 |
According to new research, the energy required to run an AI system may already be greater than the amount of electricity used to mine Bitcoin.
This may sound like good news for the Bitcoin mining industry, which has been consistently criticized for its power consumption, but there are concerns that AI is emerging as a fierce competitor to Bitcoin in the areas of electricity and equipment, according to a report from the Bitcoin Policy Institute.
Thanks to the deep pockets of the industry, AI companies can afford to charge more for the same electricity than miners. With AI offering up to 25x more revenue per kilowatt-hour (kWh) than Bitcoin, some miners are adding AI processing to their data centers or even switching from Bitcoin to AI entirely.
“This trend will continue as long as AI’s revenue per megawatt-hour is higher than Bitcoin’s,” BPI researcher Margo Paez told Cointelegraph.
The AI industry is still in its infancy, but the energy demands of generative AI models are enormous. According to Goldman Sachs, a single ChatGPT query consumes about 10 times the energy of a typical Google search. MIT Technology Review reports that generating AI images can use as much power as fully charging a smartphone.
Bitcoin mining’s high energy usage has led to a ban in Europe and a moratorium in New York. According to BPI, US Bitcoin mining facilities use about 121.13 terawatt hours (TWh) of energy annually, while AI could consume between 20 and 125 TWh in 2023 (it’s hard to get an exact figure because AI is housed in data centers used for other tasks).
But with generative AI gaining massive traction this year, the report estimates that AI will use 169 TWh in 2024, growing faster than Bitcoin mining, reaching around 240 TWh in 2027, while mining is expected to reach 160 TWh.
Data centers that house AI models also require significant amounts of water to cool the machines to maintain efficiency. A study by Shaolei Ren of the University of California Riverside estimated that ChatGPT consumes about 500 milliliters of water for every five to 50 questions asked.
By comparison, it is estimated that Bitcoin mining in the United States alone requires between 93 and 120 gigalliters of water per year, with each transaction using enough water to fill a backyard swimming pool (although these estimates are controversial due to concerns about their potential inaccuracy).
AI investors’ massive funds pose risks to miners.
The profit margins for AI computing are currently much higher than those for Bitcoin mining. While mining generates revenues of $0.17 to $0.20 per kWh, the revenues for Nvidia graphics processing units used for AI range from $3 to $5 per kWh, a difference of 17 to 25 times.
So why don’t Bitcoin miners leverage their AI-powered rigs to make more money?
Anibal Garrido, a Bitcoin miner and crypto asset advisor, told Cointelegraph that the transition is not easy because miners use ASIC (application-specific integrated circuits) equipment that is designed only to compute hashes for PoW protocols. ASICs cannot be reused for AI.
But Bitcoin miners also need to constantly update and replace their equipment, and the facilities themselves need to adapt. Paez Many Bitcoin miners are already retrofitting their facilities to accommodate GPUs, and he said he knows of at least one company that has completely transitioned from Bitcoin mining to AI.
Alex de Vries, a data analyst and researcher at Vrije Universiteit Amsterdam and De Nederlandsche Bank, told Cointelegraph that competition for electricity will only intensify.
“AI companies have much deeper pockets than the crypto mining industry,” he said. De Vries believes AI companies may already be “going after Bitcoin miners’ power contracts.”
He explained that in the current AI craze, the AI industry needs immediate access to power and equipment and cannot afford to wait for years-long construction projects to build new data centers. That means the threat to mines is real.
Flexible vs. Fixed
AI’s increased power consumption could help change the politics surrounding the electricity used for Bitcoin mining, which is relatively inexpensive by comparison.
Not all electricity consumption is created equal, and Bitcoin mining rigs are much more flexible and can be turned off or on to take advantage of surplus, wasted, or cheap electricity.
AI, on the other hand, requires “99.9% uptime” for its models to function properly. This demand means it consumes all available energy regardless of cost, which can lead to less environmentally friendly or even dangerous energy sources.
Peaker power plants, which are brought online to meet unexpected spikes in demand, often run on fossil fuels, exacerbating their environmental impact.
The flexibility that Bitcoin mining provides allows miners to negotiate with governments to stop consuming energy when the grid is saturated. Once the grid stabilizes, miners can resume their work, adding flexibility to keep the grid balanced.
According to the BPI report, between 5% and 31% of U.S. Bitcoin miners have shut down operations because electricity prices were too high or because grid operators ordered them to do so.
The study, which collected data from eight U.S. mines between July and September 2023, estimates that the shutdown prevented 13.6 kilotons of CO2 emissions. That reduction is equivalent to taking 2,951 cars off the road.
Paez said that because AI is not as flexible as bitcoin mining, “the only way they can manage their emissions is to invest directly in renewable energy.”
Also read
6 questions…
6 Questions for JW Verret – The Blockchain Professor Who Tracks Money
characteristic
Escape and Former Communications: Blowing Up Steemit with Andrew Levine
Location-independent incentives for industry and fixed and energy incentives
Another important difference between the two technologies is their location requirements. While Bitcoin mining is location-independent, AI requires low latency to provide ultra-fast responses, which requires data centers to be located near major metropolitan areas.
That means AI data centers have to consume available energy in a specific location, while Bitcoin miners can move to locations with surplus energy, such as renewable energy facilities in remote areas where hydroelectric power, solar, or wind power are abundant.
Supporters argue that Bitcoin mining could support the transition to renewable energy, provide predictable demand during periods of low demand, and provide financial stability for new projects.
The location-independent nature of Bitcoin mining also allows for the utilization of wasted energy. This includes mining from isolated energy sources in remote hydroelectric power plants, capturing excess methane emissions, reusing wasted heat to electrify heating, or utilizing renewable energy from solar and wind energy sources that may be isolated due to transmission constraints.
Can AI become more efficient?
Juan Calvo, senior data engineer and Gen AI engineer at Datatonic, thinks AI needs to become more sustainable, echoing the push to use renewable energy for Bitcoin mining.
“We need to assess whether the ability to do something justifies its implementation, and we need to emphasize the importance of ethical and sustainable choices in technology development.”
The engineer explained that AI developers have a variety of techniques to improve energy efficiency, including fine-tuning existing models, using smaller models for specific tasks, and leveraging cloud solutions, which can significantly reduce overall energy consumption.
Hardware advancements can also play a role. Graphics manufacturers like Nvidia are leading the way in developing specialized hardware that improves performance while consuming less energy. The synergy between more efficient algorithms and advanced hardware can help address AI’s growing energy needs in a more sustainable way.
However, de Vries points out that the bigger-is-better dynamics of generative AI could undermine these efficiency gains. In a 2023 study, he highlighted that the increasing energy footprint of AI models is driven by incentives to develop ever-larger models, which in turn increases the demand for computational resources and energy.
“Efforts to further improve these models by increasing their efficiency and reducing their energy costs may become more feasible, thereby negating some of the efficiency gains.”
De Vries compared this dynamic to the increasing efficiency of cryptocurrency mining hardware: As mining equipment becomes more efficient, Bitcoin miners simply earn more.
Subscribe
The most interesting articles on blockchain, delivered once a week.
Daniel Ramirez-Escudero is a journalist who has been obsessed with cryptocurrencies since early 2017, and has many years of experience in the media industry. He is a crypto enthusiast, passionate about geopolitics, and interested in the financial, philosophical, and technological revolution brought about by Bitcoin.