“Are you a nuclear engineer? Welcome to Microsoft!” For several months, the professional network LinkedIn has regularly seen job offers from the Seattle firm, which is in the process of building a real nuclear engineering team. Director of nuclear technologies, director of “acceleration of nuclear development”, this team – very feminine: hats off, Microsoft! – has a clear roadmap, guaranteeing “the technical feasibility and optimal integration of SMR systems and microreactors”.
A few years ago, when a digital giant talked about energy, it was generally about renewable energies. Moreover, the sector remains a big investor in this area, decarbonization requires. With the switch to nuclear energy, something else is at stake, beyond the environmental subject: Gafam no longer only want to be green, they also want to secure gigantic quantities of electricity, permanently available.
It’s a known fact: digital is an electro-intensive sector. The International Energy Agency has just recalled this in its global electricity report: electricity consumption by data centers, artificial intelligence (AI) and the cryptocurrency sector could double by 2026. After global consumption estimated at 460 terawatt hours in 2022, total electricity consumption from data centers alone could reach more than 1,000 terawatt hours in 2026, a demand roughly equivalent to the electricity consumption of… Japan. From McKinsey to MIT, the estimates for 2030 are dizzying: in the United States, data center consumption would double; globally, they would absorb up to 21% of the electricity supply.
One reservation: in the past, we have been very wrong about the consumption projections of these famous data centers. Between 2006 and 2018, their energy consumption only increased by 6%, while computing power and storage capacity increased respectively by 6 and 25. The reason? Efficiency gains. Any industry worthy of its name is in a constant process of optimizing the use of its resources – this too.
A crazy demand for energy
The fact remains that the emergence of artificial intelligence is reviving the most crazy predictions about the coming explosion in energy demand in the sector. By 2030, AI could represent 3 or 4% of global electricity demand. Already, it absorbs 10 to 15% of Google’s energy consumption, or 2.3 terawatt hours per year. One more fantasy around this revolution? Perhaps, but the fact is that so-called learning AIs are students who are data and storage hungry: their education costs a lot of electricity.
In this context, micro and small modular reactors can only appeal to the sector. For the moment, these nuclear installations are still a technological-industrial bet: will they be there in time to respond to the explosion in electricity demand, when the most optimistic – or the most daring – industrialists? – are counting on 2030 to see the top seeds emerge? The stakes are high. The state of Wisconsin is considering delaying the shutdown of coal-fired power plants to accommodate a new data center. He is not the only one.
An idea to reduce this risk: what if we trained artificial intelligence to work on their own energy efficiency? This is the goal of the researchers at the Human Brain Project, who have found the best teacher in nature for this purpose: the human brain, the most efficient energy plant in the world! Our brain uses about 20 watts to operate, “which is equivalent to the power consumption of your single computer monitor in sleep mode. On this tight budget, 80 to 100 billion neurons are capable of performing billions of tasks. “operations, which would require the power of a small hydroelectric plant if carried out artificially”. Our secret ? Evolution. The conclusion is obvious: long live us!
Cécile Maisonneuve is founder of Decysive and advisor to the Energy and Climate center at Ifri.
.