The Cloud Has to Live Somewhere, and Somebody Has to Pay for It

Professor Ian Brighthope begins his recent argument with an observation so obvious that the language of modern technology has somehow taught us to forget it: the cloud is not in the sky. There is something almost ingenious about the metaphor. We upload photographs "to the cloud," businesses migrate their operations "to the cloud," governments put records "in the cloud," and artificial intelligence draws its extraordinary computational power "from the cloud." The terminology makes the whole operation sound weightless. Information seems to disappear from the laptop and float into an ethereal digital realm requiring neither land, water nor electricity. Nothing could be further from physical reality.

The cloud is concrete, steel, silicon, copper, fibre-optic cable, transformers, cooling equipment, backup generators and enormous quantities of electricity. It occupies industrial land, connects to transmission networks and, depending upon its cooling technology, can consume substantial amounts of water. It produces heat and noise. Behind every apparently immaterial interaction with artificial intelligence sits a chain of material infrastructure extending from the chip to the power station.

Brighthope's argument is not that this infrastructure is unnecessary. Quite the opposite. He explicitly acknowledges that Australia needs substantial domestic data-centre capacity. Hospitals, banks, universities, government agencies, telecommunications networks and businesses depend upon reliable computing and data storage. Australian-hosted computing can reduce latency, strengthen disaster resilience and keep sensitive information within Australian jurisdiction. Data centres support scientific modelling, medical research, cybersecurity, weather forecasting and the rapidly developing AI economy. This concession is important because it separates a serious argument about data centres from technological Luddism. The question is not whether Australia should possess data centres; of course it should. The question is what kind, how many, where, using whose resources, at whose expense, under whose control and ultimately for whose benefit.

That question has suddenly become urgent because AI is transforming the scale of the industry. The International Energy Agency estimates that data centres consumed around 485 terawatt-hours of electricity globally in 2025 and expects consumption to approach 950 TWh by 2030. AI-focused facilities are growing considerably faster than conventional computing facilities, with electricity consumption by AI-focused data centres increasing dramatically as companies race to construct the computational infrastructure required for increasingly powerful models.

Efficiency is improving remarkably quickly, and that qualification matters. Individual AI computations can become cheaper even while total electricity consumption rises. This is the familiar technological paradox in which making something cheaper encourages vastly more of it to be consumed. More efficient chips do not necessarily produce lower aggregate electricity demand if the world simultaneously constructs millions more applications using them.

Brighthope brings this global problem home to Australia, and particularly Victoria. Victoria wants the industry. The Victorian Government has been actively promoting the state as a destination for data-centre investment, advertising its industrial land, fibre connectivity, technology workforce and energy infrastructure. Brighthope cites a development pipeline claimed to exceed $25 billion. Melbourne already contains dozens of operating facilities, and the expansion associated with AI could be vastly larger than the traditional data centres that preceded it.

The City of Melbourne itself now acknowledges the resulting problem. Its Draft Data Centres and AI Infrastructure Position Paper says these facilities can impose substantial demands upon land, electricity, water and other infrastructure. It proposes clearer planning rules, mandatory performance standards, better disclosure, infrastructure contributions and assessment of cumulative impacts. That is an important confirmation of Brighthope's central argument because it comes not from an anti-AI pressure group but from the government of the city around which much of the development is occurring.

The City argues that data centres should be located according to infrastructure capacity, renewable-energy opportunities, sustainable water availability, climate resilience and community expectations. It wants planning controls that distinguish between facilities according to their scale and impact, publicly accessible information about energy, water, emissions, air quality, noise and backup generation, and infrastructure costs appropriately allocated to proponents and users. In other words, many of the questions Brighthope is asking are already being asked by Melbourne's own government.

His strongest case concerns electricity. Australia is simultaneously attempting an enormous transformation of its electricity system. Coal-fired generating capacity is retiring, renewable generation is expanding, storage must be constructed, transmission networks require major investment and households are repeatedly told that this transition is necessary and expensive. Into this already difficult engineering transformation comes a new class of electricity consumer requiring immense quantities of dependable power.

Brighthope cites estimates that Australian data centres consumed around 3.9 TWh in 2024–25, approximately two per cent of grid-supplied electricity in the National Electricity Market. Under the scenario he discusses, that could rise to around 12 TWh by 2030 and 34.5 TWh by 2050. Victoria provides an even more immediate warning. Brighthope reports AEMO figures showing average Victorian data-centre demand rising from 104 MW in the second quarter of 2025 to 204 MW in the corresponding quarter of 2026. Even allowing for volatility and the uncertainty surrounding proposed projects, the direction is unmistakable.

This is where the issue stops being an argument about computers and becomes an argument about political economy: who pays? If a multinational technology company constructs a hyperscale facility requiring vast amounts of additional electricity, the connection to the grid is not conjured into existence by innovation. Transformers, substations, transmission capacity, storage and firming must exist somewhere. If the system requires strengthening because enormous new loads have appeared, somebody must finance that strengthening.

Brighthope's principle is difficult to dispute: the additional cost attributable to these facilities should fall principally upon the companies creating the additional demand rather than being quietly socialised across household electricity bills. There is also an important accounting problem surrounding renewable energy. A company can announce that its data centre operates on renewable electricity because it has contracted renewable generation or purchased certificates, but if it merely claims a share of generation that already existed, the physical electricity system has not necessarily acquired enough additional generation to meet the new load. Somebody else must consume the remaining electricity available from the system.

The meaningful test is therefore additionality. A genuinely defensible large data-centre development should bring sufficient new generation, storage and supporting infrastructure to meet the demand it adds rather than simply acquiring a green label for electricity that somebody else would otherwise have consumed. The Commonwealth itself has moved in this direction, with expectations for data-centre developers calling for additional clean generation or storage, improved efficiency, appropriate contributions toward network costs and avoidance of upward pressure on consumer electricity prices.

Brighthope identifies the weakness immediately: expectations are not obligations. A government can publish an admirable statement of principles while continuing to approve developments whose economic attractiveness depends partly upon costs being borne elsewhere. Electricity systems operate according to engineering constraints rather than press releases. If a two-gigawatt load appears, the physical generating and network capacity required to supply it must ultimately exist somewhere.

Water presents a similar problem, although Brighthope is careful not to exaggerate it. Not every data centre consumes enormous quantities of potable water. Cooling technologies differ substantially, and closed-loop systems, air cooling, recycled water and increasingly sophisticated direct cooling can dramatically alter the water footprint. A sweeping claim that every data centre is draining drinking supplies would therefore be false. The sensible questions are local: how much water will this particular facility consume, where will that water come from, what cooling technology will it use, what happens during drought, can recycled water substitute for potable water, and what are the cumulative demands if several enormous facilities cluster in the same region? Those questions should be answered before approval rather than discovered after construction.

The same principle applies to land and employment. Governments understandably advertise billions of dollars of investment and thousands of construction jobs. Construction employment is real and economically valuable, but construction eventually ends. A highly automated data centre can occupy substantial land and consume extraordinary quantities of electricity while employing relatively modest numbers of permanent workers. The correct calculation therefore cannot stop at the headline investment figure.

Governments should disclose the permanent jobs, apprenticeships, local procurement, tax revenue, public subsidies, infrastructure costs and opportunity costs associated with major projects. A $5 billion development is not automatically a $5 billion benefit to Australians merely because $5 billion was spent constructing something on Australian soil. The relevant question is how much lasting value remains in Australia after the construction crews have gone home.

This leads to the most interesting part of Brighthope's argument: sovereignty. Australia could conceivably become a major global centre for data centres while capturing surprisingly little of the economic or strategic value. Imagine the arrangement. Australians provide the land, Australian electricity generators provide the power, Australian water systems provide the cooling resources, Australian governments facilitate the planning approvals and Australian infrastructure connects the facilities. Yet the intellectual property belongs overseas, the software belongs overseas, the profits flow overseas, the controlling corporation answers to a foreign board and critical technological decisions remain outside Australian control.

Brighthope describes the danger memorably: Australia could become a "digital quarry." It is an excellent analogy. Australia has spent generations digging valuable things out of the ground, shipping them overseas and buying back higher-value manufactured products. The AI version could be stranger still. Instead of exporting iron ore and importing machinery, we could supply the land, electricity and physical infrastructure required to manufacture intelligence while paying foreign corporations for access to the intelligence produced upon our own soil.

Even the Commonwealth has recognised the problem. Brighthope quotes Assistant Minister Andrew Charlton's warning that merely constructing data centres in Australia does not guarantee Australia captures the economics of AI. That admission should fundamentally alter the policy debate. The objective cannot simply be maximising gigawatts of installed data-centre capacity because gigawatts measure electricity consumption, not national prosperity. Nor can success be measured by dollars of announced investment. Governments have an unfortunate habit of treating the capital cost of enormous infrastructure projects as though expenditure itself were a social benefit.

The relevant measure is value captured by Australians. Does the investment create Australian technological capability? Does it develop Australian engineers and researchers? Does it provide Australian companies with affordable computing capacity? Does it create intellectual property here? Are profits taxed here? Does it strengthen universities and domestic AI companies? Does Australia control strategically important infrastructure? If the answers are largely no, the economic case becomes much less impressive.

Brighthope then makes a distinction that deserves much wider attention: data residency is not the same thing as data sovereignty. Residency asks where the server physically sits, whereas sovereignty asks who controls it. A server rack located in Melbourne is physically Australian, but if a foreign corporation controls the encryption keys, system administration, software, contractual terms and ultimately the company itself, Australian control may be much less complete than the geography suggests.

That becomes particularly important for health records, financial information, government data, defence-related information and critical infrastructure. The question is not merely whether the information has crossed the Australian border. It is whether another country's legal system, corporate headquarters or government can exercise authority over the company controlling it.

Brighthope therefore proposes going substantially further than ordinary data-localisation requirements. For the most sensitive national functions, he argues for beneficial Australian ownership and operational control, including control of encryption keys and senior security administration. One can debate how far that requirement should extend. Requiring complete domestic ownership could increase costs, limit access to global expertise or provoke reciprocal restrictions upon Australian firms abroad. But the underlying sovereignty question cannot simply be waved away. A country that regards electricity grids, telecommunications systems and defence facilities as strategic infrastructure cannot logically assume that control over the infrastructure containing its government's digital memory is merely an ordinary commercial matter.

There is another dimension that Brighthope handles carefully. Data centres are not inherently surveillance systems. The building containing servers is no more morally responsible for surveillance than a printing press is responsible for propaganda. The same computational infrastructure can sequence cancer genomes, forecast bushfires, design medicines, detect financial fraud or help a student understand calculus. But it can also support facial recognition, mass behavioural analysis, censorship systems, comprehensive digital identity systems and increasingly sophisticated state surveillance.

Infrastructure is politically neutral only in the narrow sense that silicon possesses no political beliefs. Its capabilities are not neutral. Once immense computational capacity exists, governments and corporations acquire possibilities that previously did not exist at the same scale. The answer is not to smash the computers but to establish democratic limits before capabilities become institutional habits.

This is ultimately why Brighthope's proposal for a temporary moratorium on inadequately governed hyperscale and AI-scale developments deserves serious consideration rather than dismissal as anti-technology. He does not propose stopping ordinary data centres or shutting down Australia's digital economy. His position is more limited: where governments cannot demonstrate adequate electricity supply, water safeguards, planning controls, infrastructure financing and public-interest protections, approvals for enormous new facilities should pause until enforceable rules exist.

That is hardly revolutionary. We would not approve a giant mine without knowing what happened to the tailings, nor a chemical plant without asking what it emitted. We would not construct a new city without considering roads, water and electricity. Yet the cultural glamour surrounding artificial intelligence encourages governments to treat enormous computing facilities as though they were simply unusually sophisticated office buildings. They are nothing of the sort; at hyperscale they are major industrial infrastructure.

Some proposed facilities make the point dramatically. Brighthope discusses the Victorian AI Hub proposal at Plumpton, whose publicly described concept involves a 350-hectare site and ultimate maximum capacity of 2.4 GW. Maximum capacity is not the same as actual initial consumption, and the project remains subject to approvals, qualifications Brighthope properly makes. But merely contemplating infrastructure on that scale tells us that the category has changed. A multi-gigawatt computing complex is not just another technology company opening an office. It is simultaneously an energy-policy, land-use, infrastructure-financing and potentially water-policy decision, and depending upon who ultimately controls the computing capacity, it may become a national-security and sovereignty decision as well.

That is why Brighthope's proposed reforms make sense as a starting framework: transparent registration of major facilities; disclosure of ownership, electricity and water requirements; cumulative environmental assessment; additional electricity generation and storage corresponding to additional load; developer payment for attributable network infrastructure; preference for recycled or low-water cooling; staged approvals; meaningful community participation; and enforceable Australian-benefit requirements rather than promises contained in media releases.

The City of Melbourne's own draft policy independently reaches strikingly similar conclusions. It calls for scale-sensitive planning, public reporting, cumulative-impact assessment, additional renewable generation for major new demand and appropriate allocation of infrastructure costs to proponents and users. That convergence matters. Brighthope's critique is not merely an outsider railing against technological change. Important elements of it overlap with concerns now being expressed by governments confronting the physical consequences of rapid data-centre expansion.

There is nevertheless a balance to maintain. Data centres are not merely parasitic electricity consumers. Computing produces enormous economic and social value, and AI may accelerate scientific discovery, improve medicine, increase industrial productivity and create entirely new industries. Australia would make a serious strategic mistake if excessive regulation drove all advanced computing infrastructure offshore. But the opposite mistake would be equally foolish: believing that because AI may be transformative, every data-centre proposal must therefore represent progress.

The correct question is whether the bargain is good for Australia. If foreign corporations wish to construct enormous computing infrastructure here, there is nothing unreasonable about Australians demanding that they bring the additional electricity they require, pay for the network infrastructure they necessitate, minimise potable-water use, disclose environmental impacts, create demonstrable domestic benefits and operate within Australian sovereignty requirements where critical national information is involved. That is not hostility towards investment. It is the elementary due diligence a competent country should perform before surrendering scarce resources to a fashionable industry.

The great achievement of Brighthope's essay is therefore to drag "the cloud" back to Earth. Once that happens, much of the technological mystique disappears. A data centre becomes visible for what it physically is: a gigantic industrial consumer sitting on a particular piece of land, connected to a particular electricity network, using particular resources, owned by particular people, operating under particular laws and producing economic benefits that ultimately flow somewhere.

Those facts can be measured, and governments should measure them. Who owns the facility and controls the data? Who supplies the electricity and builds the transmission lines? Who pays for them? How much water is consumed? How many permanent jobs remain when construction ends? Where are the profits taxed? Who owns the intellectual property? What happens if geopolitical relations deteriorate? Who can turn the system off and who possesses the encryption keys? Above all, what does Australia receive in exchange?

These questions do not obstruct technological progress. They distinguish national development from simply renting out the country. Australia unquestionably needs the cloud, but the cloud is not in the sky. It sits on Australian land, draws upon Australian infrastructure and increasingly contains the intimate digital machinery of Australian civilisation. Professor Brighthope is therefore right about the principle even where the details of particular projects require case-by-case judgment: if the cloud lives here, consumes resources here and stores our national life here, it must answer to Australians rather than merely occupying Australia.

https://ianbrighthope.substack.com/p/the-cloud-is-not-in-the-sky