Weather files are crucial in building physics analysis but with climate change, the use of historic data may not adequately represent the range of conditions faced throughout a design’s service life. This study focuses on the capability of Artificial Intelligence (AI) in creating future weather data, addressing the limitations of current future weather generation tools that are often location-specific, costly, time-consuming, or outdated. Using a 70-year historic dataset from New York City, multiple transformer-based neural network models were trained and used to forecast the year 2020. When matched against 2020 measured data, AI generated files had smaller mean absolute errors than a 2020 file created using the common ‘morphing’ methodology (CCWorldWeatherGen). The AI files were then used in a residential apartment energy simulation and moisture risk analysis. The AI files produced heating/cooling load results which more closely matched the measured 2020 weather scenario compared to historic typical meteorological year (TMY3) and the ‘morphed’ future file. Similarly, when applied to an annual moisture risk analysis, iterations between the weather files showed approximately 400 h difference in condensation potential. This gap in predictions can potentially misguide a design team and client on performance or strategy towards carbon neutrality. The efficiency of AI can be leveraged as an additional approach to enhance the breadth of building design guidance.