AI and Bitcoin Mining: A Comparative Analysis of Energy Consumption

Recent studies suggest that the energy demands of artificial intelligence (AI) systems may rival or even surpass those of Bitcoin mining.

This shift may seem favorable for the Bitcoin mining sector, frequently scrutinized for its energy usage. However, a report from the Bitcoin Policy Institute highlights that AI is emerging as a strong competitor for electrical resources and computing hardware.

The substantial financial resources of AI companies enable them to outbid Bitcoin miners for electricity. With AI generating up to 25 times more revenue per kilowatt hour (kWh) than Bitcoin, some miners are either incorporating AI processes into their data centers or converting entirely to AI operations.

“This trend is likely to persist as long as the revenue per megawatt-hour remains higher for AI compared to Bitcoin,” stated Margot Paez, a researcher at the Bitcoin Policy Institute.

Energy Consumption Comparison

Although the AI sector is still expanding, its energy requirements are significant. According to Goldman Sachs, a single request to ChatGPT can consume nearly ten times the energy of a standard Google search. Additionally, MIT Technology Review found that generating an image through AI can consume as much power as charging a smartphone.

Bitcoin mining has faced regulatory scrutiny due to its high energy consumption, leading to potential bans in Europe and moratoriums in New York State. The Bitcoin Policy Institute reported that US Bitcoin mining operations utilize about 121.13 terawatt hours (TWh) annually, while AI may consume between 20 and 125 TWh in 2023, with AI usage projected to reach 169 TWh by 2024 — potentially outpacing Bitcoin’s growth.

• US annual Bitcoin mining energy use: 121.13 TWh
• AI projected energy consumption by 2024: 169 TWh
• AI estimated energy usage in 2027: 240 TWh
• Bitcoin mining energy use by 2027: 160 TWh

The Water Footprint of AI and Bitcoin

Data centers that host AI models also demand substantial water resources for cooling systems to maintain operational efficiency. Research from Shaolei Ren of the University of California Riverside indicates that asking between five and 50 queries to ChatGPT requires approximately 500 milliliters of water.

In contrast, annual water use for Bitcoin mining in the United States is estimated between 93 and 120 gigaliters, with each transaction consuming enough water to fill a small swimming pool, although there are debates about the accuracy of these estimates.

The Financial Landscape: AI vs. Bitcoin Mining

The profit margins associated with AI computing are significantly higher than those of Bitcoin mining. Mining revenues range from $0.17 to $0.20 per kWh, while revenues from AI capabilities, utilizing specialized hardware, can range between $3 to $5 per kWh, representing a 17 to 25-fold difference.

Transitioning from Bitcoin mining to AI isn’t a straightforward process. According to Bitcoin miner and crypto asset adviser Anibal Garrido, the specialized hardware typically used for Bitcoin mining, known as application-specific integrated circuits (ASIC), is solely designed for hashing functions and cannot be reassigned for AI tasks.

Flexibility and Energy Management

The contrasting nature of energy consumption between AI and Bitcoin mining may influence future energy policies. Bitcoin mining operations exhibit greater adaptability; they can power down or switch on based on electricity prices, allowing access to surplus, wasted, or cost-effective energy sources.

Conversely, AI requires nearly continuous operational uptime, absorbing all available energy irrespective of cost implications, which can lead to the use of less environmentally friendly energy sources. This increasing demand may intensify the use of fossil fuel-driven peaker plants to meet sudden spikes in electricity requirements.

Bitcoin miners have the capability to negotiate power contracts with utility providers, shutting down operations during grid saturation and resuming once conditions stabilize, thus enhancing grid stability and contributing to emissions reduction.

Location Dynamics and Energy Incentives

A key difference between Bitcoin mining and AI operations lies in their geographical requirements. Bitcoin mining can occur in various locations, while AI necessitates proximity to urban centers to ensure low latency for quick response times. This disparity results in AI data centers consuming energy directly from their specific locales, while Bitcoin miners can move to areas with abundant energy, particularly renewable sources.

This ability allows Bitcoin mining operations to capitalize on wasted energy, demonstrating potential for positive contributions to renewable energy adoption by providing steady demand during periods of low consumption, thereby supporting financial stability for new energy initiatives.

The Path Forward for AI Efficiency

As the demand for AI surges, the technology must increasingly focus on sustainability. Various approaches to enhance energy efficiency are available to AI developers, such as optimizing existing models, deploying smaller models for focused tasks, and utilizing cloud computing to minimize overall energy expenditure.

Advancements in hardware technology, particularly from companies like Nvidia, are critical in enhancing performance while lowering energy consumption, promoting a more sustainable future for AI. However, as noted by analysts, the trend towards creating larger AI models could counteract efforts to improve efficiency, potentially leading to increased energy demands.

Trend analysis indicates that with greater efficiency in AI technology, there remains the risk that developers may continue to pursue larger models, perpetuating a cycle of escalating resource requirements.