Is Denmark Asking the Wrong Question About AI's Energy Use

As Denmark worries about the electricity consumed by artificial intelligence, this article asks whether AI could ultimately save far more energy than it uses.

By David A. Williams

Denmark's debate about artificial intelligence has become trapped inside a deceptively simple calculation. AI requires data centers, data centers consume enormous quantities of electricity, and Denmark's grid is already under pressure from electrification, industry, transport and the transition away from fossil fuels. Therefore, the argument goes, every new AI facility must be understood primarily as another burden on a scarce national resource. It is a reasonable concern, but it measures only what artificial intelligence takes from the electricity system, not what it might enable the wider economy to save. Al Gore has recently argued that AI could help eliminate as much as 5.4 billion tonnes of greenhouse-gas emissions annually by 2035 through efficiencies in electricity, transport and food systems. If that estimate is even partially correct, the political discussion should no longer be confined to how much power AI consumes. Denmark must also ask what it does with that power.

The distinction matters because data centers are highly visible consumers, while the savings created by computation are dispersed across society. A new server complex appears as a large connection request on the grid. Its electricity demand can be calculated, debated and placed alongside that of factories, heat pumps and electric vehicles. The energy it might save elsewhere is much harder to see. It may appear as fewer empty lorries on European roads, lower fertilizer use on farms, earlier detection of failing wind turbines, better management of electricity storage or more accurate forecasts of when power will be abundant. The consumption takes place behind one fence. The benefits may be spread across thousands of buildings, machines and journeys. Politics tends to notice concentrated costs more readily than distributed gains, which creates the risk that Denmark will restrict the technology it can see consuming electricity while ignoring the much larger quantities of energy being wasted throughout the economy.

This does not mean the environmental concerns surrounding AI are exaggerated. The International Energy Agency expects global electricity consumption by data centers to more than double to around 945 terawatt-hours by 2030. AI is helping to drive that increase, and Denmark cannot pretend that its own renewable electricity is unlimited simply because a large share comes from wind. Grid capacity, generation, substations and transmission cables remain physical assets that take years to build. Data centers also compete for land, water and skilled labour. Their economic contribution varies considerably, particularly when they employ relatively few people once construction is complete. The arrival of an industry with almost unlimited demand for electricity should therefore provoke scrutiny. It should not, however, produce an indiscriminate hostility to computation itself.

The Danish argument often treats electricity consumption as though it were a complete measure of social value. It is not. A megawatt used to generate disposable online advertising cannot automatically be considered equivalent to a megawatt used to identify methane leaks, coordinate renewable generation or improve industrial efficiency. Both appear identical to the grid, but their consequences are radically different. Electricity is not valuable merely because it exists; its importance lies in what it allows a society to accomplish. Denmark already makes political choices about land, taxation, infrastructure and industrial development. It may soon have to make equally deliberate choices about computing power.

Artificial intelligence can be especially valuable because the energy transition is becoming a problem of coordination as much as production. Denmark can build more wind turbines, solar parks and interconnectors, but renewable electricity arrives according to the weather rather than the habits of consumers. At the same time, millions of heat pumps, electric vehicles, batteries, buildings and industrial machines will need to respond to changing prices and grid conditions. The result is a system too complex to manage efficiently through human decision-making alone. AI can forecast supply and demand, shift consumption towards periods of abundance, identify congestion before it becomes critical and coordinate thousands of small assets as though they were parts of a single power station. The more renewable electricity Denmark produces, the more intelligence it will need to prevent that electricity from being wasted.

The same principle applies beyond the grid. Transport networks waste fuel because vehicles take inefficient routes or travel partially empty. Manufacturers lose energy through poorly timed production, defective components and machinery maintained only after performance has deteriorated. Farms apply water, pesticides and fertilizers across entire fields even though conditions may vary from one section to another. Supermarkets and distributors discard food because demand was predicted badly. Buildings are heated, cooled and illuminated according to fixed schedules rather than actual use. None of these inefficiencies look dramatic in isolation, but together they form an enormous reservoir of avoidable emissions. AI's environmental promise lies not in one spectacular invention, but in its ability to discover millions of small inefficiencies that human beings have accepted as part of normal economic life.

Gore's figure of 5.4 billion tonnes should nevertheless be treated as a potential rather than a guaranteed outcome. Artificial intelligence does not possess an environmental purpose of its own. It optimizes the objective it is given, and a company seeking higher sales may use greater efficiency to expand production rather than reduce emissions. A transport business that lowers the cost of each journey may simply operate more vehicles. A factory that consumes less electricity per product may manufacture more products. An efficient system can therefore increase total consumption, a phenomenon economists recognize as the rebound effect. AI may also be used to locate new oil reserves, accelerate extraction or persuade consumers to purchase goods they do not need. The same intelligence that reduces waste can stimulate demand.

This is why Denmark should move beyond the blunt choice between approving data centers and obstructing them. Grid access could be connected to measurable national benefits. Operators that reduce consumption during periods of scarcity should be treated differently from those demanding uninterrupted maximum power. Facilities that recover heat for district-heating systems should receive recognition for doing so. Projects promising environmental benefits should disclose what their computing capacity will actually support and publish audited figures covering electricity, water, heat recovery and emissions. A company seeking privileged access to scarce Danish infrastructure should be expected to offer more than an impressive building and a renewable-energy purchasing agreement.

Denmark could also reserve part of its future computing capacity for applications serving the public interest. Universities, hospitals, municipalities, energy companies and agricultural researchers will increasingly need access to advanced computing, yet much of Europe's AI infrastructure remains controlled by foreign technology groups. If Denmark allows its electricity to power the expansion of artificial intelligence without securing domestic access to the resulting capabilities, it may carry the physical cost while surrendering most of the economic value. The country could become a digital equivalent of a raw-material exporter, supplying clean power and suitable locations while the most valuable models, platforms and intellectual property remain elsewhere.

A national climate-computing strategy would offer a more ambitious alternative. Instead of treating data centers as isolated property developments, Denmark could connect grid access with research partnerships, flexible electricity consumption and practical climate objectives. Computing resources could be directed towards offshore-wind maintenance, district-heating optimization, agricultural productivity, shipping efficiency and the management of an increasingly decentralized electricity system. The aim would not be to declare every algorithm environmentally beneficial. It would be to ensure that the country distinguishes between artificial intelligence that merely consumes Danish electricity and artificial intelligence that improves Denmark's ability to use electricity intelligently.

There is a greater economic danger in focusing exclusively on AI's power demand. If Denmark restricts computing infrastructure without reducing its dependence on AI, the processing will simply take place elsewhere. Danish companies will continue purchasing cloud services, Danish data will continue improving foreign systems, and Danish money will continue financing data centers in countries where the electricity may be generated with more carbon. Denmark could then achieve a tidier national energy account without producing any meaningful global climate benefit. It would have exported both the emissions and the economic opportunity.

The real question is therefore not whether artificial intelligence uses electricity. Of course it does. The question is whether the value created by that electricity exceeds its environmental and social cost. Gore's 5.4-billion-tonne estimate challenges Denmark to examine both sides of the calculation. A data center can be an excessive consumer, but computation can also make a renewable grid more flexible, a transport network more efficient and a food system less wasteful. Which future emerges will depend on the conditions governments impose and the purposes they choose to reward.

Denmark should ration scarce grid capacity carefully, but scarcity is a reason to make intelligent choices, not an excuse to avoid them. Electricity devoted to pointless computation is waste. Electricity used to reduce much larger inefficiencies across society may be one of the most valuable investments the country can make. AI is not automatically a climate solution, any more than it is automatically an environmental threat. It is an amplifier. Denmark's task is to decide what it wants amplified.

Sources

Financial Times: Al Gore on AI's climate potential

International Energy Agency: Energy and AI

Climate TRACE: Global emissions monitoring using AI and satellite data

Climate TRACE: Database covering more than 352 million assets