For the nuclear power industry, GAI is both a tool to enhance the performance of nuclear reactors, but as well, a reason for a nuclear industry renaissance. GAI needs lots of power to support the data centres that work with the Large Language Models (LLMs) used to train GAI.
GAI placed five times greater demand on data servers than standard search engines according to Sarah Myers West, Managing Director of the research group, the AI Now Institute. According to an October 2023 article appearing in Scientific American, the demand for AI on servers noted that data centre electricity use to date was accounting for up to 1.5% of global demand. That was then, but GAI capacity adoption trends show significant growth in AI server unit demand with as many as 1.5 million units shipping per year by 2027 from just one vendor, NVIDIA alone. By 2027, the addition of 1.5 million servers running at full capacity will require energy producers to generate 85.4 Terawatt-hours of electricity annually. In the October Scientific American article, it notes that this would be more energy “than what many small countries use in a year.”
Where is the energy being used? There are two phases involved with GAI: training and inference. The latter covers user queries and responses. Both phases are energy intensive with Google’s historic data demand history putting the ratio between training and inference at 40:60. OpenAI’s ChatGPT numbers require less energy for inference but remain energy intensive for training.
There are data centre energy demands that go beyond power to the servers. These facilities need additional energy to remain cool which could add 50% to electricity requirements.
Machine learning is already in use in existing nuclear power plants. AI algorithms provide real-time monitoring and predictive maintenance by sifting through the data being generated and identifying anomalies that human operators might miss. Jeremy Renshaw, an expert on technology innovation at the Electric Power Research Institute, describes how AI is enhancing precision noting that a human operator and inspector “only looks at pertinent data. Instead of searching for the ‘needle in the haystack,’ we remove the haystack.”
AI used in existing nuclear power operations doesn’t replace human analysis. It enhances and speeds up operator responses. Machine learning can spot minor structural problems like cracks in tanks and pipes that are hard to detect when just monitoring gauges.
AI combined with a digital twin that runs in parallel with a real nuclear plant would dramatically enhance power production, and bolster efficiency and equipment performance.
AI combined with robotics can be used to do maintenance in areas of a power plant that would put human technicians at undue risk.
I am old enough to remember the nuclear accident at Three Mile Island that saw a partial meltdown of the reactor core in 1979. It coincided with the release of the movie, The China Syndrome, about a nuclear accident. The combination of the two made the public uneasy about nuclear power technology.
Three Mile Island was the worst nuclear industrial accident before the events at Chornobyl, in the Ukraine (then part of the Soviet Union), and later at Fukushima in Japan.
In the news in the last few weeks, the name Three Mile Island has reemerged because Microsoft wants to use the power from a reopened unit at the nuclear plant to provide electricity to its AI data centres.
Microsoft isn’t alone among the technology giants seeking nuclear energy to power existing and future data centres.
In a world where AI spending is projected to reach US $632 billion by 2028 and $1.81 trillion by 2030, it means energy demand for GAI will see rapid growth as well. GAI spending by 2028 is estimated to reach US $202 billion or 32% of overall AI spending.
The demand for nuclear energy, geothermal, renewable and thermal sites will grow as GAI and AI demand increases. The renewed interest in nuclear with the latest news about Three Mile Island still has to take into consideration the many challenges this highly regulated form of energy production undergoes. New nuclear projects in North America can take decades to be approved and built.
Will the approval process speed up in the future? Despite lobbying by technology companies looking for more power to fulfill energy demands for GAI, untangling the regulatory process because of past problems with this form of energy production could probably use AI and digital twin solutions to ease safety and operational concerns. New fission reactors will also need to address an old and persistent problem that impacts existing operations, the disposing of spent nuclear fuel, a current environmental problem that seeks a permanent solution.
]]>
The research found that the training of a series of AI models emitted almost 284,000 kilograms (626,000 pounds) of carbon dioxide (CO2), five times the average lifetime emissions of an automobile including the CO2 released during manufacturing.
New AI models based on neural network computing require many hours of training and exposure to vast amounts of data. In many cases, the hardware used is top-of-the-line supercomputers or multiprocessor server farms. The energy requirements to power the hardware, plus the manufacturing of the systems themselves, if not sourced from renewables produces net new carbon emissions. This is particularly true when the hardware being used to develop the AI models is sourced from often, energy-intensive in-house data centers, and not from cloud sources which more recently have become greener by developing accompanying wind, solar and battery storage farms.
For the purpose of the study, the authors looked at model training for natural language processing (NLP), teaching AI to understand human language. The training of NLP models has involved mega datasets of scraped pages from the Internet, computationally expensive, and an enormous energy consumer. In studying the four models that represent state-of-the-art NLP performance, they were able to calculate resource usage, and total energy consumed over the model training process. They then converted the energy and resources consumed into CO2 equivalents.
What they found was that the more complex the AI model, the higher the environmental cost in CO2 equivalent emissions, with the most costly model of the four producing a carbon footprint equal to 635 kilograms (1,400 pounds) of CO2, equal to a single fare round-trip flight from New York City to Los Angeles. The CO2 amount described here covers only one modeling effort. So imagine when you are testing a number of options, how quickly the emission equivalent values would rise. For example, in a six-month AI training effort involving 4,789 modeling options, the amount of CO2 emitted equaled 35,800 kilograms (78,000 pounds). Multiply that last statistic by the geometric growth in AI research, and particularly, using large neural networks, and you have a growing and massive carbon footprint issue that needs to be addressed.
In the paper, the authors address this issue by looking at the shared computing resources offered by companies such as AWS, Google, and Microsoft, looking at advantages and disadvantages of doing the research using third-party commercial data farms. What they found out is that in most cases, the economic benefit is not immediately apparent using cloud-based commercial server farms proving to be twice as expensive as relying on home-built resources. And once built, the home-based servers can be repeatedly used for subsequent projects for the cost of energy and operational maintenance.
The authors post for consideration that because of the burgeoning field of AI research, a good national strategy consider cloud-based data farms built by governments for common use and that these facilities be powered from renewable energy sources. Continuing down the current path, the authors conclude is not feasible because the computational resource requirements for advancements in AI will in time become environmentally, energy, and cost prohibitive.
