<p>The building sector is a primary contributor to global energy consumption and CO₂ emissions. As climate policies mandate a shift towards sustainable practices, accurately forecasting energy demand and emissions is critical for effective planning. This study addresses the need for a robust, multi-model framework to predict energy consumption and subsequent CO₂ emissions in both traditional and Nearly Zero-Energy Buildings (nZEBs), thereby enabling proactive, data-driven strategies for decarbonization. Traditional forecasting methods often fail to capture the complex, non-linear interactions between climate variables, building characteristics, and energy use. While various AI models have been applied individually, their standalone performance can be inconsistent. A hybrid ensemble framework, which leverages the strengths of multiple models, offers a promising solution to enhance predictive accuracy and reliability. This research introduces a novel multi-level hybrid ensemble framework that uniquely reconciles disparate data granularities, ranging from high frequency nZEB sensors to sparse traditional building records. Unlike existing studies focused on single-model or high-resolution data, this approach integrates six diverse AI algorithms to maintain predictive reliability across inconsistent datasets. The study provides a first-of-its-kind comparative analysis of long-term decarbonization pathways up to 2050, offering a strategic roadmap for balancing legacy and modern building infrastructures. Our results demonstrate that the proposed hybrid ensemble framework significantly outperforms individual models in forecasting energy demand, achieving a substantial reduction in prediction error. The framework’s ability to accurately project energy consumption allowed for a detailed analysis of long-term CO₂ emission pathways. Our results project that after 30 years, the increase in energy demand will be 120.7% for nZEBs, while for traditional buildings, it will be significantly higher at 199.1%. This research presents a robust predictive framework whose key insight is the quantification of a ‘climate resilience gap’: traditional buildings show ~ 1.65 times greater climate-induced energy demand sensitivity (199.1% vs. 120.7% increase by 2050) than nZEBs under a BAU scenario. By integrating multiple AI models, our framework provides a powerful tool for strategic energy planning, supporting sustainable development and informed decision-making to mitigate future CO₂ emissions.</p> Graphical Abstract <p></p> <p>This graphical abstract provides a quick, visually appealing summary of the research on a Hybrid AI Ensemble for Forecasting Energy Demand in nZEB and Traditional Buildings. It’s designed to give readers a rapid understanding of the study’s core findings and methodology without requiring them to read the entire manuscript. The visual starts by outlining the Hybrid Ensemble AI Approach, which combines several machine learning models, including ANN, RF, XGBoost, and RBFN, along with an Autoencoder and Decision-tree. This approach is applied to Energy Consumption Data from both traditional and nearly zero-energy buildings (nZEB). The AI Modeling Comparison step shows how the different models are evaluated before being combined into the final Hybrid Ensemble and fed into the Prediction Engine. The output of this engine is used for Strategic Energy Planning to 2050, which is powered by a LSTM (Long Short-Term Memory) model for temporal prediction. The key results are visually represented as a reduction in CO₂ Emission, support for Sustainable Planning, and providing Stakeholder Insight. The research aims to Optimize Energy Use, Support Sustainability, and Inform Decision-Making for a more sustainable future. This clear flow of information effectively communicates the research’s process, findings, and overall importance in a concise format.</p>