Welcome to the State of AI – a collaboration between MIT Technology Review and Financial Times. Writers from both publications are going to debate a different aspect of the AI revolution that is reshaping the global power every Monday over the next six week.
Casey Crownhart is a senior energy reporter at MIT Technology Review. Pilita Clifton, FT columnist and Casey Crownhart discuss how China’s rapid build-out of renewables could allow it to leapfrog ahead in AI.
Casey Crownhart writes:
Energy is the greatest barrier to AI progress. Not money. This is especially true in the US where there are massive data centres waiting to be brought online. It doesn’t appear that the nation will have the infrastructure or power to support them.
This wasn’t the case in all cases. Data centers could offset the increased demand for about 10 years before 2020 by improving efficiency. The US electricity demand has risen, and efficiency improvements are not keeping up. The strain on power grids is increasing as fewer new capacity comes online. Electricity costs are rising for those who live near data centres.
The US must learn from other countries about energy abundance if it wants AI to be able to fulfill big promises, without driving up electricity prices for us. Look at China.
China will have installed more than 6 times as much new capacity in power generation in China by 2024 than the US.
China generates a large amount of electricity using coal. However, this is a decreasing share. Instead, China is focusing on solar, wind and nuclear power, as well as gas, at record-breaking rates.
In the meantime, US is focusing on revitalizing their ailing coal industries. Polluting coal-fired plants, and their operation, are expensive. The US’s aging plants are less reliable, producing electricity only 42% of time, as opposed to 61% in 2014.
This is not an ideal situation. If the US doesn’t do something about it, they risk being consumers rather than innovators of energy tech and AI. China already earns more money from renewable energy exports than does the US from its oil and gas.
It would be a good idea to build and permit new renewable energy plants, as they are currently the fastest and cheapest way to get them online. Wind and solar power are not popular with the administration. The obvious choice is natural gas, but there are some concerns regarding the delays in key equipment.
Data centers could be made more flexible. They could come up with new AI infrastructure without having to build any additional energy infrastructure if they decided not to use electricity during stressful times.
According to a study by Duke University, if data centres agreed to reduce their energy consumption for just 0.25% (roughly 22, hours per year), they could supply power equivalent of 76 GW. This is like adding 5% to the grid without having to construct anything.
Flexibility would not be sufficient to meet the surge in demand for AI electricity. Pilita, what do you think? What could the US do to get out of this energy crisis? What else should we be considering when AI is concerned with energy usage?
Pilita Clark replies:
I am in agreement. I agree. We need to see more agreements like the ones that give data centers cheaper electricity in exchange for allowing power utilities to use their backup generators. The need for more power plants is reduced by both, and this makes sense no matter how much AI uses.
It is important that countries around the globe understand how much energy AI will consume.
The forecasts of what will be needed in data centers within five years range from less than double the current rates up to four times more.
It’s partly due to the lack of data available about AI’s energy requirements. We don’t even know what the future efficiency of these systems is going to be. Last year, the US chip designer Nvidia announced that their specialized chips were 45,000 times energy-efficient over the past eight years.
We have also been wrong before about the energy requirements of tech. At the height of the dot-com boom in 1999, it was erroneously claimed that the internet would need half the US’s electricity within a decade–necessitating a lot more coal power.
Some countries have already felt the strain. Data centers in Ireland consume so much electricity that they have restricted new connections around Dublin.
Several regulators have proposed new regulations that would force tech companies to produce enough electricity to meet their demands. Such efforts should continue. Also, I hope AI helps increase power availability and accelerate the global energy shift needed to fight climate change. OpenAI’s Sam Altman stated in 2023, “once we develop a super-intelligence that is really strong, it will be easy to combat climate change.”
So far, the evidence isn’t promising. This is especially true in the US where projects to promote renewable energy are being cut. The US could end up as an anomaly in a global market where renewables, which are becoming cheaper and more efficient, accounted for over 90% of the new capacity that was added last year.
Europe aims to run one of the largest data centers in Europe primarily with batteries and renewable energy. China is the leader in green energy.
In the 20th century, fossil-fuel rich countries dominated the world. The US wants to extend this reign. China may be the first green electric state in history. It will be a significant chapter in the history of economics, technology, and geopolitics if it can do this while winning an AI race that has been dominated by the US.
Casey Crownhart replies:
Your skepticism about tech executives who claim that AI can be an important tool in the fight against climate change is well-founded. Fair enough, AI progresses rapidly. We don’t need to wait around for technology that makes big promises but has no evidence to support them.
Experts say that AI could help in planning, and perhaps even with operating the grid. However, these experiments are experimental.
In the meantime, a large part of the globe is making progress in transitioning from older, less-green forms of energy to more modern, cleaner ones. It remains to be determined how this will impact the AI boom. AI will change our grid, our world and it is important to understand the implications.
Continue reading
MIT Technology Review Reporters did the math to determine the amount of energy required for an AI query.
AI energy requirements still have a few positive aspects.
You can also find out more about the following: The FTThe team at’s Visual Data Team takes a look Inside the race to AI capability.
Globalization is a global phenomenon. The FT Reporters ask if data centres can be truly green.