From: redpointai
The rapid growth of artificial intelligence (AI) is creating an unprecedented demand for energy, bringing the topic of power generation to the forefront of technological and environmental discussions . Even without AI, the United States needs to increase its electrical grid capacity fivefold by 2050 to meet goals such as converting all gas-powered vehicles to electric and expanding manufacturing for materials like steel and cement . Technology is seen as the primary solution to enable 8 billion people globally to live in comfort and safety . The core challenge is making a significantly larger amount of energy available in a sustainable and affordable manner .
The demand from hyperscalers in particular is accelerating the deployment of new energy technologies . If the cost of energy could be reduced tenfold and widely distributed, it would unlock the potential to significantly improve living standards for many people worldwide . Lowering energy costs by 10% year-over-year for the next 20-30 years could lead to transformative changes, including universal AI compute, widespread air-conditioned comfort, and advanced manufacturing capabilities . Energy is currently the limiting factor for AI applications and market potential across various industries | hyperscalers .
Current Energy Landscape and Future Directions
The energy market is undergoing significant changes, with new solutions emerging.
Solar Power
Solar energy is currently the most cost-effective way to add new electricity to the grid in the United States, accounting for 80% of new energy in 2024 . However, solar power is intermittent, operating only about 25% of the time, making it unsuitable for applications requiring 24/7 power, like a data center filled with AI chips . Efforts are being made to combine solar with battery backup to extend its usability .
Battery Technology
Lithium-ion batteries, introduced in 1991, have become 97-98% cheaper since their inception and continue to decrease in cost by over 10% annually . The ability to mass-manufacture batteries cheaply in large factories is compared to the efficient production of chips for computing .
Fusion Energy
Fusion is considered a highly promising energy source . It is a reaction that works, occurring at the center of the sun, and humans have successfully replicated it multiple times . Fusion is incredibly power-dense; theoretically, one supertanker could fuel the entire United States grid for a year, or a pickup truck could power a major plant for a year . It also lacks concerns about waste associated with other nuclear processes . The technology could enable a factory to produce energy with minimal material input in about two years .
Offshore Compute Platforms
Companies like Panasa are developing offshore compute platforms, which are 200-meter tall structures that float in the ocean . These platforms generate energy by harnessing wave power and combine energy generation with cooling capabilities due to their oceanic immersion . This approach could lead to the cheapest inference platforms on the planet, opening up possibilities for massive increases in compute capacity .
Geothermal Energy
AI is being applied to geothermal exploration. For example, a company named Zanar uses AI to analyze data and identify optimal drilling locations where hot water or steam can be extracted, making the process of finding these energy sources more efficient than traditional drilling methods . This concept extends to finding other deposits like copper or hydrogen .
AI’s Role in Energy and Climate Change
The intersection of AI and energy is a key area of discussion . The demand for AI compute, particularly for running full-time AI agents, will require substantial power . As AI models, especially reasoning models, shift compute demand from training to inference, the cost of these requests will become crucial . Cheaper inference requests could lead to nearly infinite demand for inference time reasoning compute . This drives companies like Panasa and next generation fusion technologies to become vital for hyperscalers to grow their businesses .
Hyperscalers are actively engaging with energy providers, with most announcing power purchase agreements for existing or new nuclear power plants . Some, like Meta, have issued RFPs for next-generation fission plants .
Short-term Challenges and Long-term Vision
In the short term, the energy demands for AI may lead to a reliance on natural gas power plants, as they can quickly add gigawatts of power to the grid within a year or two . While some believe that AGI will eventually solve climate change, it’s considered insufficient to “punt the problem down the hill” . A dual approach is needed: planning for short-term needs while investing in long-term sustainable solutions like solar, battery backup, next-generation geothermal, and fusion .
AI Applications for Climate Solutions
Beyond energy generation, AI is also being leveraged for various climate-related applications:
- Weather Prediction: AI can improve weather prediction and related areas like insurance risk assessment .
- Materials Discovery and Optimization: AI is crucial for inventing and optimizing new materials for carbon capture, catalysts, and other reactions . This involves searching multi-dimensional spaces to find materials with specific properties quickly .
- Home Efficiency: AI can help consumers improve home efficiency and comfort. By using a thermal camera attached to a phone, AI can process videos of a house and provide recommendations for improvements like adding insulation or replacing appliances, with clear payback periods .
The effectiveness of AI solutions in these areas depends on how much of the overall time and cost they address in bringing a solution to market . Solutions that directly impact a critical activity, like geothermal exploration, are more impactful .
Future AI Progress and Energy
The core challenge for humanity’s progress is making enough energy available cheaply and sustainably . The demand for AI compute, particularly for inference, is seen as nearly infinite if the price is right . This reinforces the need for massive, affordable energy generation.
The speaker believes that the significant progress in AI, from convolutional neural nets in 2013 to current large language models, stems from the ability to scale compute and data, following clear “S-curves” of development . Similarly, energy technologies like solar and batteries are on their own S-curves, with costs continuously decreasing . The general trend of technology making things “better, faster, cheaper” is driven by investing heavily in R&D and scaling manufacturing of standardized components .