Artificial fires through a lot of tools. Perhaps a more energy-efficient AI is likely to result in more energy being used over the long run, as a conundrum was first discovered in the 1860s.
For most consumers, “large speech models” such as OpenAI’s ChatGPT function like instinctive search engines. However, AI models return information they’ve created from scratch, in contrast to traditional online searches that locate and retrieve data everywhere along a worldwide network of servers. Like powering up a nuclear furnace to use a computer, this designed process is quite wasteful.
One study suggests the AI industry will be consuming somewhere between 85 and 134 terrawatt-hours ( TWh ) of electricity by 2027. That’s the same amount of energy that the Netherlands consumes annually. One well-known researcher predicts that over 20 % of US electricity will be used to power AI data centers ( huge warehouses stuffed with computers ) by 2030.
Big tech companies have often vowed to be significant investors in wind and solar energy. However, most people are developing their own atomic options due to AI’s desire for 24/7 energy. Microsoft also plans to revive , the legendary Three Mile Island , power flower, the image of America’s worst-ever legal radioactive accident.
Despite Google’s ambitious goal of being carbon neutral by 2030, the agency’s AI improvements mean its emissions have climbed 48 % in the past few years. Additionally, each month, the processing power required to train these concepts increases tenfold.
Nevertheless, Chinese start-up DeepSeek claims to have created a fix: a design that fits the effectiveness of established US foes like OpenAI, but at a fraction of the cost and carbon footprint.
An environmental game changer?
DeepSeek has created a powerful open-source, relatively energy-lite model. The company claims it spent just US$ 6 million renting the hardware needed to train its new R1 model, compared with over$ 60 million for Meta’s Llama, which used 11 times the computing resources.
DeepSeek uses a “mixture-of-experts” architecture, a machine-learning method that allows the model to scale up and down depending on the complexity of prompts. The manufacturer claims that its model can train and store more data without using sizable amounts of pricey processor chips.
Following investor concerns that AI companies would reevaluate their energy-intensive data center developments, US chip manufacturing and energy stocks fell. As the world’s largest supplier of specialist AI processors, Nvidia saw its share price fall by$ 589 billion, the biggest one-day loss in Wall Street history.
Paradoxically, as well as upsetting the performance of US tech stocks, improving the energy efficiency of AI platforms could actually worsen the industry’s environmental performance as a whole.
With tech stocks crashing, Microsoft CEO Satya Nadella tried to bring a longer-term perspective:” Jevons paradox strikes again”! he posted on X. ” As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of”.
The Jevons paradox
For more than a century, people have been saying that using less energy isn’t always beneficial to Earth’s resources. In his book” The Coal Question” in 1865, a young Englishman named William Stanley Jevons argued that Britain’s position as an industrial superpower might soon be ended as a result of its rapidly dwindling coal reserves.
But to Jevons, frugality was not the solution. He claimed that” the idealism that the sparse use of fuel is equivalent to a decreased consumption is completely confounds all other ideas. The very contrary is the truth”.
According to Jevons, any increase in resource efficiency generates an increase in long-term resource consumption, rather than a decrease. Higher energy efficiency has the effect of lowering energy’s implicit price, which in turn raises the rate of return and demand.
Jevons gave an example of the British iron industry. If advances in technology enabled a blast furnace to produce iron with less coal, profits would increase and new investment in iron production would be drawn. Price reductions would also encourage higher demand. He concluded:” The greater number of furnaces will more than make up for the diminished]coal ] consumption of each”.
Since the dawn of human civilization, the economist William Nordhaus has used this concept to improve lighting efficiency.
In a paper published in 1998, he came to the conclusion that the typical worker in ancient Babylon might need to work for more than 40 hours to buy enough fuel to generate the same amount of light as a typical lightbulb for an hour. However, an average American would need to produce the same amount of work by 1992.
Throughout time, efficiency gains haven’t reduced the energy we expend on lighting or shrunk our energy consumption. We now, in contrast, produce so much electric light that areas without it have turned into tourist attractions.
Warming and lighting our homes efficiently, driving our cars, mining Bitcoin and, indeed, building AI models are all subject to the same so-called rebound effects identified in the Jevons paradox. And because of this, it will be impossible to guarantee that an energy-use reduction in the overall industry is achieved.
A Sputnik moment
In the 1950s, the US was horrified when the Soviets launched Sputnik, the first space satellite. America spent more money on the space race, not less, as a result of the development of a more effective rival.
DeepSeek is Silicon Valley’s Sputnik moment. In an arms race that is no longer limited to US tech giants, more distributed and powerful models will likely mean more distributed and powerful models.
AI offers superpower status, and the floodgates may now be fully open for the UK and other global competitors, as well as China. What’s for certain is that in the long term, the AI industry’s appetite for energy and other resources is only going to increase.
Peter Howson is assistant professor in international development, Northumbria University, Newcastle
This article was republished from The Conversation under a Creative Commons license. Read the original article.