Introducing “AI and Energy Dynamics” – a collaborative series by leading technology and financial publications exploring the intersection of artificial intelligence and energy infrastructure. Each Monday for six weeks, experts will delve into how generative AI is transforming global power landscapes.
Casey Crownhart, an energy-focused senior reporter, teams up with Pilita Clifton, a seasoned columnist, to analyze how China’s aggressive renewable energy expansion might accelerate its AI advancements.
Casey Crownhart explains:
Contrary to popular belief, the primary obstacle to AI development isn’t funding but energy availability. This issue is especially pressing in the United States, where numerous data centers remain offline due to insufficient power infrastructure.
Historically, data centers managed to meet rising demand through efficiency gains for nearly a decade before 2020. However, with billions of AI-driven queries daily, electricity consumption is surging beyond efficiency improvements. The result: strained grids and escalating electricity costs in regions hosting dense data center clusters.
To sustain AI growth without inflating energy prices nationwide, the U.S. must adopt strategies that prioritize energy abundance. China offers a compelling example.
By 2024, China is projected to add approximately 429 gigawatts (GW) of new power capacity-over six times the net increase expected in the U.S. during the same period.
While coal remains a significant part of China’s energy mix, its share is shrinking as the country rapidly scales solar, wind, nuclear, and natural gas installations.
Conversely, the U.S. is attempting to revive its declining coal industry, despite coal plants being costly, polluting, and increasingly unreliable-operating at just 42% capacity in 2023 compared to 61% in 2014. Without decisive action, the U.S. risks falling behind as a consumer rather than a pioneer in both energy and AI technologies. Notably, China’s renewable energy exports now generate more revenue than U.S. oil and gas exports combined.
Accelerating the deployment of renewable energy facilities is a practical solution. Wind and solar projects remain the fastest and most cost-effective to commission, yet they face political resistance in the U.S. Natural gas offers a transitional option but is hindered by supply chain delays for critical equipment.
In the short term, enhancing the operational flexibility of data centers could alleviate grid stress. If data centers agreed to reduce power consumption during peak demand periods, new AI infrastructure could be integrated without immediate upgrades to energy systems.
A Duke University study revealed that if data centers curtailed their electricity use by a mere 0.25% annually-equivalent to about 22 hours-they could free up enough capacity to supply 76 GW of power. This would effectively boost grid capacity by 5% without constructing new plants.
However, such flexibility alone won’t suffice to meet the anticipated surge in AI-related energy demand. Pilita, what strategies could the U.S. adopt to overcome these energy limitations? What additional factors should we consider regarding AI’s power consumption?
Pilita Clifton responds:
I concur that demand-responsive data centers should become standard practice rather than exceptions. Innovative agreements offering discounted electricity to data centers in exchange for utility access to backup generators can reduce the need for new power plants-benefiting both sides regardless of AI’s ultimate energy footprint.
Understanding AI’s future electricity consumption is critical, yet current projections vary widely-from doubling to quadrupling today’s data center energy use within five years.
This uncertainty stems partly from a lack of publicly available data on AI systems’ energy requirements and efficiency improvements. For instance, Nvidia reported last year that its specialized AI chips have become 45,000 times more energy-efficient over the past eight years.
Past predictions about technology’s energy demands have often been inaccurate. During the 1999 dot-com boom, some forecasts claimed the internet would consume half of the U.S.’s electricity within a decade, prompting fears of increased coal dependency that never materialized.
Nonetheless, some regions are already feeling the strain. In Ireland, data center power consumption is so high that new grid connections around Dublin have been restricted to prevent overloads.
Regulators worldwide are considering mandates requiring tech companies to generate or procure sufficient renewable energy to offset their consumption. Such policies are vital and should expand. Ideally, AI will not only drive energy abundance but also accelerate the clean energy transition essential for combating climate change. OpenAI’s CEO Sam Altman suggested in 2023 that superintelligent AI could make addressing climate challenges significantly easier.
Yet, current trends in the U.S. are discouraging, with renewable projects facing cancellations even as renewables accounted for over 90% of new global power capacity additions last year.
Europe is advancing plans to power major data centers primarily through renewables and battery storage, while China leads the global green energy surge.
In the 20th century, fossil fuel-rich nations dominated geopolitics. The U.S. aims to maintain this legacy, but China is poised to become the first “green electrostate.” If China leverages this to win the AI race-historically led by the U.S.-it will mark a pivotal moment in economic, technological, and geopolitical history.
Casey Crownhart replies:
I share your reservations about optimistic claims that AI will revolutionize climate action. While AI evolves rapidly, we cannot afford to wait for unproven technologies to deliver on such promises.
Experts suggest AI could enhance grid management and planning, but these applications remain experimental.
Meanwhile, many countries are making tangible progress toward cleaner energy systems. The impact of these shifts on the AI boom remains uncertain, but it’s clear AI is reshaping both energy grids and society. Vigilance regarding its environmental footprint is essential.
Additional Resources
- In-depth analysis of AI’s energy consumption per query by leading technology journalists.
- Visual data explorations of the accelerating global AI capability race.
- Investigations into the feasibility of truly sustainable data centers worldwide.
