windmills
By Cully Cavness, Co-Founder, President and COO
One of the great paradoxes facing artificial intelligence (AI) is whether it will pose a major climate risk… or be an integral part of the climate solution. Recently, numerous articles and news stories exclaim that AI is driving unprecedented increases in power demand and if unchecked will lead to runaway carbon emissions.
Yes, AI is energy-intensive and consumes a significant quantity of electricity. The IEA forecasts that its growth will drive more than a doubling of data center electricity demand between 2022-2026. However, relative to total global emissions, AI’s incremental emissions will be modest as discussed below. The counterbalance to AI’s incremental emissions will be the emissions reduction from AI-generated technologies, which may greatly outweigh the emissions increase from AI’s energy use. AI is a critical tool that will accelerate the technological innovations we need to solve the challenges of climate change in ways that are economical and practical.
Without fundamental technological changes, our path to any widely-held climate goals is increasingly dubious. 2024 was the warmest year on record globally, and also the first year that average global temperature surpassed 1.5°C above pre-industrial level – exceeding the global climate goal that was set at the Paris Climate Agreement. As we shared in our last blog, the scenarios advocated by the IPCC to limit warming to no more than 1.5 or 2.0oC by 2100 also appear increasingly unrealistic when viewed in the context of current technologies and simple human nature. Our best shot is breakthrough innovation, and there is no one single technological solution that can easily reverse this trend. We’re going to need a plethora of technological breakthroughs including but not limited to:
An all-of-the-above energy strategy that includes improved efficiencies for renewable energy and fossil fuel energy sources across the value chain from production to consumption
New battery chemistries that enable long-term storage at lower costs, enabling a greater share of intermittent renewables over time
Smarter and more dynamic grids
New clean energy sources like modernized fission processes and breakthroughs in fusion
Breakthroughs in carbon capture that radically reduce the price per ton of CO2 abatement
Improved climate adaptation and mitigation technologies such as wildfire prediction and more accurate climate modeling
And we need all of this innovation as soon as possible, which is why AI is essential. The technological innovations that AI will enable increasingly represent humanity’s best new opportunities to combat climate change.
AI Energy Demand in Context
We can view the climate risks and rewards of AI like an investor. On one hand, we may “risk” an incremental 1-2% of global emissions through the aggressive scaling of AI training and inference compute infrastructure. On the other hand, the “return” may be measured in tens of percentage points of reduced emissions through technological breakthroughs such as fusion, new battery chemistries, carbon capture systems innovation, wildfire prediction and prevention, or innovative materials science (not to mention other societal benefits ranging from cures to chronic disease to autonomous robots and self-driving cars that save humanity countless hours of toil). Real progress is already underway in many of these areas, such that the risk-reward increasingly skews towards a positive long-run climate payoff from AI.
We should not sugar coat the reality of AI power demand, but we should view it in its appropriate context. IEA estimates that data centers account for approximately 1-1.3% of global electricity today. To put this in context, this is about half of the electricity used by computers, phones, TVs and other household IT appliances. Data transmission networks account for another 1-1.5% of global electricity consumption. In terms of greenhouse gas (GHG) emissions, data centers and data transmission networks were responsible for approximately 1% of global energy-related GHG emissions.
AI is driving up total global power consumption as we energize new and existing data centers to meet the growing electrical load requirements of the GPUs needed to support advanced AI training and applications. McKinsey & Co. estimates that U.S. data center electricity consumption will rise to approximately 600 terawatt-hours (TWh) per year in 2030, or nearly 12% of total U.S. electricity demand. 12% of electricity sounds like a big number, but for emissions impact comparisons we should compare data centers against total U.S. primary energy consumption (a metric that includes energy for heating and transportation), which is approximately 27,500 TWh. In this light we see that even in an aggressive growth case the energy consumption of data centers is perhaps 2% of total energy consumption in the U.S.
Similar power demand trends are expected on a global basis. Due to the growth of AI, data centers’ total electricity demand globally could exceed 1,000 terawatt-hours by 2026, more than doubling the 460 TWhs they consumed in 2022. However, even with this significant jump forecasted, the IEA points out that data centers are not the dominant driver of global electricity demand growth in the near term. While the rise of AI is certainly a factor, global electricity demand growth is expected to increase by 6,750 TWh by 2030 in the IEA’s Stated Policies Scenario. Continued economic growth, the growing adoption of electric vehicles and air conditioners, and the growth of electricity-intensive manufacturing are all cited as bigger drivers while the contribution of data centers to global electricity demand is merely “modest.”
It is important to remember that data centers are critical for many digital services from emails and streaming to mobile use and cloud gaming, not just for AI. Even with its rapid growth, Goldman Sachs forecasts that AI will only represent 27% of data center power demand by 2027. Furthermore, thanks to the growing proportion of zero-emissions power generation on the grid and the increasing focus of data centers on using renewable and clean energy sources, growing electricity demand will not increase emissions by a commensurate amount.
The Climate Needs Disruption
When it comes to mitigating climate change, time is of the essence. The scientific evidence that greenhouse gas emissions are leading to global warming and a changing climate is ample. Yet scalable solutions to humanity’s emissions trajectory have not materialized, allowing emissions and global temperatures to rise unabated.
The IPCC publishes scenarios of socioeconomic and technological development with associated projections of temperature rise. A cursory familiarity with history, human nature and recent political trends will lead most fair-minded readers to conclude that the scenarios leading to a sub 2oC outcome are increasingly non-credible.
Without technological breakthroughs, we are headed for scenarios with 3-4oC warming (or more) by the end of the century, far greater than the 2oC that scientists warn may lead to significant harm to biodiversity on our planet and increased risk of runaway “tipping point” scenarios. AI may be the essential ingredient to solving the climate challenge by speeding up innovation and driving down costs.
AI-Driven Technologies Offer Climate Solutions
We are on the cusp of a new crop of revolutionary energy and environmental technologies to be enabled and scaled on the back of AI. The Global Partnership on AI’s report, Climate Change and AI, highlights how AI’s unique capabilities enable it to accelerate climate action and transform industries and sectors. AI can help quickly distill large amounts of unstructured raw data, turning it into actionable information, a task that is slow and laborious for humans. It can also combine data to improve predictions, juggle many variables and factors to optimize complex systems, and accelerate scientific modeling, experimentations, and discoveries through rapid simulations.
Within the electricity system in particular, AI can help in a number of ways, including by more accurately forecasting electricity supply & demand to help balance power grids efficiently. This is critical to our ability to integrate more renewables, which are intermittent, onto the grid. AI can also improve algorithms for electricity scheduling and storage, and manage decentralized microgrids. All this allows us to optimize energy production, use and storage, improve efficiency, and enable more clean energy to flow onto our grids.
Batteries–key to electrification and storage of clean energy—are another great example demonstrating AI’s potential. Improving battery life, battery chemistries, and their long-term storage capacity will be a key part to effective decarbonization. One of Crusoe’s customers, SES AI, a leading developer of advanced lithium-metal batteries for EVs and urban air transport, is using Crusoe Cloud to accelerate material discovery to improve batteries. By using AI to map the universe of small molecules in electrolytes, they hope to better optimize electrolytes to unlock the full potential of lithium-metal batteries and revolutionize battery chemistry, enhance energy storage solutions, and advance energy efficient technologies.
AI can also help us unlock energy solutions beyond renewables such as clean fuels. Avalanche Fusion (formerly Avalanche Energy) is developing a first-of-its-kind fusion microreactor that can provide clean power for a wide variety of applications from microgrids to transportation, and an important part of their design process is rapid simulation to quickly evaluate and analyze different configurations of the microreactor. Avalanche ran their underlying simulations on Crusoe Cloud, which informed in-lab experiments that can start within hours of completing a simulation, reducing the time needed to develop their mini reactor.
AI can also be used to help manage climate-related risks. The ability to analyze large climate datasets allows for faster and more accurate predictions and assessments of extreme weather events, sea level rise, and changes in precipitation patterns, which can be used by businesses, governments, and insurers to make more informed, data-driven decisions about both mitigation and adaptation strategies. One of Crusoe’s clients Jua created foundational weather and climate prediction models and used AI to speed the training of these models to more accurately predict global weather patterns for the optimization of renewable energy trades and to improve the management of physical climate risks.
Meeting AI and Data Center Energy Demand in a Sustainable Way
AI is both a climate risk and a climate solution. However, the magnitudes of the potential risks and the potential benefits are not equal.
The risk that AI poses to the climate is measured in low single digit percentages of incremental emissions, and can be mitigated to a degree with low-emissions energy procurement by data centers. The potential benefits of AI can be measured in many tens of percentage points of emissions reduction, not to mention other wide-ranging opportunities for societal benefit. The short term incremental emissions from AI energy consumption appear worthwhile when viewed in this context.
To mitigate that short-term risk, Crusoe focuses on reducing the environmental impacts of data centers by targeting clean and renewable energy sources to power our AI data centers. In this way, Crusoe enables the innovation of our customers in an environmentally-aligned capacity so they can build the products and solutions that can help solve the climate challenge (and many others).
In conclusion, while the growing energy demand of AI presents a legitimate concern, its potential to accelerate critical climate solutions far outweighs the risks. AI is not merely another electricity consumer. It is a powerful tool that can unlock breakthroughs in energy efficiency, renewable energy, grid management, and carbon capture. By harnessing AI's capabilities, we will accelerate the development and deployment of technologies needed to mitigate climate change and transition to a more sustainable future.
Liked what you just read? Share it
