A physicist once described civilisation as only possible in an energy surplus. Without it, collapse becomes inevitable. This insight, explored in Planet: Critical’s episode "The Energy Collapse" with Louis Arnoux, highlights a fundamental truth: energy underpins every aspect of modern life. No matter how advanced society becomes, infrastructure, food production, and even artificial intelligence face systemic risks if energy supply fails to meet demand.
In January 2025, US President Donald Trump announced the Stargate Project, a USD 500 billion private-sector initiative involving OpenAI, SoftBank, Oracle, and MGX to build AI infrastructure and data centres across the US, creating over 100,000 jobs. Meanwhile, DeepSeek has developed an AI model that has sparked debates about the future of artificial intelligence. Yet, a critical question remains largely unaddressed as we celebrate these rapid advancements. Are existing energy systems prepared to sustain AI’s growth?
AI’s Insatiable Appetite for Energy
Artificial intelligence is power-hungry. Recent studies show that data centres, the physical backbone of AI, are responsible for billions of dollars in public health costs due to their reliance on fossil fuels. Research from the University of California, Riverside, and Caltech found that US data centres incurred more than USD 5.4 billion in health-related expenses over the last five years, with USD 1.5 billion in 2023 alone. These figures highlight how digital progress is deeply connected to the environmental and economic costs of energy consumption.
As AI adoption accelerates, projections suggest that US AI data centres could require an additional 14 gigawatts of power capacity by 2030. That is roughly the equivalent of adding several large nuclear power plants to the grid. This growing demand raises concerns that AI could become a technology reserved for those with privileged access to vast energy resources, if energy systems do not evolve simultaneously. This would increase technological inequality and force difficult trade-offs between AI and other essential sectors such as healthcare, manufacturing, and agriculture.
Furthermore, these estimates only account for direct energy use. They do not include the wider supply chains, rare earth mining, human labour, and social costs that underpin AI’s infrastructure. The extraction of rare earth elements, such as lithium, cobalt, and neodymium, can cause severe environmental damage and exploitative labour practices. Mining operations, many of which are concentrated in politically unstable regions, often rely on child labour and dangerous working conditions. The water-intensive processing of these minerals also leads to pollution and long-term ecological harm.
AI is often framed as a weightless, abstract revolution, but in reality, it is built on physical resources and carries material consequences.
Energy: The Hidden Constraint on AI’s Future
AI is frequently presented as a seamless, digital solution, existing in the cloud and requiring nothing more than data to function. This view is misleading. In reality, AI is deeply physical, relying on vast amounts of energy, hardware, and global supply chains.
If AI is treated as a universal solution, it risks becoming an expensive and inefficient tool, deployed in situations where it does more harm than good.
Consider a gun. In some cases, it is the right tool for the job, such as in military or law enforcement settings. In others, it is not only dangerous but completely inappropriate. AI is no different. It excels in areas like medical research, logistics, and complex problem-solving, but in many cases, it is being forced into applications where it is unnecessary, inefficient, or even counterproductive.
A colleague once put it another way: "You can cut down a tree with a hammer, but it probably isn’t a good idea". Theoretically, with enough effort, a hammer could be used to fell a tree. But the time, energy, and sheer volume of hammers required would make the process absurdly inefficient. If we insist on using AI for everything, regardless of whether it is the right tool for the job, we create an artificial demand for data centres, electricity, and computational power that may be better spent elsewhere.
A sustainable AI future requires selectivity. AI should not be the default choice for every task. Instead, it should be used strategically, maximising impact in areas where it is truly effective while acknowledging that in many cases, simpler, lower-energy solutions are preferable.
A Scramble for AI Energy Solutions
Rather than making the harder choices, such as prioritising renewable energy expansion and imposing stricter efficiency standards, companies are taking various approaches to securing energy sources for AI.
The Nuclear Pivot
AI firms increasingly seek nuclear energy to meet their growing power demands. In 2024, Google partnered with Kairos Power to develop small modular reactors (SMRs), with plans to generate 500 megawatts of nuclear power by 2030. Microsoft, meanwhile, is using AI to streamline nuclear power plant approvals, hoping to fast-track the construction of new reactors.
This shift signals a growing awareness that AI cannot be sustained on existing grids alone.
Carbon Capture: A Stopgap or a Distraction?
ExxonMobil has predicted that data centres could represent 20% of the carbon capture market by 2050, triggering major investment in carbon capture and storage (CCS). Some view this as a practical bridge to a cleaner energy future. Others argue that CCS is a distraction, allowing industries to delay genuine decarbonisation by relying on a technology that remains costly, inefficient, and difficult to scale.
Hybrid Models: AI-Powered Energy Optimisation
Beyond securing energy sources, AI is also being used to optimise energy efficiency. AI-driven models are already being deployed to balance electricity grids, predict maintenance needs, and improve energy use in wind and solar farms. If applied correctly, AI could reduce its own footprint by making energy systems more efficient.
The Power of AI: A Recent Example
AI’s potential is undeniable. In early 2025, researchers used AI to crack a virus structure in seconds - a problem that had taken scientists years to solve. This breakthrough demonstrates AI’s extraordinary ability to accelerate medical research, predict viral mutations, and develop life-saving treatments.
However, these benefits will only be realised if AI remains widely available. Without reliable, sustainable energy, the advantages of AI could be limited to the most powerful corporations and governments, leaving others locked out.
The Takeaway: A Question of Surplus
A sustainable food supply is essential for human survival. Likewise, a sustainable energy supply is critical for the future of AI and modern technology.
But Arnoux’s insight remains central: civilisation is only possible in an energy surplus. The real question is not just how to power AI, but whether society can afford to sustain it at its current trajectory. If we fail to secure clean, reliable energy, we may be forced to choose between AI’s expansion and the stability of other essential systems.
This is AI’s paradox. It is a tool that can help solve the world’s most pressing challenges, but if left unchecked, it may become one of them. If AI absorbs too much of the global energy surplus, it could destabilise the very systems it is meant to improve. The future of AI depends not just on energy availability, but on energy wisdom, deploying it where it matters most, and knowing when a simpler tool is the better choice.
Let’s not overlook the small stuff, because it’s what holds everything together.