Facing the national strategic transformation of intelligent manufacturing and full-life-cycle health management during the "14th Five-Year Plan" period, the reliability and safety of core energy units in high-end equipment have become a key bottleneck. As the core of these units, the accurate assessment of the state-of-health (SOH) of lithium-ion batteries is crucial for ensuring system safety and optimizing operation and maintenance strategies. However, the strong nonlinearity and multi-factor coupling characteristics of the battery degradation process pose significant challenges to traditional methods. To address this challenge, a new paradigm for lithium-ion battery SOH estimation driven by a foundational large model is proposed. This framework deeply integrates the excellent long-term temporal reasoning capability of the foundational large models with the advantage of fine-grained feature decoupling in deep learning through an innovative hierarchical temporal token encoding mechanism and a residual-enhanced feature mapping network, achieving end-to-end integrated modeling from raw sensing data to SOH prediction. Specifically, the framework first converts multi-dimensional battery feature sequences into semantic tokens processable by the large model. Weak degradation-sensitive features are then extracted and enhanced using a deep residual network. Finally, the context awareness and self-attention mechanism of the large model are leveraged to accurately capture and predict the capacity decay trajectory across hundreds of cycles. Comprehensive evaluations on four public datasets (HUST, MIT, XJTU, and TJU) show that compared with mainstream benchmark models, the proposed method achieves the optimal or near-optimal performance in terms of root mean square error (RMSE) and mean absolute error (MAE), demonstrating excellent prediction accuracy, stability, and generalization ability. This study not only provides a theoretical perspective and engineering practice scheme for lithium battery life prediction but also offers an example for the in-depth application of large model technology in the field of intelligent operation and maintenance of high-end equipment.