AI energy consumption is an increasingly prominent issue, for example, large AI models need to consume a lot of computing resources during the training phase. According to the 2023 Artificial Intelligence Index Report released by the Stanford Institute for Artificial Intelligence, GPT-3 consumes up to 1,287 megawatt hours of electricity for a single training, which is equivalent to emitting 552 tons of carbon dioxide, which is roughly equivalent to the carbon emissions of 126 households in a year. In addition, AI models also consume a lot of power during the inference phase. A single user request may consume 2.96 watt-hours of power, which is significantly higher than the power consumption of a traditional internet search.
As the widespread adoption of AI technology has led to an increase in the number and size of data centers, so has the demand for electricity. It is predicted that by 2030, AI data centers alone are expected to add about 323 terawatt hours of electricity demand in the United States, which is about six times the current annual electricity consumption in New York City.
Globally, electricity consumption in the data center, AI, and cryptocurrency industries is expected to double in the next few years, which is significantly faster than the average growth rate of global electricity demand.
"The end of AI is energy", this point of view has caused heated discussions in the industry. As the main carrier of computing power, the power consumption and energy supply methods of data centers are changing. For data center enterprises, how to reduce energy consumption and improve energy efficiency is an eternal issue. In the face of the energy demand of AI, photovoltaic and energy storage technology have become the solution.
Figure: Is photovoltaics the end of AI?
As a clean and renewable form of energy, photovoltaic power generation has great potential for development. Energy storage technology provides a necessary supplement to photovoltaic power generation, and by storing excess electrical energy, it can release energy when the sun is insufficient to ensure the stable operation of the power gridThe combination of energy storage technology and photovoltaic technology can further improve energy efficiency. As an efficient and high-density energy source, nuclear energy can also provide stable power support for AI. Nuclear energy, especially fusion technology, is seen as a key technology to solve the energy crisis and climate change due to its almost unlimited fuel supply and environmentally friendly nature. Globally, progress in the commercialization of nuclear fusion is accelerating, and if successful, it will provide humanity with almost unlimited access to clean energy.
While energy technologies such as photovoltaics, energy storage, and nuclear energy can provide important energy support for AI, the energy needs of AI are complex and diverse. Therefore, it is necessary to adopt a comprehensive solution that combines multiple energy technologies to meet the energy needs of AI in different scenarios.
For example, in critical facilities such as data centers, photovoltaic and energy storage technologies can be adopted to reduce dependence on traditional energy sources; In areas where energy supply is unstable, efficient energy technologies such as nuclear power can be used to ensure the stable operation of AI.
Tech giants have realized the close connection between AI and energy, and have begun to actively deploy it. Companies such as Microsoft, Google, and Amazon are not only looking for nuclear energy as an energy solution for data centers, but also advancing nuclear energy technology through investment and partnerships. Amazon, for example, has partnered with Talen Energy to ensure that data centers are guaranteed to receive power around the clock.
