jamie Guthrie and James Frixou unpack where GPU-backed financings are heading, why energy strategy is now make-or-break for project bankability, and what needs to change in Australia for the opportunity to be realised.