Tech Giants’ AI Surge Straining Energy Grids, Competing with Bitcoin’s Power Consumption
The rapid advancement of artificial intelligence (AI) systems by tech giants is placing unprecedented strain on global electricity grids. Surprisingly, the energy demand from this AI boom is beginning to rival the substantial power consumption of Bitcoin. However, the cryptocurrency has taught us valuable lessons on how to mitigate the impact.
Several major technology companies, including Amazon.com, Alphabet Inc, Microsoft Corp, OpenAI, Meta Platforms Inc, Baidu Inc, Tencent Holdings Ltd, and Alibaba Group Holding Ltd, are aggressively developing and deploying generative AI products. The soaring interest in these AI systems is amplifying the pressure on energy grids worldwide and threatening to match Bitcoin’s energy consumption.
Nvidia Corp, the prominent US chipmaker, has experienced a doubling of data-center revenue in the last quarter due to the growing demand for generative applications like ChatGPT. Nvidia’s processors, although essential in this AI revolution, come at a significant cost in terms of both price and energy efficiency. For instance, the GH200 Grace Hopper Superchip, their latest flagship, draws a staggering 1,000 watts of power, equivalent to a portable heater.
While most customers may opt for less advanced chips, they often purchase them in bulk to create massive AI servers. This is where the real hunger for electricity arises. A study published last year assessed the energy consumption required to train a single large-language model used for generating text in multiple languages.
According to the study’s authors, the startup HuggingFace utilized 176 billion parameters and 1.6 terabytes of data in their BLOOM project. This undertaking took a cluster of 384 Nvidia A100 graphical processors (GPUs) over 118 days to complete. The extensive use of so many GPUs for such an extended period likely generated approximately 24.7 metric tons (54,000 pounds) of carbon dioxide. However, if we account for the network connections and idle time of the entire system, the true cost doubles to 50.5 tons.
It’s crucial to note that training a model merely marks the beginning of the AI journey. Amazon.com Inc estimates that approximately 90 percent of the expense associated with running artificial intelligence arises during the subsequent phase when users make queries to get results. This inferencing stage, such as asking ChatGPT for specific information like chocolate-cake recipes, requires a significant amount of energy. Although challenging to calculate precisely, the energy expenditure for inferencing is believed to be roughly ten times higher than that required for training. In essence, it accounts for around 500 tons of CO2 emissions. Astonishingly, a single generative AI query may have a carbon footprint four times larger than a typical Google search.
Notably, Bitcoin’s design inherently involves extensive number crunching, which explains the global rollout of semiconductors and servers in the quest for mining the virtual currency. According to an ongoing study by the University of Cambridge, Bitcoin is responsible for 72.5 million tons of carbon dioxide emissions. However, if all Bitcoin mines were powered by hydroelectricity, this figure could drop as low as 3 million tons. When compared to the wastefulness of cryptocurrency, the 500 tons of carbon dioxide emissions resulting from a single round of AI training and deployment may seem insignificant. Nonetheless, it is still equivalent to driving one million miles in a gasoline-powered car or taking 500 flights from New York to Frankfurt.
The AI industry is still in its early stages, and at least a dozen major technology companies are eagerly striving to develop and implement generative AI products. With fierce competition among these companies, they continue to purchase power-hungry processors to analyze ever-increasing amounts of data. Moreover, once the analysis is complete, they will race against each other to provide consumers with college essays, deepfake videos, and synthetic versions of popular music, such as Pink Floyd.
Adding to the complexity is the fact that most AI training is currently powered by fossil fuels. These server farms have emerged rapidly in existing locations, often far away from hydroelectric dams or solar power arrays. However, due to the issue of network latency when responding to internet requests, they must be situated near the end user instead of thousands of miles away.
Fortunately, the AI industry can learn valuable lessons from Bitcoin. Power-hungry crypto mines have taken advantage of cold climates rich in renewable energy, such as Iceland with its arctic air and abundant thermal energy. China, too, may repurpose the numerous hydroelectric power stations that once attracted mining rigs to meet their rising power needs. Although foreign AI providers may not have access to such resources, Chinese technology giants recognize the value of having them nearby.
Furthermore, unlike Bitcoin, generative AI offers tangible value to the world. While the cryptocurrency attracted speculators and billions of dollars in investment without adding significant value, AI technology drives innovation and offers practical applications. Just ask ChatGPT.
In conclusion, the surge in AI development by tech giants is placing immense strain on global electricity grids, posing a challenge comparable to the energy consumption of Bitcoin. As the demand for AI systems continues to intensify, it becomes crucial to find sustainable solutions and adopt renewable energy sources. The lessons learned from Bitcoin’s approach to energy efficiency can guide the AI industry toward a more responsible and eco-friendly future.
