For decades, electricity was an almost invisible variable in the digital economy. A company chose where to locate a data center according to land availability, taxation, connectivity or proximity to customers. The power grid followed. Additional capacity was connected, infrastructure was reinforced, and the cost of that expansion was absorbed into a system designed to accommodate economic growth.

In the Australian state of Victoria, that order is beginning to reverse.

On September 22, the state government announced that new data centers would be expected to secure their own renewable energy supply and storage capacity. Developers would also bear the cost of connecting their facilities and of the grid upgrades required to accommodate them. Drinking water would no longer be treated as an ordinary cooling resource: new projects would be expected to prioritize recycled or non-potable water. Siting requirements would also become more restrictive to limit the effects of large facilities on residential areas and local infrastructure.

The decision may appear local. Yet it reaches into a problem emerging simultaneously across several of the regions most heavily engaged in the artificial intelligence race: computing capacity is now expanding faster than some of the physical systems capable of supporting it.

The digital economy is entering a phase in which the availability of chips is no longer necessarily its principal constraint. Electricity, storage, transmission networks, water and connection times are becoming strategic variables in their own right.

The Grid Is No Longer Background Infrastructure

Artificial intelligence has fundamentally changed the physical scale of the digital economy. Data centers existed long before large AI models, but the concentration of GPUs and other accelerators in increasingly large facilities is creating enormous electrical loads at a limited number of sites.

That change in scale produces a simple contradiction. Technology companies can order equipment, mobilize billions of dollars and build computing infrastructure relatively quickly. Electricity networks operate on a different timetable. New power plants, transmission lines, substations and storage capacity require permits, capital and often years before they become operational.

The constraint, therefore, is no longer simply the total amount of electricity available. It is the ability to deliver substantial amounts of power at the right location, at the right time and with the degree of reliability required by digital infrastructure designed to operate continuously.

As projects become larger, this asymmetry becomes increasingly difficult to ignore. A data center is no longer merely another industrial consumer. In some regions, it is becoming large enough to alter the planning of the entire system that hosts it.

This is precisely what Victoria's emerging doctrine is intended to prevent from becoming a burden on the rest of the economy. The government is not rejecting data centers. On the contrary, it considers them strategically important for digital services, investment and employment. What it is changing are the conditions under which they are welcomed: their expansion should not automatically transfer their energy and infrastructure requirements to other users.

The distinction matters.

The question is no longer simply whether a territory has enough electricity to accommodate AI. It is who should pay for the additional capacity required to generate and deliver it.

From Power Consumer to Energy Infrastructure

That shift could transform the data center business model far more profoundly than a conventional environmental requirement.

If an operator must secure generation, install or contract storage, finance its connection and contribute to grid reinforcement, its project gradually ceases to be merely a computing facility attached to an external energy system. It becomes a hybrid project operating at the boundary between digital infrastructure and energy infrastructure.

The industry itself is beginning to move in this direction without waiting for governments to force it.

On September 16, Emerald AI, Google and NVIDIA launched the AI Energy Management Alliance. Its approach rests on a different but complementary idea: instead of treating a data center as a fixed electrical load that the grid must satisfy continuously, some computing workloads could be adjusted according to conditions on the electricity system.

Flexible workloads could be shifted in time, batteries deployed when the grid is under pressure and some computing activity temporarily reduced. The objective is to transform part of computing demand into a controllable resource rather than an inflexible load.

The change sounds technical. It is actually economic.

Since the emergence of cloud computing, one of the model's great advantages has been its abstraction from physical infrastructure. To the user, computing power could appear almost instantly available. For the largest operators, expansion largely meant adding servers, buildings and connections to existing networks.

AI is abruptly reintroducing geography into that abstraction.

A gigawatt available several years from now is not equivalent to a gigawatt available today. Generation capacity hundreds of kilometers away is not necessarily accessible without new transmission lines. A region with abundant electricity but severe water stress does not offer the same conditions as another. And a jurisdiction willing to socialize the cost of grid expansion offers a fundamentally different economic proposition from one requiring the developer to pay.

The price of computing is beginning to incorporate the price of its physical anchoring.

Europe Is Beginning to Count as Well

The movement extends beyond Australia and the United States.

On September 21, the European Commission proposed stronger reporting requirements for data centers with installed IT power demand of at least 500 kilowatts. The framework would increase transparency around energy and water efficiency, exposure to local water stress and the potential contribution of facilities to energy systems, including through waste-heat recovery. Data centers already account for roughly 2.5% of the European Union's electricity consumption, while European computing capacity is expected to expand significantly with the development of AI.

Victoria, the European Union and emerging American initiatives are not constructing the same regulatory system. Their convergence should therefore not be mistaken for an already established global doctrine.

They are, however, responding to the same physical problem.

Governments want more computing capacity. They increasingly regard artificial intelligence as strategic infrastructure, seek to attract investment and simultaneously fear dependence on capacity located elsewhere. Yet the more they attempt to territorialize computing, the more they must territorialize what makes computing possible: electricity, transmission, water, land and storage.

Digital sovereignty is therefore encountering a constraint rarely emphasized in the rhetoric surrounding it: computing capacity cannot be relocated without relocating at least part of the energy system supporting it.

A New Geography of Computing

That constraint could gradually redraw the global map of data centers.

For years, the most attractive locations were those combining connectivity, institutional stability, proximity to major markets and favorable tax conditions. None of those criteria will disappear. But another variable is likely to join them: the ability to add very large quantities of electricity rapidly without destabilizing the existing grid.

Regions with abundant generation, storage potential, space for new infrastructure and relatively rapid connection procedures will enjoy a growing advantage. Conversely, metropolitan areas with already constrained grids may remain technologically attractive while becoming physically incapable of absorbing certain projects.

The change may also alter the industry's financial structure.

A data center associated with power generation, batteries, transmission infrastructure and potentially flexibility contracts requires more capital and broader expertise than a conventional computing facility. Operators capable of integrating these components will gain an advantage. Smaller developers may encounter an increasingly significant barrier to entry.

Victoria is implicitly accepting this trade-off: investment remains desirable, but it must increasingly arrive with some of the infrastructure it requires.

That represents a subtle reversal in the relationship between territory and the digital industry.

During the first phase of cloud computing, territories largely competed to attract data centers. In the phase now beginning, some are starting to ask what those data centers will contribute to the systems expected to accommodate them.

The Real Cost of Intelligence

It would be premature to conclude that data centers are becoming energy-independent. Most will continue to rely on interconnected electricity grids, and that interdependence remains economically rational. Even facilities combining on-site generation and storage generally benefit from maintaining a connection to the wider network.

The break lies elsewhere.

The grid is gradually ceasing to be treated as a passive and unlimited resource available to the digital industry on demand. Its capacity is becoming a scarce asset, and that scarcity must be allocated, priced or compensated.

The technology industry is attempting to respond through flexibility. Governments are beginning to respond through regulation. Power producers see a new market. Grid operators, meanwhile, must learn to manage consumers whose requirements can reach the scale of major industrial facilities.

These responses remain different, but they converge on the same reality: the expansion of artificial intelligence no longer depends solely on how much computing power the industry can manufacture.

It increasingly depends on how much physical power territories can devote to it.

For years, the promise of the cloud was to make us forget where the machine was located. AI is producing precisely the opposite effect. The larger computing becomes, the more its territory reappears.

And with it returns a question far older than artificial intelligence: who pays for the infrastructure that makes growth possible?

Main sources

Government of Victoria — Sustainable Data Centre Action Plan and announcement of September 22, 2026.

Reuters — Victoria proposes mandating new data centres source their own green power, September 22, 2026.

European Commission / Reuters — new European requirements concerning data-center energy and water efficiency, September 21, 2026.

NVIDIA / Emerald AI / Google — launch of the AI Energy Management Alliance, September 16, 2026.