Among all the evidence that productive capital has developed a genuine appetite for yield-bearing industrial assets in Australia, the most convincing is the one that has received the least examination in these terms. The flow of private and institutional capital into digital infrastructure — data centres specifically — over the past five years represents, in my view, the clearest market confirmation that the productive yield thesis is not aspirational but operative. This capital is not speculating on appreciation. It is not seeking negative gearing benefits or capital gains tax discounts. It is responding to a clean income yield from a contracted, net-leased asset that generates predictable returns at a scale the residential investment market structurally cannot match. The hyperscale data centres now being developed and expanded across Australia’s major corridors are exactly what the productive reallocation of capital is supposed to look like: long-duration assets, anchored tenants, infrastructure-grade income streams, and an underlying demand driver — the exponential growth of digital computation — that is not going away. I have found it genuinely encouraging to watch this capital move. What I have found less encouraging is where it is moving to, and what it is consuming in the process.
The same physical inputs that make a site suitable for a data centre make it suitable for advanced manufacturing. Serviced industrial land with high-voltage grid connection, access to water, proximity to logistics corridors and port infrastructure, and sufficient parcel size for a large-footprint facility, this is the specification that both a hyperscale computing campus and a precision manufacturing precinct require. The difference is not in the location logic; it is in the economics of the race. Data centre operators move fast, pay premium rents, offer long lease terms on stabilised assets, and generate the kind of contracted income streams that capital markets price at low capitalisation rates. They are, on conventional investment metrics, exceptional tenants. Advanced manufacturing operators move more slowly, require longer lead times for fit-out and commissioning, often need staged capital commitment, and operate on longer payback periods that stretch the patience of return-oriented capital. In a market where serviced industrial land is scarce and developers must make allocation decisions, the question of which use gets the site is not a close one. Data centres win. The problem is that they are winning in a market where the scarce input they are consuming, serviced land with large-load grid connection, cannot be easily or quickly replaced once it is allocated.
To understand why grid connection is the binding constraint rather than land itself, it is necessary to understand the economics of substation capacity. A substation in a metropolitan industrial zone serves a geographic catchment of industrial parcels and has a finite total capacity, the aggregate load it can simultaneously supply. Ordinary industrial users, a logistics warehouse, a food processing facility, a light manufacturing operation — draw modest loads relative to their site area, cycle between operational and non-operational phases, and can often accept some degree of demand management during peak periods. A hyperscale data centre operates on fundamentally different terms. Its load is continuous — servers run at close to maximum utilisation twenty-four hours a day, every day of the year — and its draw is measured in hundreds of megawatts, rising to over a gigawatt for the largest campuses. A single facility of this scale does not share substation capacity in the way that a mixed industrial precinct does; it consumes it. The capacity allocated to a hyperscale data centre is continuously occupied, not drawn on intermittently, which means that the remaining capacity available to adjacent land parcels is not the theoretical maximum of the substation minus average data centre load — it is the theoretical maximum minus continuous peak data centre load. The practical consequence is that land in the substation’s catchment zone, even if it is correctly zoned, fully serviced with water and road access, and geographically ideal for industrial use, may be effectively stranded for grid-dependent manufacturing operations because the available connection headroom has been fully committed.
The water dimension compounds this. Large-scale data centres require continuous cooling. The thermal management of server infrastructure at hyperscale generates heat at a rate that demands an industrial-volume water supply for cooling systems, or alternatively, air-cooled systems that themselves consume significant power. A major facility can draw tens of millions of litres of water annually. In the same corridors where industrial water allocation is managed within environmental licence limits, the combination of a data centre’s continuous electrical and water draw creates a zone of infrastructure intensity that may foreclose other industrial uses not because the land is unavailable but because the enabling infrastructure has reached capacity. The grid queue data makes this concrete: for large industrial loads, the connection scale required by advanced manufacturing, green hydrogen production, critical minerals processing, connection times in constrained metropolitan corridors are running to twenty-four to thirty-six months under current demand. That queue exists because the total connection volume being sought by all users exceeds the rate at which network operators can augment substation capacity. Data centres, drawing at continuous high load, are not the only entrants in that queue — but they are among the largest and most persistent, and their infrastructure footprint does not diminish once established.
What I am describing is an uncomfortable structural parallel, and I want to name it precisely because the discomfort is analytically important. The preceding pieces of this series have argued that speculative residential property crowded out productive investment not through any deliberate act of harm but through the rational operation of a framework — tax treatment, prudential risk-weighting, collateral valuation — that made residential the optimal allocation decision for the capital most capable of deploying at scale. The digital infrastructure boom is operating through a different framework and toward entirely different ends; these are productive assets generating real economic output, not speculative instruments inflating in the absence of fundamental yield. But the structural dynamic has a family resemblance: capital responding rationally to the best available return signal is consuming scarce physical inputs ahead of a longer-horizon, lower-return-profile set of productive uses that would generate greater economic complexity. The market mechanism is working correctly by the metrics it is designed to optimise. The question is whether those metrics are the right ones for allocating the physical inputs that a complexity-building industrial transition requires.
This is not an argument against data centres. Australia requires digital infrastructure at scale, the demand for computing capacity is not a policy choice but a structural condition of a modern economy, and the capital building that infrastructure is deploying productively in the genuine sense. It is an argument about sequence and allocation: whether the market, left to sort the distribution of scarce physical inputs across the productive economy, will produce an outcome compatible with the broader complexity transition, or whether the allocation of those inputs — grid capacity, serviced industrial land, water — is a strategic question that market mechanisms alone are structurally unable to answer correctly. The capital doing the allocating has no mandate to optimise for national economic complexity. It has a mandate to optimise for risk-adjusted return. In a context of genuine physical scarcity, those mandates may not converge.
In Western Sydney’s data centre corridor — one of the fastest-developing digital infrastructure zones in the Asia-Pacific region — a cluster of hyperscale facilities is now operating at a combined load that tests the capacity of the transmission infrastructure serving the zone. The industrial land parcels in that corridor are zoned correctly for manufacturing use, sized appropriately for large-footprint operations, and located in one of the most strategically significant freight and logistics corridors in the country. What they do not have is accessible grid connection headroom. The substations serving the zone are committed — continuously, at high load — to the data centres that arrived first. The queue for augmented capacity runs to years. A critical minerals processing facility, a precision manufacturing operation, an advanced industrial user of any kind that requires a large-load grid connection cannot occupy that land on any near-term horizon, not because the land is unavailable, not because it has been rezoned or speculated away, but because the enabling infrastructure was allocated — rationally, productively, in direct response to the yield signal this series has been building toward — to a different class of productive asset that got there first. The transition’s fastest-moving capital arrived. It consumed what was available. What it left behind for the transition’s longer-horizon ambitions is the subject of the next piece.
