AI vs Bitcoin mining: Which consumes more energy?
If AI and crypto are the technological behemoths of our era, which one has a bigger appetite for energy?
Illustration: Nadia Méndez
It’s no news that artificial intelligence (AI), the technology of our time, consumes enormous amounts of energy. Large AI models like GPT run on massive data centers as energy-hungry machines that perform intricate computations. On the other hand, crypto mining, especially Bitcoin mining, has notoriously consumed “the same amount of energy as certain countries”, or so we’re told.
AI models run on servers equipped with multiple GPUs (or TPUs for Google), which are usually supplied by Nvidia. Each AI server consumes two kilowatts continuously, according to Shaolei Ren, a University of California Riverside researcher whose studies on modern AI models are frequently cited in the media.
But how many servers are needed for one large AI model like the GPT? SemiAnalysis, a semiconductor analysis firm, estimates that using ChatGPT in every Google search would require more than 500,000 of Nvidia’s A100 HGX servers. Based on this figure, it would result in daily electricity consumption of 80 gigawatt hours (GWh) and annual consumption of 29.2 terawatt hours (TWh), according to Alex de Vries, another frequently-cited researcher from the Vrije Universiteit Amsterdam School of Business and Economics.
From Google’s recently launched ChatGPT competitor Gemini to OpenAI’s secret “Q*” project, tech giants are going all-in on AI, and power demand is expected to increase exponentially. de Vries points out that AI servers could use between 85 to 134 Twh annually by 2027 in a middle-ground scenario. That figure is similar to the amount of power that Norway and Sweden each use in a year and is 0.5 percent of the current global power demand.
“I think that researchers should be much more mindful of the consequences of AI operations,” says Professor Martin Takáč, Deputy Department Chair of Machine Learning and Associate Professor of Machine Learning at Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI). At MBZUAI, researchers are required to take the carbon footprint of AI models into account when designing the models.
Meanwhile, the digital gold rush also comes at a great cost. Crypto mining is leading the charge. The University of Cambridge Centre for Alternative Finance (CCAF) offers the most widely accepted estimates on the energy consumption of Bitcoin and Ethereum. The Cambridge Bitcoin Electricity Consumption Index (CBECI) estimates that Bitcoin consumes 112.31 TWh annually, which equals nearly the entire annual electricity consumption of the Netherlands.
Bitcoin mining consumes so much energy because of its consensus mechanism and its competitive nature. Bitcoins are created as rewards when a miner adds a block to the chain by solving a complex mathematical puzzle using computer processing power. That mechanism is called proof of work (POW). The very system itself is designed to be demanding.
But not all cryptocurrencies consume so much energy. An alternative consensus mechanism, used in creating Ethereum, is called proof of stake (PoS). The block creator is selected by an algorithm based on the user’s stake. This means the more cryptocurrency you own, the more likely you will be chosen to create a block. Without a demand for massive computer processing power, Ethereum consumes around 7.08 GWh annually, a drop in the ocean compared to Bitcoin.
On the other hand, water consumption is also skyrocketing as the servers at both AI and Bitcoin mining data centers need water to keep cool.
In its latest environmental report, Microsoft’s global water consumption grew 34 percent from 2021 to 2022 to nearly 6.4 gigaliters (GL), a sharp increase compared to previous years. That much water could fill more than 2,500 Olympic-sized swimming pools. Similarly, Google reports a 20 percent growth in water usage in the same period.
Ren’s paper estimates that every time a user asks ChatGPT five to 50 questions, the chatbot gulps up 500 milliliters of water. The figure becomes astronomical if you consider the millions of users asking ChatGPT (definitely more than five) questions daily. According to the paper, global AI data centers may use 4,200 to 6,600 GL of water by 2027, more than four to six times Denmark’s total annual water withdrawal.
“In total, it is estimated that Bitcoin consumed more than 1600 GL of water in 2021 and is likely consuming more than 2200 GL of water in 2023,” de Vries, who extensively studies the resources consumed by crypto mining, said in an interview with Greenpeace.
In a famous case earlier this year, environmental groups in the US sued Greenidge Generation, a Bitcoin mining and power-generation company, for allegedly discharging large volumes of hot water into Seneca Lake in New York. Residents around the lake raised concerns that the warm water discharges were heating it beyond regulated standards. However, the lawsuit was later dismissed.
Although AI and Bitcoin mining data centers consume similar amounts of energy, it seems that Bitcoin mining has a much higher dependence on green energy than AI data centers. According to Bijan Alizadeh, CEO and co-founder of Phoenix Group, one of the largest Bitcoin miners in the world, 70 percent of Bitcoin mining’s energy consumption is green. Alizadeh explains that the desire of Bitcoin miners to generate power themselves has contributed to this result. “That’s the future of Bitcoin mining,” he says. “Bitcoin mining is getting consolidated from 900 companies to 300 now.”
However, “the majority of electricity powering AI data centers comes from fossil fuels,” according to an article by MIT’s Lincoln Laboratory Supercomputing Center. To solve that problem, tech giants like Microsoft plan to deploy nuclear energy in small modular reactors to power its colossal numbers of data centers. “It seems like a bit of an extreme solution,” says de Vries. “There’s going to be plenty of applications where AI just doesn’t make any sense at all,” he says. “This year, we’ve seen so much hype and a lot of fear of missing out – everyone’s just pushing AI into their products regardless of whether it’s a good idea. That’s ultimately where the real waste is.”
In the UAE, crypto mining facilities have found a unique use case in balancing the electricity grid. In a region where ACs are blasting for more than half of the year, the grid has a significant disparity as December hits. The sudden drop in temperature results in significantly less AC usage and a prominent increase in excess energy. One solution to stabilize the grid when this happens is to switch on the Bitcoin mining facilities due to their flexibility, according to Saqr Mashhor Ereiqat, Co-Founder of Crypto Oasis, a well-established MENA-focused blockchain ecosystem. “With that, you actually can store electricity by transforming that electricity into a Bitcoin,” he says.
While this is a temporary solution to balancing the grid, it is still worth debating whether the consensus mechanism of Bitcoin mining is problematic. Meanwhile, the fast-moving tech world also awaits feasible solutions to move data centers away from fossil fuels.
