Nuclear Power Can't Save AI
Nuclear Power Can't Save AI
The limits of this technology are starting to hit hard.
AI is getting power-hungry. Estimates suggest OpenAI’s ChatGPT already uses 500,000 kWh of electricity each day! That’s about as much as 17,241 average US households use per day. But that is only the energy ChatGPT uses when operating. As I covered in a previous article, as AIs are trained on ever larger datasets in an attempt to make them perform better, the energy required to train them increases exponentially. As such, if next-gen advanced AIs are to be created, deployed and used profitably and without damaging the environment, the AI industry needs an energy revolution. US Energy Secretary Jennifer Granholm, along with many AI companies, has suggested that AI data centres could have their own nuclear power plants to not only break past this energy barrier but also solve some of the nuclear industry’s biggest problems. However, nuclear-powered AI is sadly a pipe dream, at least for the foreseeable future. Let me explain.
So, why are AI companies, and even the Energy Secretary, drawn to nuclear power? Well, it has carbon emissions that are as low, if not lower, than renewable energy. Moreover, unlike renewable energy, it is on-demand and relatively compact. As such, it is currently the only power source that can deliver the vast amount of power needed to AI data 24/7 without interruptions or excess emissions.
Not only that, but Garnholm has suggested that AI can solve one of nuclear energy’s most significant issues: bureaucracy. You see, there is an insane amount of regulation around atomic power, as you would expect. As such, it can take many, many years to find a suitable location, design suitable plans, approve designs and start building a nuclear power plant. This incredibly expensive paperwork is a massive hurdle to building new nuclear power plants, as it makes them significantly slower and more costly to deploy than any other power source. According to Garnholm, AI could be used to accelerate this process by automating it. As such, a symbiotic relationship between AI and nuclear power can be created, where they both work together to make each other better.
Sadly, there is no evidence that AI can be used to speed up nuclear bureaucracy. Considering AI still stumbles when used in basic legal settings, I highly doubt it can be used to automate or even significantly speed up nuclear bureaucracy in the near future. After all, nuclear power plants are some of the most complex and potentially risky systems we have ever built. There is a reason the regulations around them are super tight and lengthy. But even if it could, nuclear energy simply can’t be AI’s saviour.
Firstly, nuclear takes far too long. Take the 3.2 GW Sizewell C nuclear plant being built in the UK. Despite being built on a reapproved site, it was initially set to take 12 years to build and is set to blow way past that timeframe. This is very typical of modern nuclear power plants, as many take literal decades to build. The AI industry simply can’t wait that long for their energy. The AI industry is rapidly developing, and if any company stagnates, it risks being made obsolete overnight.
But there is also a cost issue. Sizewell C will cost a mind-blowing £30 billion ($36 billion), and again, that is typical of the nuclear industry. This price tag means that AI companies can’t fund their own nuclear power plants, as they simply don’t have those billions lying around, and even if they did, they would definitely need to see a return on such expenditure within a year or two, not in over a decade. This is also why nuclear power is one of Earth’s most expensive forms of energy. Coal, one of the cheapest fossil fuel energies, costs around $109 per MWh, while nuclear power sits at $155 MWh, making it 42% more expensive! Considering how much energy ChatGPT is already using, this price disparity would be enough to bankrupt any AI company that tried to purely power itself with nuclear energy.
But it isn’t like the CEOs of these AI companies don’t know this. OpenAI CEO Sam Altman has gone on record saying that an energy breakthrough, like nuclear fusion, is needed to make advanced AI viable. You see, he thinks that next-gen nuclear power plants like SMRs (Small Modular Reactors) and nuclear fusion energy will have all of the benefits of current nuclear power but will be way faster to deploy, far cheaper to build, and produce energy at a significantly lower cost.
But there is a problem here. Fusion power is at least 20 years away, if not more, as we have yet to start to produce a net gain in energy from a fusion reactor. SMRs and other cheaper, faster-deploy nuclear reactor designs are coming, but they won’t be commercially available until the mid-2030s. Even then, these will be the first few, and they are set to be only marginally faster and cheaper than current nuclear power. It will take decades for the next-gen nuclear industry to establish itself, solidify regulations around their new technology, scale manufacturing, and bring costs down.
Simply put, nuclear energy can’t provide power soon enough, or at a low enough cost, to solve AI’s ballooning energy problem. Quite frankly, nothing can. Renewables can’t expand quickly enough, geothermal energy isn’t that readily available, and fossil fuels destroy the planet. The only way for AI to break past this energy barrier is to become far more energy efficient.
Thanks for reading! Content like this doesn’t happen without your support. So, if you want to see more like this, don’t forget to Subscribe and follow me on BlueSky or X and help get the word out by hitting the share button below.
Sources: NS, TNR, The Conversation, Will Lockett, Will Lockett, TelcoDR, Will Lockett