Time Series Foundation Models (TSFMs) represent a paradigm shift in energy forecasting, yet their practical value beyond statistical accuracy remains unexplored. This study reveals that the true potential of TSFMs extends beyond point forecast accuracy to encompass robust uncertainty quantification that fundamentally improves economic outcomes in energy management systems. We evaluate state-of-the-art TSFMs against conventional methods for electrical load and photovoltaic generation forecasting within a Model Predictive Control framework optimizing battery energy storage operations. While TSFMs demonstrate competitive forecasting performance, our key finding challenges conventional wisdom: forecast accuracy does not directly correlate with economic performance under realistic pricing structures. Most significantly, conservative probabilistic forecasting using high quantiles reduces operational costs despite exhibiting higher statistical errors. This counterintuitive result emerges from asymmetric cost structures where peak demand charges create non-linear relationships between forecast errors and economic outcomes. Our analysis demonstrates that energy forecasting requires a fundamental shift from accuracy-centric toward contextaware, objective aligned evaluations. TSFMs’ primary value lies in uncertainty quantification capabilities, enabling adaptive forecasting strategies that align prediction targets with operational requirements rather than statistical metrics.