Artificial intelligence (AI) is a revolution in plain sight, and the world surrendered before it even began to resist. Unregulated, electrified, and accelerating faster than our ethics can catch up, it has embedded itself into every corner of modern life with astonishing speed.

But beneath the marvel of its capabilities lies a question too few are asking: What is the cost to the Earth? Not in metaphor, not in theory, but in carbon, in water, in waste. The deeper truth is coming into focus: AI may be the most extractive and environmentally disruptive technology we’ve ever unleashed, and unless we interrogate its impact now, we risk waking up in a world permanently altered, not by the brilliance of its design, but by the relentless ascent of its unchecked rise.

Artificial intelligence does not run on magic. It runs on power. The data centres that fuel these systems — from the large language models we chat with to the recommendation engines shaping our daily lives — are ravenous consumers of electricity. In 2022, AI-driven data centres consumed between 240 and 340 terawatt-hours of electricity globally, a staggering figure that accounts for more than one per cent of the world’s total electricity usage that year.

That is more power than the entire United Kingdom consumes annually. This is not a projection; it’s already happening.

The International Energy Agency warns that this demand will double by 2026 as AI adoption accelerates across every sector. Now, imagine that usage doubling annually.

AI’s energy appetite should have sounded the alarm in a world urgently trying to decarbonise. Instead, it has been repackaged as a growth strategy. Across the United States, utilities are doing the unthinkable: delaying the retirement of coal-fired power plants and reinvesting in fossil fuel infrastructure, driven in no small part by the surging energy demands of AI and data centres. In Georgia, for example, Georgia Power received approval to keep ageing coal plants online specifically to power new server farms, despite previous commitments to shut them down.

Some argue that AI-related electricity use is growing so rapidly that some analysts estimate it could soon outpace Florida’s entire yearly consumption.

Similar reversals have emerged in Virginia and West Virginia, where regulators now warn that more than 9,000 megawatts of fossil fuel generation (once scheduled for closure) may remain active to meet AI-driven electricity loads. And from Utah to Wisconsin, utilities are quietly shelving renewable energy roll-outs to keep pace with this digital surge.

Moreover, if the electricity footprint is staggering, the water toll is perhaps even more deceitful because most people have no idea that artificial intelligence is incredibly thirsty. The same servers that crunch data at superhuman speed run extremely hot, and just like the human body needs sweat to cool down, data centres require massive cooling systems to prevent overheating.

For many facilities, the cheapest and most common method involves water — millions of litres of it. Therefore, cold water is pumped through systems to absorb heat from the server racks, then either evaporated in cooling towers or run through energy-intensive chillers to be reused. This cycle, repeated endlessly at scale, creates a digital thirst that is becoming deeply problematic.

A study by the University of California, Riverside, estimated that training a single AI model, GPT-3 (one of the earlier versions powering generative text tools) consumed more than 700,000 litres of clean water.

Conservatively, that’s enough to meet the daily indoor water needs of 370 people living in the US, and that’s just for one training cycle. Once deployed, every interaction with such models adds a little more to the water bill. Multiply that by billions of daily prompts, queries, translations, and searches, and the picture becomes alarming.

Big tech companies have already begun reporting water usage spikes linked directly to AI development. In 2022, Google disclosed that it had used 5.6 billion gallons of water, much of it for data centre cooling. Microsoft’s water consumption rose by 34 per cent in just one year, primarily due to AI-related projects.

The truly troubling part is where this water is being drawn from. Many of the world’s largest data centres are located in water-stressed regions: Arizona, Iowa, and parts of The Netherlands. Communities in these areas are beginning to push back, asking whether it’s fair or even moral to prioritise machines’ cooling over people’s hydration.

We are not just at an ecological tipping point, but at an ethical one. The real question is: Whose future is being protected, and whose basic needs are being sacrificed to keep the digital economy running cool and smooth?

And then there is another cost, the one we can’t yet see but will soon feel: the waste.

The specialised chips and machines required to build and run AI systems are short-lived. Rapid upgrades, intense processing needs, and tight hardware cycles mean we are producing a new wave of electronic waste that the world is unprepared to handle.

By 2030, generative AI is projected to produce between 1.2 and five million metric tons of e-waste annually. For perspective, that’s nearly 1,000 times more than the AI industry produced just a few years ago. By the end of this decade, that could mean up to 2.5 million tons of discarded servers, chips, and cooling components annually, a figure on par with the world’s total electronics recycling capacity.

Based on history, we already know where much of this waste ends up. Not in pristine recycling labs or high-tech recovery facilities, but rather in landfills, informal e-waste dumps across the Global South, where children and workers burn plastic casings to extract metals, breathing in lead, mercury, and arsenic. It ends up in groundwater, rivers, lungs, and the future.

Solutions to mitigate these harms exist: component reuse, longer hardware lifespans, circular design, but unless we demand them they won’t be implemented. Why? Because they’re not yet profitable. Moreover, they require government regulation, which is moving far slower than the technology.

We in the Caribbean cannot afford to be passive observers of this trend. As small island states facing rising seas, salt-water intrusion, and climate displacement, we know all too well the cost of negligence. We must now raise our voices, not to slow innovation, but to guide it.

Let us be clear-eyed. AI is not evil. But it is being built, trained, and deployed in ways that ignore our planetary limits. So if we are silent now, we risk letting the digital revolution trample the very ecosystems we are fighting to protect.

Let us insist on climate audits for AI models, push for transparency in energy and water usage, and demand that e-waste be tracked, minimised, and dealt with ethically, not outsourced to poorer nations under the cloak of digital advancement.

Innovation must not be a privilege that poisons others; intelligence, artificial or not, must be judged not just by what it can create, but by what it refuses to destroy.

We must not only allow our digital future to be as intelligent as it claims to be, but let it be as ethical as the Earth requires it to be.

Lisa Hanna is Member of Parliament for St Ann South Eastern, People’s National Party spokesperson on foreign affairs and foreign trade, and a former Cabinet member.

Lisa Hanna