Since the big AI model represented by ChatGPT burst into flames at the end of 2022, generative AI technology has been expanding from the generation of language, video, and images to the application end of human-computer interaction, AI assistants, intelligent manufacturing, drug development, and other applications, and will also help humans explore more unknown areas in the future. However, the infrastructure end that supports the rapid growth of generative AI, including energy, arithmetic and data, has faced considerable challenges since the beginning, especially the supply of energy.
OpenAI recently released a ‘Blueprint for U.S. AI Infrastructure’ in Washington, D.C., making clear that it plans to work with the new government on AI policy and help it build an AI data center that is expected to consume 5 gigawatts of power (which would be five times larger than the one the company is currently developing). The blueprint also mentions how to build an artificial intelligence (AI) data center, including calling on the US government to expand energy capacity such as nuclear power.
Global Data Centre Capital Expenditure Soars
In addition to OpenAI, US tech giants Microsoft, Google, Amazon, Meta and others have been very active in AI and data centers in recent years, with notable investments. Data shows that Microsoft’s quarterly capital expenditure in the first quarter of this year was $14.9 billion, a record high, and a 50 percent increase over the same period last year.
According to the latest U.S. Census Bureau data, spending on building private data centers in the U.S. has surged to nearly $30 billion a year, more than double what OpenAI will be spending by the end of 2022 when it launches ChatGPT. Today, US companies are spending more of their construction budgets on data centers than in categories that still accounted for a significant amount of capital last year, such as hotels, retail and leisure facilities.
BlackRock and Microsoft have also recently been rumored to be investing $100bn in expanding and building AI data centers, with NVIDIA supporting their partnership. And Amazon is also planning to invest more than $100 billion in data centers over the next decade. NVIDIA has also announced that it will invest $3 trillion in building data centers globally over the next five years.
The investment of these tech giants will have a great driving and demonstration effect on the global AI data center. Currently, North America occupies the largest share, about 40% to 50% of the capital expenditure, mainly concentrated in the United States. Going forward, Asia-Pacific will be the fastest growing region, especially in China and India, where capital expenditure is expected to grow at an average annual rate of more than 10 percent over the next few years. According to market research firms, global data center capex will exceed $400 billion by 2025, benefiting from the further spread of cloud computing and artificial intelligence.
Energy Issues Arising from AI Development
Although generative AI shows great market potential, energy consumption and the resulting environment cannot be ignored. Data centers, as the main place of operation for generative AI, are rapidly increasing their power consumption as a proportion of the global total. Currently, data centers already consume between 1% and 1.5% of global electricity use, and this proportion is expected to rise significantly in the coming years. By 2030, data center electricity demand could account for 4.6% to 9.1% of total US electricity generation.
While technological advances present opportunities to improve energy efficiency, such as reducing energy consumption through algorithmic optimization and more efficient hardware design, overall, the energy demand for generative AI continues to grow rapidly. Some companies are even considering developing small nuclear power plants to ensure a stable energy supply. Data from the International Energy Agency suggests that global data center power demand is expected to exceed 1,000 terawatt hours by 2026, accounting for 2% of global power demand.
Looking at current technology trends, the training and application of AI models have posed new challenges to energy consumption. According to research, the energy consumption for training a large deep learning model is equivalent to the use of a household appliance for an entire year. Therefore, how to reduce the environmental impact of AI arithmetic through green energy is an urgent issue for the current technology industry.
AI Energy Supply Targeting Nuclear Power
Nuclear power, as a clean and efficient energy option, has attracted much attention in recent years. The potential of nuclear energy is gradually being rediscovered in global efforts to combat climate change and the energy crisis. Microsoft’s restart of this nuclear power plant, located in a relatively technologically advanced but resource-poor region, demonstrates the company’s ambition for energy self-sufficiency. At the same time, this strategy also highlights the growing energy demands of cloud computing and AI development, reflecting the urgent need for sustainability in today’s society.
On the corporate front, US tech giants are also looking to domestic power supplies and are targeting the 92 nuclear reactors in operation in the US. Amazon Web Services (AWS), for example, is close to a deal that would see it receive its power supply directly from Constellation Energy, the owner of the largest nuclear power plant in the US. Microsoft also plans to use the next generation of nuclear reactors, known as Small Modular Reactors [SMRs], to support its data centers and AI projects. Bill Gates has even called for increased investment in fusion technology to cope with increased energy demand from generative AI. And OpenAI has plans to build new solar and wind plants and clean up unused nuclear reactors for use, and has even proposed expanding its nuclear power output by utilizing a small reactor partly intended for submarines by the US Navy.
Other Challenges to AI Development
Of course, in addition to power consumption, generative AI also puts a huge strain on water resources and the environment. For example, every time a generative AI such as ChatGPT is used to answer a question, a certain amount of water is consumed to cool the data center. At the same time, people are questioning the safety of an increasing number of nuclear power plants.
Currently, despite the commitment of some tech companies such as Microsoft and Google to become carbon neutral, their greenhouse gas emissions have increased significantly over the past two years due to the rise in AI adoption. As a result, experts are calling on tech companies to be more transparent and take steps to reduce carbon emissions and energy consumption. Undoubtedly, to find a balance between technological innovation and environmental protection in the future, it is also necessary to reduce their ecological footprints by optimizing algorithms, improving hardware design and adopting cleaner energy sources.
