Australia’s data centre pipeline is large, it is growing quickly, and on current evidence a meaningful share of it may not convert on the timeline the market is currently pricing it in. Although compute demand is real and, if anything, understated in most public forecasts, and as per latest AEMO QED Q2 2026 report ~9GW of projects currently working through the connection queue (with 12 project accounting 7.6GW being in application stage) underpinning a firm floor of the DC opportunity.

It is more of a statement about binding constraints before a GW of announced capacity actually converts. Infrastructure that physically exists when the project needs it, and development certainty sufficient for capital to commit to assets, which are returning their investment over 20+ years investment horizon. Both conditions are currently challenged at the scale, the ambition requires, and the reason both matters together is that they compound.

This article sets out why the gap between ambition and readiness is widening rather than closing. Instead of waiting for regulation to solve the problem for data centres, we start seeing more innovative structures leveraging creative funding vehicles spreading out cashflow risks and implementing guarantee structures to underwrite warehousing risk more effectively, similar to vendor finance structures observed in other asset classes, all aiming at unlocking institutional capital at scale.

The ambition-reality gap leads to increasing growing pains

According to AEMO’s 2026 ISP, Data centre electricity consumption is forecast to rise roughly sevenfold over the next decade, from around 5 TWh currently to approximately 34 TWh, taking data centres from about 2% of grid-supplied electricity to somewhere between 6% and 12% depending on the horizon. In New South Wales specifically, the share is expected to move from around 4% to 11% by 2030.

AEMO’s own consultation material identifies around 225 known data centre projects and a connection pipeline of roughly 5.4 GW of large-scale, transmission-connected proposals above 5 MW, concentrated in New South Wales and Victoria. According to the same database though, more than 40% of data centre projects since 2025 have either dropped out of the queue or regressed in connection status.

This is precisely the gap a Senate inquiry into AI and data centres has been examining through 2026. Submissions from the Planning Institute of Australia and the Green Building Council of Australia made a structural point highlighting data centres not being recognised as a distinct land use in most state planning frameworks, so they are assessed under the same pathways as any other commercial or industrial development, with no nationally coordinated approach to siting, no integrated treatment of cumulative water and energy demand, and no dedicated monitoring function tracking the pipeline against grid, water, and workforce capacity in aggregate.

Hence, regulators are being asked to approve gigawatt-scale, essential-infrastructure-grade projects using tools built for warehouses. The inquiry’s recommendations, including dedicated planning pathways, national monitoring of cumulative load, and “additionality” obligations tying data centre power purchase agreements to new generation, are sensible, and the government’s stated timeline potentially points to legislation being introduced in early 2027 at the earliest.

The ISP-level ambition is running well ahead of the regulatory architecture needed to convert it, and every quarter that gap persists adds to the share of the pipeline that stalls rather than converts.

Infrastructure is the key constraint, and it is binding on several fronts

Energy is the most visible bottleneck but not the only one. The gap between how quickly a data centre can be built and how quickly new firmed generation can be delivered is a more fundamental problem. With onshore wind projects typically taking anything between five and up to ten years from planning to operations (BESS and Utility Scale Solar somewhere between five and seven years), a data centre can be developed and build in around three to five years.

In practice, to ensure firmed generation – targeting data centre demand – actually getting build, developers are confronted to underwrite long-dated offtake agreements, potentially before the underlying renewable project has reached financial close, materially increasing the upfront capital a project needs before it produces any revenue. Though this might work in theory, current discrepancy in wholesale electricity prices and build costs for firmed generation would actually require data centres to pay a premium to current offtake prices in order to get the energy supply incl. firming off the ground.

That mismatch is why developer such as Zerra, the Singaporean developer behind a proposed 1 GW facility in regional Queensland, has approached gas generators about co-locating rather than waiting on new renewable build; as in practical reality BESS cannot (yet) economically deliver the 48 hours of backup a gigawatt-scale, 99.999%-uptime facility requires.

New rules developing nationally through the AEMC, and separate rules New South Wales has now legislated on its own account (discussed in more detail below), are addressing part of this by requiring data centres to fund the network augmentation and new generation their own load requires, rather than socialising those costs across household consumers. By requiring registration as National Electricity Market participants, AEMO has visibility over large inverter-based loads.

The dynamic nature gets more and more considered a big issue under current rules, because basically the data centre capacity is able to swing in either direction in no time creating issues for NSPs in balancing supply and demand. Wind needs EMT models of every inverter and controller. BESS needs to demonstrate fault ride-through behaviour in simulation before connection but data centres are more or less considered as passive load (thus not required to meet performance standards or full EMT modelling obligations but rather aggregate load models which underestimate the dynamic nature) but with increasing awareness that they’re nothing like that.

Leaving system impacts aside, data centres as offtakers being confronted to pay a premium to enable dedicated supply projects to reach FID is not a temporary pricing quirk. It reflects a structural gap that the current suite of government underwriting has not closed, and is not obviously designed to close on the timeline data centres need.

The Capacity Investment Scheme (being a fiscal scheme), the Commonwealth’s national cap-and-collar tender for new generation and storage, has seen auction competition push CIS-awarded prices down to levels market participants describe as insufficient on their own to satisfy debt funders and/or Equity IRR expectations, so most CIS-contracted projects still need a separate corporate PPA to reach financial close.

That is the “buyers’ strike” this market has been living with for several years. Offtake volumes of roughly 1 to 1.5 GW a year, and only around 500 MW a year contracted ahead of financial close outside exceptional years, against the 6 GW a year of new capacity the Commonwealth’s own 2030 targets require.

New South Wales’ equivalent Long-Term Energy Service Agreement scheme (being ultimately funded by consumers) operates on the same cap-and-floor principle at state level and has helped some projects, batteries in particular, over the line, but neither CIS nor LTESA has closed this gap at the pace or scale the more than 2 GW of wind capacity referenced later in this article needs to reach financial investment decisions before the end of 2026.

South Australia’s Firm Energy Reliability Mechanism shows the same pattern at smaller scale, being a tender for long-duration firming capacity that only closed its first round in late 2025, with contracted projects not due to reach commercial operation until 2028 to 2031, and with its own reliability and liquidity obligations still undefined.

The reform intended to fix this properly at a national level, the Electricity Services Entry Mechanism, would have a central buyer warehouse long-tenor contracts specifically to bridge the mismatch between the roughly fifteen-year revenue certainty new generation needs and the shorter-term contracts buyers prefer, but it remains in the design phase with key questions yet to be answered. A working group is due to deliver recommended term sheets in November 2026, final contract templates in 2027, and a first tender only in late 2027, with commentators already questioning whether a structure built for trading liquidity can be made bankable for financiers on that timeline.

None of CIS, LTESA, or FERM is currently closing the gap between PPA and LCOE-plus-firming economics at the scale the data centre pipeline needs, and ESEM, the reform explicitly designed to do so, is unlikely to be tested in a live tender before 2028, and it is not yet clear which projects would be awarded (i.e. tender vs. open book) and how support would be structured.

Water and workforce sit alongside energy rather than behind it. Sydney Water’s own forecasts for example, puts data centre water demand at up to 90bn litres annually by 2035, representing 15% to 20% of the city’s total water supply, a figure well above industry’s own more conservative estimates and one that is already prompting utilities toward “take or pay” forward-funding arrangements in which data centre operators pre-fund augmentation ahead of approved network upgrades.

Workforce capacity is arguably tighter still. Powering Skills Organisation puts Australia’s additional electrician and energy-trades shortfall at 72,000 workers by 2030, of which the data centre sector alone may account for ~13,000, drawing from the same finite pool of skilled trades as the 1.2mn homes committed under the Housing Accord and the broader renewable energy build-out. Industry bodies describe the practical effect as “poaching”, thus large projects recruiting apprentices trained by the small and medium businesses that cannot compete on wages.

None of energy, water, or workforce constraints is decisive on its own, but a project needing all three simultaneously, on a compressed construction timeline, in a market where all three are already tight, faces a materially higher delivery risk than the headline MW figure may suggest.

The capital allocation problem this creates

Superannuation and infrastructure capital is the natural long-term owner of digital infrastructure. It values what a mature toll road, regulated utility, or contracted energy asset offers, being long-dated, contracted with good credit counterparties or regulated revenue, a limited and well-understood risk profile, and a long-term deployment of capital.

But the data centre assets actually being financed to date do not meet those criteria. Credit analysis of hyperscaler-backed financing shows the sector increasingly relies on project finance structures, ring-fenced vehicles borrowing against a specific project’s cash flows rather than a hyperscaler’s full balance sheet, with global data centre lease obligations now estimated at roughly USD 1.4 tn, of which around USD 1.1 tn sits off-balance-sheet until leases commence operationally.

Spreads on hyperscaler paper are already trading wider than comparably rated bonds, a function of heavy and concentrated debt supply rather than deteriorating credit quality. Analysts are explicit that credit quality beyond 2027ff depends on whether cloud and AI compute demand and pricing power hold up, not on the presence of an offtake agreement alone. That is a fundamentally different risk than the one an infrastructure investor would be able to underwrite.

Layering the Australian-specific cost allocation reforms on top and the mismatch sharpens further. Requiring data centre operators to fund or underwrite their own network augmentation, underwrite additionality-linked renewable PPAs and Tolls ahead of financial close, and forward-fund water infrastructure may be the correct policy response to the alternative, which is socialising those costs onto household electricity and water bills. But it means a much larger share of total project capital commitment by the data centre developer itself, well before the facility is generating revenue, at precisely the point in the project lifecycle where risk is highest and least suited to infrastructure-style capital.

Two possible outcomes emerge, with both carrying consequences for consumers even though the reforms were designed to protect them. Either developers absorb the augmentation cost directly, in which case only the most well-capitalised sponsors, typically hyperscalers or backed platforms willing to accept development-stage risk, can compete, concentrating the pipeline in fewer hands and pricing out the smaller and mid-market developers who might otherwise diversify supply. Or the connecting infrastructure, generation, transmission, water networks, is instead funded by utilities, network businesses, or the state / commonwealth, in which case that capital ultimately seeks its return through regulated asset bases and network tariffs that do flow back to consumers, just through a different and less visible channel than the direct cost-pass-through the reforms were meant to prevent.

The market’s immediate response may be engineering the certainty capital needs

Two developments through 2026 show the market may be moving to solve the certainty problem directly rather than waiting for policy to solve it, one on the energy side and one on the financing side, and both are worth examining for what they reveal about who is now underwriting what.

On energy, New South Wales has stopped waiting for a national settlement and legislated its own regime, empowering its Energy Minister to regulate data centres independently of the National Electricity Rules. The state offers a 75-day fast-track approval to projects that source 40% of their electricity from onshore wind, alongside entry bonds for connection applicants, a Major Network Upgrade Fee of $200k per MW in the Sydney-Newcastle-Wollongong network and $100k per MW elsewhere, and a requirement that operators pay for the full capacity reserved for them whether or not they draw it, with distribution businesses able to recover their transmission upgrade costs directly from data centre customers.

Queensland and the Northern Territory have declined to adopt an equivalent mandate, and federal AI infrastructure standards now in development are expected to go further again, requiring data centres to supply their own additional renewable generation. The practical effect is to make behind-the-meter generation attractive less because it is the cheapest option and more because it is the only way for a developer to control timing and step outside the state fee schedule altogether.

At least 2 GW of wind capacity is currently in direct negotiation with data centre offtakers in NSW, racing towards final investment decisions before the end of 2026, including Squadron Energy’s 414 MW Uungala project, the only one with construction actually underway, Tilt Renewables’ 700 MW Liverpool Range Stage 1 targeting a decision by year end, and Squadron’s 700 MW Spicers Creek, still seeking a contracted offtaker. Market participants describe the bar for a bankable agreement as having risen sharply on both coverage and tenor, and note that data centres rarely disclose their energy sourcing arrangements publicly, which slows the matching of supply to demand at exactly the pace the fast-track approval windows assume it will happen.

In effect, the data centre has become both offtaker of last resort for the wind farm and financier of first resort for its own grid connection, roles that used to sit with a utility, a retailer, or a government-backed underwriting scheme.

The same step-up is happening in project finance, at a much larger scale, through parties such as Nvidia. Nvidia was reported to be considering guaranteeing up to USD 250bn of the debt behind OpenAI’s Ohio data centre campus; the figure was scaled back, first to under USD 120bn and then to a confirmed USD 105bn, after the original number triggered a share price fall on investor concern about circular financing, the pattern in which the world’s dominant chip supplier increasingly helps finance the very infrastructure that creates demand for its own chips.

The revised structure has Nvidia guaranteeing certain lease and power payments and backstopping the infrastructure’s value if OpenAI defaults, while investing USD 1.5bn directly into SB Energy, the SoftBank-backed developer, and remaining the exclusive chip supplier for the facility’s first phase.

A fortnight later, Nvidia announced a separate and far larger initiative, with financing platforms established with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR intended to mobilise more than USD 500bn of third-party capital, explicitly targeting pension and insurance money, into compute infrastructure. Nvidia’s own framing of the rationale is rather than offering a conventional credit enhancement, the company argues its compute is inherently investable because it is “fungible and transferable across customers and operators,” while Goldman Sachs has said the goal is to build “a market for credit backed by NVIDIA compute.”

Both can be described as doing what infrastructure-grade capital has been asking data centres to do all along, providing offtake and increasing revenue visibility and certainty needed to unlock projects. The difference is that here the guarantor is also the equipment vendor whose own revenue depends on the project proceeding.

Why the shape of underwriting matters

Other industries have gone through a similar playbook before, with mixed results. Vehicle and equipment manufacturers have long used captive finance arms to underwrite residual values on leased equipment specifically to keep sales volumes moving through a dealer network, subordinating the profitability of the financing itself to the manufacturer’s broader goal of continued unit sales. Where the guaranteed residual reflects genuine, diversified secondary demand, the arrangement works well and has done for decades. Where the guarantee exists mainly to keep volume flowing to a single seller, the manufacturer is effectively underwriting its own sales pipeline, and the risk does not disappear but rather it simply moves further from the point where it originated.

The financial crisis produced a larger-scale version of the same dynamic, in the warehousing and securitisation of mortgage risk into structures marketed as diversified and investment grade, where the true quality of what was being pooled mattered far less to buyers than the presence of a rating and a structure that looked familiar. Nvidia’s own language about building “a market for credit backed by NVIDIA compute” sits deliberately in that lineage, attempting to do for compute capacity what securitisation did for mortgage pools and what project finance does for toll roads and utilities, turning an operating asset into something a pension fund can hold at scale.

That is not inherently a problem. It becomes one only if the certainty being sold is manufactured rather than real, meaning it depends on a small set of related parties continuing to transact with each other rather than on independent, diversified end demand. The Bank for International Settlements made close to this exact point in its 2026 Annual Report, naming the concentration of AI infrastructure financing as one of three major threats to global financial stability, on the basis that revenue, debt, and supplier investment are increasingly concentrated among the same counterparties, which makes conventional credit assessment “less reliable than it looks.”

The starkest version of the pattern, and one worth keeping in view precisely because it looks nothing like a hyperscaler, is P&R Container, the German shipping-container leasing scheme that collapsed into insolvency in 2018. Investors bought containers, P&R leased them to shipping lines on the investor’s behalf, and P&R promised steady quarterly rental income plus a guaranteed buyback at 65% of the original purchase price after five years. The scheme ran for more than a decade, raised roughly EUR3.5bn from around 54,000 investors, and was sold almost entirely on the strength of that guarantee.

When it unravelled, administrators found P&R had sold roughly a million more containers than it actually owned so the guarantee had never been backed by a real, growing asset pool, but funded, Ponzi-style, from new investor money. Nobody is suggesting any compute financing platform is anything like that, and the comparison is not meant to imply otherwise.

The narrower and more useful point is that a guaranteed buyback or a promised steady return is only as sound as the independent asset value and cash flow standing behind it, and the more a guarantee’s credibility rests on the guarantor’s own continued willingness or ability to keep transacting, rather than on a diversified pool of end users independently paying for a service they value, the closer that guarantee sits to a first-loss piece of the capital stack than to a genuine transfer of risk.

For anyone assessing Australian data centre capital, that is a useful discipline to apply to every underwriting or guarantee structure now emerging, whether it is on a project basis, a network cost-recovery arrangement, or a chip-vendor-backed lease. The relevant question is not whether a guarantee exists. It is who actually absorbs the loss if compute demand, power delivery, or resale value falls short of plan, and whether that party’s interest in the deal proceeding is independent of the outcome it has been asked to guarantee.

What needs to happen for the pipeline to convert

The practical implication for anyone advising on, developing, or allocating capital to this sector is that revenue and offtake certainty cannot be treated as a financing detail to be resolved after planning and grid connection are secured, and it cannot be taken at face value simply because a guarantee has been announced. It is the mechanism that determines which pool of capital, infrastructure-grade or development-grade, can actually fund the project, and that choice in turn determines who ultimately bears the cost of the connecting infrastructure, energy, water, and network alike.

Structuring data centre projects so that infrastructure-style capital can participate earlier, through genuinely long-dated hyperscaler leases with credit support that survives beyond a single AI capex cycle and does not rely on the guarantor’s own commercial interest in the outcome, through clearer national rules on cost allocation and additionality that remove jurisdiction-by-jurisdiction uncertainty, and through planning reform that gives data centres a defined approval pathway rather than a planning scheme never designed to assess them, would do more to convert the announced pipeline into operating capacity than any further upward revision to the demand forecast, or any further scaling of vendor-backed guarantees.

The gap Australia needs to close by 2030 is not one of ambition, and increasingly it is not one of capital supply either. It is the gap to front investment in enabling infrastructure, relieve pressure on consumer prices and making sure that expected demand contributes it’s fair share in setting it up and commits long term so that infrastructure owners are able to return their investment, all whilst not eroding data centre investment returns over proportionally.

Stefan Dorp is Associate Director, Capital Advisory at Rennie Advisory. If you’d like to have a chat on how we can support your projects getting off the ground in the current market environment, get in touch with myself or anyone from the Capital Advisory team.