AI may have finally tipped the scales in favour of nuclear energy again

Digital giants, Amazon, Google and Microsoft are all exploring the use of nuclear energy. Indeed, nuclear power, which has been under a PR cloud for decades, may make a big comeback as a direct consequence of the advent of artificial intelligence (AI) and, to some extent, cryptocurrencies. 
 

Google has partnered with Kairos Power to source energy from small nuclear plants to power its AI-driven data centres. Microsoft has a deal with Constellation to use energy from the Three Mile Island plant, which was the site of a major nuclear accident. Amazon, too, has agreements to invest in small modular reactors (SMRs) to meet its future energy needs.
 

The sophisticated manipulation of energy is inextricably linked to the sophistication of civilisations. The Egyptians used water-wheels and hydraulics to build their pyramids. The Age of Enlightenment was driven by steam. Electricity added many productive hours to the day and enabled innovations like the telegraph. 
 

Power demand rises with economic growth, and even as AI promises to push the global economy into an upcycle, massive amounts of power will be needed for the data centres that are its lifeblood. Cryptocurrency mining also adds to power demand. 

The spectre of climate change associated with high carbon footprints makes thermal power unattractive. Hence, the renewed interest in nuclear, which offers the advantages of thermal without the same carbon footprint. 
 

Among renewables, solar, wind and water are intermittent sources, which require storage solutions, and power evacuation from remote locations (especially for offshore wind). Geothermal and tidal power are location-specific, while green hydrogen remains unproven and is currently very expensive.
 

Thermal is a reliable 24x7 source but polluting if generated from fossil fuels like coal, gas, biogas, or naphtha. Nuclear is a cleaner version of thermal. A normal thermal plant burns coal to heat water and produce steam that turns turbines. A nuclear plant uses nuclear reactions to heat water to steam.  
 

A normal thermal plant has a big carbon footprint. Nuclear plants produce radiation, but they are zero-carbon emission. Also nuclear energy in theory is very efficient, since tiny amounts of fissile material can generate lots of power. If you shield and contain radiation, you have a cheap and clean long-term source of reliable 24x7 power. Naturally, hyperscale data centres and cloud service providers, which all need 24x7 stable and reliable power are interested. 
 

The issues with nuclear can be discussed in broad brushstrokes. Spent fuel must be disposed of, or reprocessed. Spent fuel can be lethal for centuries. If it’s to be disposed of, you need shielded landfills, which must remain secure for hundreds of years to prevent any breach.
 

If fuel is reprocessed, it can be reused. But the reprocessing turns it into weapons-grade fuel, which means it can also be used to make big bangs. This had led to many international safeguards on reprocessing, (and the potential threat of Israeli military intervention as has occurred in Iraq, Syria and Iran). Eventually, a nuclear plant needs to be shut down and that’s a vast expense as well as being tricky from the shielding aspect. Levelised nuclear power is cheap to generate but not so cheap if you factor in the costs of mothballing plants and fuel disposal. 
 

Three major accidents led to public opinion going anti-nuclear. Chernobyl 1986 was the worst, requiring the sequestering of 900 square kms of land around the plant. The Fukushima disaster of 2011 was triggered by a tsunami. The Three Mile Island (TMI) incident was a meltdown of TMI-unit 2 (Microsoft has asked Constellation to reopen TIMI unit-1, which ran safely for decades). Germany shut down its nuclear plants after Fukushima and Japan cut back. France is the only major EU economy that still uses nuclear power at scale (and reprocesses fuel). 
 

Nuclear design has improved a lot since those three incidents at older plants. But an accident is always a non-zero probability at any power plant. Adopting nuclear involves offsetting the admittedly low probability of an accident, which could be horrific if it does occur, versus the high probability that you can generate cheap, clean power for many decades. This calculation has been discussed many times with indeterminate results. AI may have finally tipped the scales in favour of going nuclear again.