Portfolio optimisation is one of the main tasks in investment to maximise returns while minimising risk under given constraints. However, real-time market conditions and geopolitical factors affecting the financial markets are rapidly changing, and thus long-term historical data become less reliable for capturing current dynamics. Generative Adversarial Networks (GANs) offer a solution by generating synthetic financial data that reflects realistic market behaviour. To address this, we propose a Gated Recurrent Unit-based GAN (GRU-GAN) with Wasserstein Distance (WD) loss, a sequential GAN designed to generate synthetic financial time-series data by capturing temporal dependencies in stock returns, such as those from major US tech companies (AAPL, MSFT, GOOGL, AMZN, META, TSLA). Compared with the traditional V-GAN (Vanilla GAN), which relies on fully connected layers and is suited to static data distributions, the GRU-GAN offers a dynamic modelling approach that better reflects evolving market patterns. This proposed framework enhances portfolio optimisation by augmenting datasets with realistic synthetic data, improving risk assessment, and trading strategy development. The performance metrics demonstrate that the proposed GRU-GAN outperforms the V-GAN, achieving a higher annualised return (2.3624 vs. 1.5054) and a superior Sharpe Ratio (7.3852 vs. 4.1680), reflecting better risk-adjusted returns.