Training AI requires large amounts of electricity energyAnd that's why Microsoft is looking into the possibilities of using small nuclear reactors for its centers.
A strange news came to us from The Verge, according to which Microsoft is considering the possibility of using next-generation nuclear reactors (small modular reactors = SMR) to power its data centers and artificial intelligence.
Training large language models is an incredibly energy-intensive process that has a huge carbon footprint. Keeping data centers running also requires a large amount of electricity which could generate significant amounts of greenhouse emissions, depending of course on the source of the energy.
And so Microsoft is turning to nuclear to train its AIs. But what is certain is that he will have many other obstacles to overcome. First, a functioning small nuclear power plant should be procured. Next, she'll have to figure out how to get her hands on the highly enriched uranium fuel these small reactors typically require. Finally, it will have to find a way to store nuclear waste for the long term.
In addition to nuclear fission, Microsoft is also investing in nuclear fusion, a far more ambitious endeavor given decades of research, but which has yet to lead to a practical power system. However, the company signed an agreement market of electricity with Helion, a startup fusion founded this year by its CEO OpenAI Sam Altman, with hopes of buying electricity from it by 2028.
The development of digital technology as it seems it is not so simple and painless.