Artificial Intelligence (AI) presents opportunities and challenges for accelerating a sustainable future, industry experts warn at the InnoVEX 2024. While AI can drive innovation and efficiency, its growing computational demands raise environmental concerns over energy, carbon emissions, and water usage.
The semiconductor industry, driving AI capabilities, is projected to grow from US$600 billion in 2023 to US$1 trillion by 2030, with AI chips surging from $20 billion to $150 billion in the same period. Raymond Chik, Board Chair at chip startup Untether AI, cited research predicting that 20% of the world's electricity could be used by AI computing within a decade.
Training large language models like GPT-3 requires 1,300 megawatt-hours of energy, equivalent to the monthly consumption of 1,500 US households. Even generating a single AI image can consume as much energy as a smartphone charge. "We need sustainable manufacturing processes and energy-aware AI computing infrastructure," Chik emphasized, advocating for AI-accelerated material discovery, optimized manufacturing, and energy-efficient chip architectures like Untether AI's memory-centric design.
Sengmeng Koo, Senior Deputy Director at AI Singapore, quantified the resource demands, stating that training a large language model consumes enough energy to power an average American home for 120 years. A single interaction with such a model uses 20 times more water than a Google search.
Koo further discussed the carbon emissions from AI model training, which contribute to 1-2% of global energy-related greenhouse gas emissions, equivalent to the entire aviation industry. He highlighted that AI model training can emit as much CO2 as five cars over their lifetimes. He also cautioned that the growing demand for hyperscale data centers could exacerbate water scarcity issues, particularly in regions already facing water shortages.
Niven Huang, Managing Director at KPMG Sustainability Consulting, emphasized the need for a "paradigm shift" from a financial-driven to an ESG mindset in business operations. "We are running out of time to work out the dilemma we are facing in environmental and social and economy," he warned, stressing the urgency for responsible resource use and climate action.
Huang believes AI can guide this transition by revealing facts, trends, and real solutions beyond mere compliance reporting. However, he cautioned that AI itself consumes significant energy, potentially increasing carbon emissions. "We still have a lot of concern," he said, questioning whether AI's benefits outweigh its environmental impact.
Sustainable Solutions on the Horizon
Despite these challenges, the speakers highlighted emerging opportunities. Huang believes transformative business models, new materials, energy sources, mobility solutions, and manufacturing processes enabled by AI and advanced computing can drive sustainability.
Chik highlighted prospects in renewable energy integration, water recycling processes, and designing energy-efficient data centers from the ground up. For technology providers, developing faster and more efficient products, algorithms, and chip architectures like memory-centric design could significantly reduce computing demands and operational costs.
Koo echoed similar opportunities, adding that AI Singapore has implemented sustainable practices such as specialized cooling, power management systems with a low Power Usage Effectiveness (PUE) of under 1.2, repurposing hardware, and dynamic resource allocation for efficient utilization. In addition, he cited Google's success in reducing energy consumption by 40% through machine learning algorithms.
As AI continues its exponential growth trajectory, pioneers across industries must confront its environmental paradox head-on. Balancing the technology's transformative potential with sustainable practices and responsible resource management will be crucial for securing a prosperous future for all.
