Carbon Emission Modeling for High-Performance Computing-Based AI in New Power Systems with Large-Scale Renewable Energy Integration

Under the global impetus toward carbon peak and carbon neutrality, large-scale renewable energy integration has become a key driver in transforming traditional power grids into new power systems. Meanwhile, the growing adoption of advanced artificial intelligence (AI) approaches, especially large-scale models, heavily relies on high-performance computing (HPC) resources, which pose significant sustainability challenges due to their energy consumption and carbon emissions. This study introduces a newly developed carbon emission model (CEM) that accounts for both embodied and operational emissions in HPC systems. The CEM integrates parameters such as energy intensity coefficients, workload distribution patterns, and renewable deficiency rates, providing a lifecycle perspective of emissions in HPC-based AI applications for power systems. Results reveal that operational emissions dominate, constituting 87% of the total lifecycle footprint. Different regions exhibit varying carbon emissions, and on average, increasing the renewable energy share from 20% to 50% reduces total emissions by 43%, while a full transition to renewable energy achieves a 92% reduction. Circular economy practices, including hardware recycling and sustainable design, are also highlighted to mitigate embodied emissions. This study offers quantitative evidence and actionable insights for power industry stakeholders, enabling the balance between high-performance AI computations and ambitious carbon neutrality goals in renewable-integrated systems.