Electricity demand forecasting is crucial for South Africa’s energy planning, especially amid supply constraints and a growing renewable energy sector. This study develops physics-informed neural networks (PINNs) to forecast national electricity demand using Eskom’s data for 2021–2026. The PINNs model integrates domain physics by incorporating solar photovoltaic (PV) and wind generation as constraints in its loss function, ensuring that forecasts reflect the physical impact of renewables on net demand. We forecast hourly and monthly demand over horizons comparable to the official Eskom forecasts and evaluate performancePerformance against the official Eskom projections. The proposed PINNs achieved substantially lower forecast error, improving mean absolute error (MAE) and mean squared error (MSE) by an order of magnitude, and a higher \(\textrm{R}^2\) compared to Eskom forecasts. The results demonstrate that embedding renewable generation data as a physical constraint yields more accurate and potentially generalisable demand predictionsPrediction. This study contributes to the national energy management and grid planning, highlighting how improved demand forecasts can guide capacity expansion, integration of renewables, and demand-side management strategies in South Africa.