This study addresses the evolving challenges of portfolio risk management in the era of Artificial Intelligence (AI), using NVIDIA Corporation (NVDA) as a representative high-tech asset. We propose an integrated AI-augmented framework that combines a hybrid Long Short-Term Memory (LSTM)-GARCH (1,1) model with alternative data, including macroeconomic indicators and financial sentiment derived from news and social media. Trained on daily data from 2018 to 2022 and evaluated out-of-sample in 2023, the hybrid model significantly outperforms ARIMA, standalone LSTM, and GARCH benchmarks in forecasting both NVDA returns (MAE: 1.72%) and conditional volatility (Mincer-Zarnowitz R²: 0.68). Risk analysis based on model outputs reveals NVDA's pronounced exposure to semiconductor supply chain disruptions, regulatory shifts, and tech-sector sentiment. A dynamic portfolio strategy that adjusts NVDA allocation according to predicted volatility achieves a higher Sharpe ratio (3.38 vs. 2.95) and lower maximum drawdown (−28.5% vs. −42.1%) than a static benchmark. Our findings demonstrate that adaptive, AI-driven risk models are essential for managing the non-linear dynamics and tail risks characteristic of AI-centric investments.
