It’s hard to separate the reality from the hype around Artificial Intelligence (AI). Will it be a game-changer, affecting the world economy, employment and our way of life? Whatever the answers to that, AI is already having a huge impact on energy consumption in developed economies, particularly the US.
AI models need a lot of computing power and they are driving investment in massive data centres. While, up to recently, Ireland had one of the biggest concentrations of data centres in the world, AI means the focus is moving elsewhere, mainly to the US.
As data centres become dramatically larger, existing centres could rapidly become obsolete or redundant. In future, when an Irish data centre needs to replace equipment, it may be cheaper for the owners to buy in capacity from a super-centre than to invest in new chips and the water-cooling capacity needed to operate them.
That possibility of future redundancy means that Irish data centres should pay for all their lifetime infrastructure costs up front.
It costs a fortune to feed the AI monster, but is it worth it?
Nuclear generators, hydrogen and battery storage are options we should explore to generate clean electricity
Growth in Irish living standards is likely to disappoint over the next 20 years
Spending on infrastructure makes sense, cutting VAT on hospitality is madness
The computer chips needed to run the latest AI models require a dramatic increase in electrical power, concentrated into a tight space. This high-energy usage means AI data centres need large amounts of water to keep them cool.
So, the energy and water demands of data-hungry AI are dramatically greater than Ireland’s current data centres have capacity for. The Financial Times reports that Meta is building a centre in Louisiana that will require 2 gigawatts of electricity generation capacity. That’s enough to power every household in Ireland twice over.
Why is Ireland not considered a truly rich country?
Even in US terms, that’s pretty big. By 2028, data centres could account for 12 per cent of all electricity generated in the US, and between 3 and 4 per cent to US greenhouse gas emissions. Ramping up energy demand for data centres is coming at a time when electricity prices are already rising for households and businesses because of pressures on generation capacity.
The US will find it extremely difficult, expensive and carbon-intensive to deliver all the electrical energy that this wave of AI data centres requires.
[ Forget the futuristic dystopias: AI is changing the world right nowOpens in new window ]
The scale of investment and energy consumption involved could pose major challenges for the US economy. This year, the investment for AI will amount to over 2 per cent of US gross domestic product (GDP). Management consulting firm McKinsey estimates that by 2030, the cumulative cost of the investment needed to make AI work will be over 20 per cent of US GDP. This raises a key question – will the investment pay off?
To make money, the income generated from this investment will first have to pay for the capital cost – all the computers, extra electricity generation capacity and transmission equipment involved.
It will also have to pay for the energy consumed each year. Investors will want a return on the money that have put in. Together, that means that by 2030, the AI models will need to generate income of 3 per cent to 5 per cent of US GDP.
Thus, if the AI models are worth what they are costing, they would have to raise US productivity by well over 3 per cent by 2030. As of now, this seems difficult to achieve. Much of AI is currently trained on the mixture of fact and fiction found online, making it unreliable without skilled human oversight.
Alternatively, to make a return, AI models would need to replace well over 5 per cent of all jobs in the US by 2030, which is even more unlikely.
[ ChatGPT-5: Maybe Sam Altman should cool the jets on new AI iterationsOpens in new window ]
For investors, even if their AI models make a huge difference to the US economy, there is the question of how they will be paid for their services. Today, we pay for our searches using Google or Edge by being exposed to ads. Will college students pay substantial sums to have their assignments written by AI in the future? Will businesses spend large amounts on buying in the wisdom of AI?
It is likely that there will be one or two winners, making big profits from their investment, while a lot of the investors currently rushing into AI will lose their money. This means that the returns investors hope to make from their investments need to be high, to reflect the riskiness of their bet.
The new AI modelling capacity undoubtedly brings major advantages for the modern economy. However, there is a danger that these benefits are being exaggerated in the rush to win the AI race. It takes far more energy to train an AI model than to stream videos. For all their sophistication, AI models lack the judgment capacity that only humans have. Educating our children remains much more energy efficient than relying on computer brains.
