Coal is not clean, it’s not beautiful, and we don’t need it for AI

Trump didn’t need to sign an executive order keeping the furnaces burning to ensure America’s dominance of AI, writes Chris Wright

This week, US president Donald Trump issued an executive order aiming to revive “America’s beautiful, clean coal industry” such that it can power the rise in electricity demand from, among other things, artificial intelligence. However, this nostalgic push ignores the reality that has been commonly understood since Dickens: nothing about coal is clean.

What it also ignores are the reasons behind the industry’s 20-year decline in the US, as well as the incredible uncertainty surrounding what might be the future energy demand of AI and the data centres that give it life.

Coal is a dirty business. From the moment you dig it up, often with the help of some of the world’s biggest explosive rigs, you release methane gas, which warms the planet more than 82 times faster than carbon dioxide over 20 years. Across the Appalachian mountains in West Virginia and Kentucky, the most popular type of coal mining literally involves blowing the tops of mountains off, covering streams in rock and dirt, and filling the nearby valleys with pollutants and heavy metals.

Then, when you burn coal to create electricity, you immediately release carbon dioxide, the main driver of climate change. Burning coal currently releases approximately one-fifth of America’s total carbon dioxide emissions, along with a host of other air pollutants known to cause asthma and respiratory diseases. Thanks to the Clean Air Act of 1970, originally signed into law by one of Trump’s Republican icons, Richard Nixon, the level of sulphur emissions from coal-fired power plants has been significantly reduced. The level was due to fall even further thanks to measures introduced by Barack Obama. But these have already been on Trump’s chopping block, as were important updates to protect coal miners from dust inhalation.

The other reality of the coal sector is that it has been in decline since 2008, and coal plants now generate only 15 per cent of America’s electricity. Back in 2008, the US produced over a billion short tons of coal each year. By 2023, production had halved, and despite Trump’s support for the industry, more than 10,000 coal mining jobs were lost during his first term. Today, just over 45,000 people work in US coal mining, a number that is dwarfed by employment figures in other sectors, such as fast food: McDonald’s, for instance, employs nearly 150,000 people.

Times have changed. Those mines are ageing, productivity is falling, and America’s energy landscape has shifted dramatically. Thanks to the shale gas boom, “natural gas” (which is just methane) has become the dominant source of US electricity. Meanwhile, the cost of renewables has plummeted, and wind and solar have already overtaken coal in the US power mix – just as they have in Canada, Germany, and the UK.

Until this week, that terminal decline was expected to continue. Ironically, on the same day as Trump issued his executive order, the US Energy Information Administration released its short-term energy outlook, projecting that coal production will fall another 9 per cent by 2026. This, they noted, was due to “coal’s continued competition with natural gas and renewables in the electric power sector”.

Donald Trump meets coal workers last week, on the day that he signed an executive order intended to revive the industry (Reuters)

However, while coal has faded from the American energy landscape, a whole new set of power demands seems to be waiting on the very near horizon. Artificial intelligence (AI) and Bitcoin mining are already driving an explosion in data-centre electricity consumption in the US, and many analysts are predicting that this could prove a critical challenge for America’s grid in the near future.

One of the biggest problems is that AI technologies such as ChatGPT are thought to require enormous amounts of computing power to run and train. Early estimates suggest that each AI query might use up to 10 times more energy than an average Google search – at least, before the Gemini chatbot took over. If you follow this trend, it makes sense to assume that the electricity demand created by AI could be almost impossible to curtail.

In January, Trump announced his plan for AI dominance, backed by more than $500bn in investment for an artificial intelligence super-project known as Stargate. While it’s uncertain whether this project will ever come to fruition, according to a report from the International Energy Agency this week, the electricity consumption of the world’s data centres could double in the next 5 years alone.

In the US, the electricity requirements of data centres could surpass those of all other energy-intensive industries – including steel, cement, and chemicals – by 2030. On a per capita basis, it is estimated that data centres in the US could consume as much electricity per person as it would take to drive an electric car from Los Angeles to New York and back.

But these early estimates are unreliable, partly because no one really knows how AI will play out – especially in terms of the innovations that could be lurking just around the corner. And innovation is the norm in this industry. As Conal Campbell, senior policy lead at Ireland’s EirGrid, noted this week: “Between 2015 and 2022, global internet traffic increased by 600 per cent, but energy used by data centres grew by less than 70 per cent thanks to technological improvements.”

While America’s tech giants are willing to pour billions into maximising the AI industry as quickly as possible, their competitors in China and India have already highlighted potential step-change improvements in efficiency. Initial testing of China’s DeepSeek indicated that simple queries could be solved using between 50 and 75 per cent less energy than is needed for ChatGPT’s solutions, using more advanced and energy-intensive processors.

In light of the growing energy demand, many tech companies are sounding the alarm and calling for governments to build new nuclear reactors. Microsoft has already teamed up with Constellation Energy to restart a nuclear reactor at the Three Mile Island nuclear plant in Pennsylvania, and other companies are certain to make significant investments to secure their own power supply. But the line between truth and investor speculation is hard to see.

Understanding just how world-shifting the AI industry will be is one of the most critical issues of our age, and the demands on the power sector that will accompany its growth will play a critical role in shaping the way the world’s energy is produced over the coming decade.

In some coal-dependent states, coal power may help to shore up the short-term growth in energy demand driven by new data centres, but it’s difficult to see how the cost efficiency of solar and wind can be ignored in the longer term. Perhaps that’s why Trump needed an executive order to bring back the coal sector this week, because the free market – and the country’s leading energy agency – had already signalled its decline.

Trump’s executive order may act as a double espresso for the industry’s short-term future, but it’s hard to see that energy high lasting much longer.