As artificial intelligence (AI) applications sweep across various industries, the rapidly escalating demand for energy becomes increasingly apparent, revealing a potential crisis lurking beneath the surface. Take, for example, generative AI models such as ChatGPT. Each instance of these sophisticated systems can consume as much electricity in a single day as 180,000 average U.S. households. Even more daunting is the energy footprint associated with training models like GPT-4, which devoured over 50 gigawatt-hours—equivalent to roughly 0.02% of California’s yearly electricity production and a staggering 50 times the energy needed for its predecessor, GPT-3. As the adoption of AI technologies continues to swell, it has become essential to examine how this need for electricity intersects with existing infrastructure and environmental considerations, particularly in relation to the U.S. electrical grids.
California’s Pivotal Role in AI Energy Dynamics
California stands at the forefront of this emerging energy crisis. The state’s largest utility, PG&E, anticipates a doubling in demand for electricity by 2040, primarily driven by the rise of AI, manufacturing initiatives supported by federal subsidies, and the proliferation of electric vehicles. This surge presents a multifaceted challenge: data centers, which must maintain constant and reliable power, may increasingly need to pivot towards more traditional energy sources. Renewable technologies like solar and wind, while crucial, cannot alone meet the demands given their intermittent nature; this reliance on inconsistent outputs forces a greater dependency on fossil fuels and nuclear energy. As we project demand increases of 13%-15% annually within the data center sector through 2030, it is critical to find viable solutions that align with a sustainable energy future.
Hydropower: The Forgotten Utility
Among the potential solutions is the long-neglected realm of hydropower, which appears to offer remarkable promise in addressing growing energy demands. According to Shon Hiatt, an associate professor at USC Marshall School of Business, significant untapped potential exists within America’s hydropower landscape. Historically, this renewable energy source has not been fully utilized, with estimates indicating that the U.S. could generate an additional 10 gigawatts of energy simply by upgrading existing hydropower plants. Moreover, fewer than 3% of the nation’s over 90,000 reservoirs currently contribute to energy generation, underscoring a glaring opportunity waiting to be seized. By installing turbines and generators on these existing reservoirs, the U.S. could harness an additional 12 gigawatts of power, providing a timely and effective response to the rising energy needs of the AI sector.
Balancing Act: The Trade-offs of Energy Sources
In any discussion about energy solutions, it is crucial to recognize the trade-offs associated with different sources. Solar energy, while progressive, requires considerable land and has a linear scaling issue that could limit its efficacy. Wind energy has its own environmental concerns, impacting wildlife and ecosystems. Nuclear power, on the other hand, has both a small footprint and the capacity for substantial energy output. A particularly compelling option may lie in run-of-the-river hydropower, which exerts a lower ecological impact as it minimizes the need for reservoirs. The U.S. Department of Energy estimates an astonishing 65 gigawatts of unexploited hydropower potential in this sector alone. However, the path to realizing these gains is often obstructed by lengthy government licensing and permitting processes, which can delay important developments.
Future Pathways: Natural Gas and Nuclear Solutions
With demand for electricity projected to soar in the coming years, one pragmatic expectation is that new combined cycle natural gas facilities will likely fill the gap before renewable solutions can be fully implemented. Natural gas plants can be erected swiftly, require less land, and benefit from the current oversupply and low costs associated with this energy source. Small modular nuclear reactors (SMRs) may also present a viable long-term solution, although their development is not expected to yield operational facilities until at least 2030.
The intersection of AI and energy presents a formidable challenge, demanding innovative solutions that prioritize sustainability. As we consider the available avenues, hydropower emerges as more than just a nostalgic relic; it can be a foundational component in powering our future technologies. By leveraging existing resources and actively engaging in infrastructure upgrades, the U.S. has a golden opportunity to navigate the complexities of energy consumption amidst a technological revolution. Our approach to these challenges will dictate not just the success of AI, but our commitment to fostering a sustainable energy ecosystem for generations to come.