NextFin

AI Demand Is Structural, but Australia’s Capital Cycle Is Getting Less Forgiving

Summarized by NextFin AI
  • AI infrastructure demand is structural: generative-AI electricity use could rise from 91 TWh to 857 TWh by 2040, requiring substantial power, grid and data-center investment.
  • Data-center returns remain cyclical: scarce connected power and contracted customers may create durable value, while speculative capacity faces utilization, hardware and financing risks.
  • Iron ore shows a volume-to-value disconnect: export volumes rose 4.4% to 248 million tonnes, but earnings declined, with government forecasts falling to A$79 billion by 2030–31.
  • Private credit is expanding under scrutiny: Australian loans may exceed A$250 billion, while opacity in valuations, governance and refinancing could amplify cyclical stress without implying an immediate systemic crisis.

NextFin News - Australia’s latest investment signals point in different directions: Macquarie’s research says generative-AI electricity demand could rise ninefold by 2040, while government forecasts show iron-ore export earnings falling from A$123 billion in 2025–26 to A$79 billion in real terms by 2030–31, and regulators are scrutinizing a private-credit market estimated at more than A$250 billion. The tension is not that the physical economy is disappearing. It is that structural demand can coexist with cyclical losses when capital mistakes scale for cash flow.

The question is whether investors are financing the scarce parts of the system—power, transmission, land, cooling, customer contracts and transparent collateral—or simply capitalizing persuasive themes. That distinction reaches from data centers to miners and lenders across Australia and New Zealand. It also explains why the three stories belong together without implying a single regional crisis.

The Three Signals Are One Capital-Allocation Test

Macquarie’s October research supports a durable build-out of AI infrastructure while showing why the economics will not be uniform. Its base case implies around 120 gigawatts of installed data-center capacity by 2040, about 106 GW more than today, and roughly $US4.5 trillion to $US5.6 trillion of construction and related infrastructure investment in today’s dollars. The estimate includes data centers, generation and grid connections as well as IT equipment, so it describes a capital cycle rather than a software upgrade.

A capital cycle can be real and still destroy value. Macquarie estimates generative-AI electricity demand could rise from about 91 terawatt-hours today to roughly 857 TWh in 2040, while its high case reaches 1,219 TWh. Yet the same research identifies generative AI as the newest and fastest-growing component of the broader investment story—the part most exposed to incorrect assumptions about utilization, hardware life and customer demand. Aggregate demand can rise while individual projects fail if power arrives late or a customer does not take contracted capacity.

Iron ore supplies the contrasting signal. The Department of Industry, Science and Resources says Australia’s iron-ore export volumes rose 4.4% year on year to 248 million tonnes in the June quarter of 2026, while export earnings fell 1.5%. More tonnes did not compensate for a lower export unit value. The department forecasts export earnings to decline from A$123 billion in 2025–26 to A$107 billion in 2026–27 and A$79 billion in 2030–31, in real terms and based on 2026–27 dollars.

New Zealand’s private-credit story supplies the financial transmission channel. The Reserve Bank of New Zealand says offshore lenders, including private-credit providers, are becoming more active in commercial property and development finance. It also says the non-bank sector remains small by international standards and that systemic risks remain limited given its size. In Australia, ASIC’s November report says some estimates put private-credit loans above A$250 billion in 2026 and identifies continuing weaknesses in governance, disclosure, valuation, reporting and distribution.

These are not three versions of the same trade. They are three tests of whether capital can distinguish durable demand from temporary pricing, productive finance from opaque leverage, and volume growth from cash-flow growth.

AI Demand Is Structural, but Data-Center Returns Are Cyclical

The strongest conclusion from the data-center evidence is structural: electricity demand from AI is not a one-quarter phenomenon. Macquarie’s base case projects generative-AI electricity use from about 91 TWh today to roughly 857 TWh in 2040, even after modelling hardware and software efficiency gains. The transmission mechanism is straightforward. More inference and agentic workloads increase compute demand; compute increases server utilization; servers increase cooling and electricity requirements; electricity demand creates a need for generation, transmission and firm capacity.

The second-order effect matters more than the headline demand number. Power becomes the scarce input that determines which data-center capacity can be monetized. A server hall without a timely grid connection is a claim on future infrastructure, not a fully functioning digital asset. That shifts bargaining power toward utilities, transmission owners, landowners and governments that control permits. It also changes the credit profile: lenders must underwrite not only a tenant’s ability to pay but the timing of interconnection, the durability of power contracts and the risk that a customer’s preferred hardware changes before the facility reaches full utilization.

Macquarie’s cost split puts the physical challenge in context. Around 40% of the $US4.5 trillion to $US5.6 trillion required through 2040 would cover the data center, power generation and grid connection, with the balance in IT equipment. That makes the sector less like a conventional property boom and more like a layered infrastructure system. The useful asset is the combination of land, electricity, cooling, fiber, equipment and contracted demand. Remove one layer and the residual value can fall faster than the headline capacity number suggests.

“AI-driven power demand growth is real and significant.” — Macquarie Asset Management, From Prompts to Power: Generative AI and Implications for Data Centre Demand, October 2026.

The sentence is a statement about demand, not a blanket endorsement of every data-center valuation. The cyclical component sits in monetization. Customers can delay leases, concentrate spending in the most efficient facilities or renegotiate capacity. Efficiency can reduce electricity per unit of compute while lower costs stimulate more usage, supporting total demand but still hurting projects built on the wrong specification.

This is why a generic AI-bubble debate is incomplete. A bubble forms when supply outruns cash-generating demand at prices that justify new capacity. The evidence supports a structural build-out, but not uniform economics across the build-out. The durable asset is scarce, connected power with credible customers. The cyclical asset is capacity whose economics depend on a forecast of utilization, hardware longevity or a single tenant’s expansion plan.

Iron Ore Is a Different Warning: More Volume, Less Value

Iron ore is a mature industry confronting supply-demand normalization rather than a new secular shortage. The Australian government’s September quarterly report says export volumes increased 4.4% year on year to 248 Mt in the June quarter of 2026, yet export earnings declined 1.5%. It attributes the earnings pressure largely to a lower export unit value, including a change in the benchmark iron content from 62% in the June quarter of 2025 to 61% in the June quarter of 2026.

The first-order effect is lower national income when prices fall. The second-order effect runs through company margins, government revenue, the Australian dollar and domestic investment. When producers expand output into a softer price environment, productivity may protect individual operations while aggregate supply keeps the marginal price under pressure. This is the volume-versus-value trap: operational success at the producer level can coexist with weaker export income at the national level.

The department forecasts iron-ore export earnings to fall from A$107 billion in 2026–27 to A$79 billion in 2030–31 in real terms, while iron ore will still account for more than 25% of all Australian resource and energy export earnings over the outlook period. That combination matters. A declining share of a large export base can still influence the currency and fiscal outlook, particularly as domestic investment is redirected toward energy transition and digital infrastructure.

The structural judgment is mixed. The price decline is partly cyclical: Chinese steel demand, inventories and global construction can mean-revert, and lower prices can force high-cost producers out. The supply response is more durable. New capacity from Australia and other producers, combined with declining ore grades and a changing Chinese growth model, means the old assumption that high export volumes automatically produce high export income no longer holds in the government’s forecast horizon.

That is the second-order link to data centers. Both sectors compete for infrastructure and capital, but their cash-flow signatures differ. Iron ore generates revenue through a globally priced commodity whose marginal value can fall as supply rises. Data centers require large upfront capital and then seek long-duration contracted cash flow. Treating both as “real assets” can obscure the difference between price-taking exposure and contracted infrastructure exposure.

The counter-thesis is that iron-ore prices could remain above official forecasts because supply disruptions, Chinese stimulus or stronger Indian and Southeast Asian steel demand absorb new tonnage. The government says markets outside China may provide support. That case would delay the earnings decline, but it would not remove the medium-term volume-to-value pressure. The falsifying signal for the bearish earnings view would be a sustained reversal in the unit-value trend, with iron-ore export earnings rising for at least four consecutive quarters despite higher volumes and the department materially revising its 2030–31 forecast above A$79 billion.

Private Credit Turns Opacity Into a Macro Variable

Private credit matters because it is where a real-economy investment boom becomes a balance-sheet question. ASIC says some estimates put Australian private-credit loans above A$250 billion in 2026 and says the sector needs stronger governance, transparent disclosure, sound valuation, reporting and distribution practices. Its surveillance reviewed 24 funds with credit strategies, primarily private-credit funds, between November 2025 and August 2026.

The mechanism is not that private credit is inherently unstable. Illiquidity and limited disclosure can delay recognition of losses. A bank loan is monitored through a regulated balance sheet, capital rules and recurring supervisory reporting. A private loan can remain marked near par while a borrower depends on payment-in-kind interest, covenant amendments or refinancing that assumes asset values will recover. When refinancing fails, the loss can appear as a discontinuity rather than a smooth price adjustment.

New Zealand is a smaller but revealing case. The Reserve Bank says offshore lenders, including private-credit providers, are becoming more active in domestic commercial property and development finance, while the non-bank sector remains small and systemic risk remains limited given its size. That combination—small today, growing at the edge of regulated visibility—is why monitoring matters. The risk is not necessarily a banking shock. It is a credit-allocation shock that can first hit developers, construction firms and property valuations, then affect banks indirectly through shared borrowers, collateral and refinancing demand.

“While systemic risks from non-bank lenders in New Zealand remain limited given their size, their growing role in credit provision highlights the importance of ongoing monitoring.” — Reserve Bank of New Zealand, Financial Stability Report, November 2026.

The second-order consequence is a change in the value of information. As private credit expands, investors need visibility into loan-to-value ratios, sector concentration, maturity walls, valuation policy and sponsor support—not only a headline yield. In an environment where AI infrastructure and property development both require large amounts of capital, opacity can become procyclical: easy money funds more projects, rising valuations support more lending, and a small deterioration in take-up or rents forces a large repricing.

ASIC’s counter-thesis is important. Private credit fills a financing gap, supports growth and productivity, and can offer diversification and attractive risk-adjusted returns. Banks may be less willing to finance specialized assets, while private lenders can tailor structures and monitor borrowers directly. That advantage is strongest when underwriting is disciplined and collateral is transparent. The bearish thesis would be weakened if ASIC’s next surveillance cycle found materially improved disclosure and valuation practices while realized losses remained below provisions through a full refinancing cycle. Until then, the issue is not the existence of private credit but whether its growth is outrunning the infrastructure needed to price it.

The Adversarial Case: This Is Not a Unified Warning

The strongest argument against a unified capital-allocation warning is that the three developments have different economics and time horizons. AI infrastructure may be productive even if some projects fail. Iron ore can suffer a cyclical price downturn while established miners remain cash-generative. Private credit can grow safely because lenders hold senior claims and negotiate directly with borrowers. Combining them risks turning sector-specific observations into a macro narrative broader than the evidence.

That counter-thesis deserves weight. Macquarie’s demand forecast extends to 2040 and assumes efficiency gains; it is not a short-term revenue forecast. The Australian iron-ore report says lower prices might force high-cost producers out, eventually rebalancing supply. The Reserve Bank explicitly says systemic risks from New Zealand non-banks remain limited. Those facts argue against calling an imminent regional credit crisis or a general collapse in real assets.

But the common thread is not synchronized collapse. It is the growing penalty for confusing scale with resilience. Ninefold AI electricity growth does not guarantee that every gigawatt earns an acceptable return. Higher iron-ore export volumes do not guarantee higher national income. More private-credit issuance does not guarantee better diversification if funds hold similar commercial-property and development exposures. The cross-sector risk is an expectation gap: capital may price the first-order story—AI demand, mining volume and private-credit yield—while underpricing the constraints that determine cash flow.

The single falsifying signal for this broader thesis would be a simultaneous improvement in the underlying conversion ratios: data-center capacity additions accompanied by rising contracted utilization and falling delivered-power costs; iron-ore volumes rising with export unit values and earnings; and private-credit assets growing while valuation, disclosure and realized-loss metrics improve through a refinancing cycle. If those conditions appear together, the warning will have described an adjustment that never became a regime problem.

The evidence points to a split verdict. AI infrastructure demand is structural, while return dispersion within it is cyclical. Iron ore’s price pressure is cyclical, but the erosion of the volume-to-value relationship is a durable medium-term pressure. Private credit’s benefits are real, but opacity is a structural vulnerability that becomes visible during cyclical stress.

What Comes Next for Australia and New Zealand

In the short term, sentiment and liquidity will dominate. Data-center developers, utilities and infrastructure funds will be most sensitive to power-connection milestones and customer commitments. Australian miners remain exposed to daily commodity pricing and the Australian dollar, while New Zealand property borrowers face the terms offered by offshore and domestic non-bank lenders. These market reactions can arrive before the long-run thesis changes.

Over the medium term, cash flow will separate the winners. For data centers, the key evidence is contracted utilization, power availability and the share of capital expenditure that produces usable capacity rather than speculative shells. For iron ore, the decisive metric is export earnings relative to tonnes. For private credit, the focus is non-accruals, restructurings, valuation marks, maturity extensions and exposure to commercial property and development finance.

Over the long term, the structural beneficiaries are owners of scarce, connected power and transmission, infrastructure with contracted demand, and lenders with transparent senior claims and conservative valuations. The exposed are projects dependent on uncontracted power, miners whose strategy assumes price will follow volume, and funds whose reported stability depends on infrequent marks rather than realized exits.

The base case is a two-speed adjustment: AI infrastructure investment continues, but capital becomes selective; iron-ore earnings decline toward the government’s lower long-run path without an abrupt supply shock; and private credit grows under heavier scrutiny rather than triggering systemic stress. The upside case is faster AI monetization, stronger non-China steel demand and improved private-credit transparency, lifting investment without broadening risk. The downside case is a failed AI return cycle combined with weaker Chinese steel demand and commercial-property refinancing stress, pressuring miners, developers and lenders at the same time.

The next decisive evidence will not be another headline about AI demand. It will be whether power-constrained projects secure long-term customers, whether iron-ore export earnings continue to fall faster than volumes, and whether regulators make private-credit marks informative before borrowers need to refinance. The structural thesis is wrong if those conversion ratios improve together; it is right if capital keeps expanding faster than the cash flows and information needed to support it.

Australia’s warning is not that real assets have become unreal. It is that the next cycle will reward scarce infrastructure and transparent cash flow—not exposure to a theme, a tonne count or a yield label.

Explore more exclusive insights at nextfin.ai.

Insights

What makes AI power demand structural?

How much data center capacity is forecast?

How does a capital cycle allocate money?

Why is power scarce for data centers?

How can efficiency hurt AI returns?

Why can more iron ore mean less value?

What drives iron ore earnings declines?

How does private credit transmit risk?

Why does private credit opacity matter?

What does ASIC seek from private credit?

How does New Zealand credit differ?

Can AI demand grow while projects fail?

Which assets deliver durable cash flow?

Can AI, mining, and credit risks align?

How do AI assets differ from iron ore?

What did ASIC find in its 2026 review?

Will regulation tighten private credit?

Search, ask or analyze anything...