There was a time, barely a decade ago, when the digital economy sold itself as weightless. Clouds, not coal; code, not combustion. The promise of net zero emissions seemed perfectly compatible with a future of frictionless computation. That illusion is now collapsing under the brute arithmetic of electricity demand.
Artificial intelligence and cryptocurrency have dragged the digital world back into the physical one. Servers need power. Chips generate heat. Data centres consume land, water, and above all, energy. And the scale is no longer trivial. Global data centres already consume hundreds of terawatt-hours annually, with projections suggesting they could approach or exceed 1,000 TWh within this decade. That is not a rounding error, it is the electricity consumption of a major industrial nation.
The uncomfortable reality is that AI is now the primary growth driver of that demand. Estimates vary, partly because companies are opaque, but AI workloads may already account for a substantial share of data centre energy use, potentially rising to 40–50% or more by the end of the decade. Some projections go further: AI alone could soon consume as much electricity as entire countries, or rival the total demand of all existing data centres just a few years ago.
Cryptocurrency, long the villain of environmental critiques, suddenly looks almost static by comparison. Bitcoin mining alone still consumes on the order of 150–170 TWh per year, a staggering figure in isolation. But AI is catching, and in some estimates surpassing it, not because crypto has become efficient, but because AI demand is exploding faster.
This is the first tension: net zero assumes stabilisation or reduction of energy demand, while AI and crypto assume exponential growth in computation. Both cannot be true at the same time without heroic assumptions about energy supply.
The standard response is technological optimism. Renewable energy will scale. Efficiency will improve. Data centres will become greener. And to be fair, there is some basis for this. Efficiency gains have historically offset part of the demand growth, and future scenarios suggest that improvements in hardware and operations could limit data centre electricity to a smaller share of global consumption than feared.
But this optimism quietly assumes something stronger: that efficiency gains will outrun usage growth. That assumption is historically fragile. Every major efficiency breakthrough, from steam engines to microprocessors, has tended to increase total consumption, not reduce it. The rebound effect is not a theoretical curiosity; it is the default pattern of industrial civilisation.
AI intensifies this pattern because it does not merely replace existing activity, it creates new demand. Billions of daily queries, real-time inference, generative media, autonomous systems, each layer multiplies computation. Even if each individual operation becomes cheaper, the total number of operations explodes. As one analysis notes, the wealth and productivity gains from AI may themselves drive even higher overall energy use across the economy.
Crypto, for its part, is structurally tied to energy through proof-of-work systems. Its consumption is not incidental but intrinsic: security is purchased with electricity. While shifts to less energy-intensive mechanisms exist, the economic logic of crypto ensures that whenever prices rise, energy consumption tends to follow.
Put bluntly, both AI and crypto embed energy demand into their core functioning. They are not just digital tools — they are energy conversion systems, turning electricity into intelligence, verification, and economic activity.
This creates the second tension: net zero depends not just on clean energy, but on constrained demand. Yet the political and economic incentives surrounding AI push relentlessly in the opposite direction. Governments want AI leadership. Firms want computational advantage. Investors reward scale, not restraint. Even countries like Australia are being warned that without massive energy expansion, they risk missing out on the AI boom altogether.
So, the system locks itself into a paradox. To remain competitive, nations must expand energy supply for AI. To meet woke climate targets, they must simultaneously limit emissions and often total consumption. The result is what might be called the energy trilemma intensified: clean, reliable, and abundant power — pick all three, but at escalating cost and complexity.
There are, of course, partial resolutions. Data centres are being collocated with renewable generation. New architectures aim to use stranded or surplus energy. Distributed micro-centres promise efficiency gains. But these are adaptations to a deeper reality, not solutions to it. They manage the growth — they do not eliminate it.
And that is why the conflict is not going away.
Net zero was conceived in a world where digital growth seemed energetically negligible. AI and crypto have ended that illusion. The digital economy is now one of the fastest-growing sources of electricity demand, and its trajectory is upward, not stabilising.
The future will not be decided by whether we can make AI or crypto "green" in some narrow sense. It will be decided by a harder question: can an economy built on exponential computation coexist with a zero net ideology? If the West is to survive, zero net must be firmly rejected in favour of an energy-rich future.