Is electricity and energy storage the ultimate influence on the development of AI?
Recently, NVIDIA CEO Huang Renxun stated in a public speech that the future development of AI will be closely linked to photovoltaic and energy storage. Another tech guru, OpenAI founder Sam Ultraman, also put forward a similar view, believing that the future of AI depends on energy, and society needs more photovoltaic and energy storage.
Daily consumption of 500000 kilowatt hours of electricity, 17000 times that of ordinary households in the United States
The development of AI is advancing rapidly, from text to images and then to videos, with remarkable speed of progress. But behind these developments comes a cost, and the cost of developing AI is the enormous consumption of energy.
Tesla founder Musk has publicly stated that the speed of AI development is unprecedented, with AI computing power increasing tenfold almost every six months. The Moore's Law in integrated circuits only doubles performance every 18 months, which is not worth mentioning compared to the current growth rate of AI.
However, with the rapid growth rate, there is a massive demand for energy. Musk predicts that within two years, the shortage of silicon will turn into a shortage of electricity, which will hinder the development of AI. Musk believes that the current growth of computing power is facing a bottleneck, and the next step is a shortage of transformers, followed by a shortage of electricity. By 2025, there may even be a situation where there is not enough electricity to run all chips.
And AI will become a major consumer of electricity. According to The New Yorker magazine, ChatGPT's daily electricity consumption reaches 500000 kilowatt hours, equivalent to 17000 times the electricity consumption of ordinary households in the United States, in response to approximately 200 million requests from its users.
The International Energy Agency (IEA) previously stated that global electricity demand will accelerate in the next three years, with an expected annual growth of 3.4%. The Uptime Institute in the United States predicts that by 2025, the proportion of artificial intelligence business in global data center electricity consumption will surge from 2% to 10%.
If generative AI continues to develop, its power consumption may increase. Dutch scientist Alex de Fries mentioned in his paper that by 2027, the entire AI industry will consume 85-134 terawatt hours (TWh) of electricity annually, with 1 TWh equivalent to 1 billion kilowatt hours.
For example, the annual power generation of the Three Gorges hydropower station in China has just exceeded 100 billion kilowatt hours, which means that in the future, the power of the Three Gorges Power Station just meets the electricity consumption of the entire AI industry, but the Three Gorges Power Station needs to supply electricity to multiple provinces and regions in central and southern China.
For example, Taiwan will soon raise electricity prices in April, and it is expected that industrial electricity prices will increase by 30%. The average electricity price plan for April will also increase by 10% -15%.
NVIDIA CEO Huang Renxun also pointed out in a public speech that the future development of AI is closely related to photovoltaic and energy storage. He said that in the future, people cannot only think about computing power. If we only consider computers, we will need to consume 14 Earths of energy.
The reason why AI today requires such a large amount of energy is due to the performance improvement of current large models. The optimization of algorithms is certainly important, but it relies more on the improvement of "brute force" computing power, that is, by adding more computing power cards to increase the performance of large models, which is known as "force brick flying".
This also means that the core threshold for future AI competition will no longer be simply computing power innovation, but the level and cost of electricity and energy storage required for massive computing power. Especially as the main energy supply for intelligent computing centers, the stability and sustainability of electricity are crucial for the healthy development of the computing power industry.
With the continuous improvement of the performance of intelligent computing centers, electricity and energy storage will become the key barriers to this industry. How to efficiently utilize electricity, reduce energy consumption, and improve energy storage levels will become an important factor for related enterprises to gain advantages in the future.
From the perspective of energy storage, for example, the development, training, and application of AI require a large amount of electrical energy support. Energy storage systems can effectively store and release electricity as needed, avoiding the impact of grid fluctuations and ensuring the stable operation of AI facilities. At the same time, the computational load during AI training may reach its peak instantly, leading to a surge in electricity demand. Energy storage systems can store electricity during off peak hours, release it during peak hours, alleviate grid pressure, and meet the elastic demand of AI computing for electricity.
Not only that, energy storage technology can also help promote the application of AI in edge computing scenarios, such as unmanned aerial vehicles, autonomous vehicle, Internet of Things devices, and so on. Energy storage devices can ensure that devices can still run AI algorithms efficiently without external power supply.
In terms of energy sources, more and more data centers are adopting renewable energy sources such as wind and solar power. However, renewable energy has intermittent and unstable problems. Energy storage technology can store the excess electricity they generate and use it when there is no sunlight or insufficient wind power, making the AI industry more green and sustainable.
Of course, facing the enormous energy demand brought about by AI is also a challenge for energy storage technology itself. For example, energy storage systems need to have efficient energy density, which means they can store more electricity per unit volume or mass to meet the continuous and high load operation needs of AI systems.
In addition, multiple aspects need to be strengthened, including fast charging and discharging capabilities, long cycle life, high safety and stability, intelligent management, and sustainability. The lithium batteries, solid-state batteries, sodium batteries, and all vanadium flow batteries currently applied and developed in the market are comprehensively considered to meet the needs of the future AI era.
Specifically, for example, for inverters and DC/AC converters in energy storage products, it is necessary to develop power semiconductor devices with higher conversion efficiency (such as new IGBT, SiC MOSFET, GaN and other wide bandgap semiconductor material devices), which can reduce losses in the power conversion process and improve the overall energy utilization efficiency of the energy storage system.
By utilizing advanced semiconductor chips and sensor technology, a highly intelligent energy storage management system is constructed to monitor battery status in real-time, predict remaining lifespan, and optimize charging and discharging strategies through machine learning and AI algorithms, effectively scheduling and allocating power resources.
Recently, the Shenzhen Municipal Bureau of Industry and Information Technology released the "Action Plan for High Quality Development of Computing Infrastructure in Shenzhen (2024-2025)", which has set a clear planning direction for industry development. In coordination with the upgrading of the power grid, the coordinated development of energy storage power stations and intelligent computing centers will be carried out to stabilize power fluctuations and support the development of AI large-scale model technology, which will become the mainstream in the future. It can also be seen that the development of energy storage technology has already reached the upper limit of future AI development.
summary
Energy storage technology is one of the important foundations supporting the healthy development of the AI industry. It can not only solve the large demand for energy in the AI field, but also help it achieve low-carbon and environmental protection goals. And in order to meet the electricity demand brought about by the development of AI, energy storage technology needs to be deeply researched and innovated in terms of safety, economy, system integration, algorithm support, digitization and intelligence, adaptability, technological innovation, as well as policy and market mechanisms. Through these efforts, energy storage technology will be able to better support the development of AI, while promoting the transformation and upgrading of the entire energy industry.