FNSPID (Financial News and Stock Price Integration Dataset) is a comprehensive dataset designed to bridge the gap between quantitative financial data and qualitative sentiment analysis for stock market prediction. It comprises 29.7 million stock prices and 15.7 million time-aligned financial news records for 4,775 S&P500 companies, spanning 1999 to 2023, sourced from NASDAQ, Bloomberg, Reuters, Benzinga, and Lenta. The dataset includes raw news text, four types of summarizations (LexRank, Luhn, LSA, TextRank), and sentiment scores for a subset of 50 stocks labeled by ChatGPT on a 1-5 scale. Missing sentiment values are handled via exponential decay. The authors validate FNSPID through extensive experiments with six deep learning models (LSTM, CNN, GRU, RNN, Transformer, TimesNet) for stock price prediction. Key findings show that larger training datasets significantly boost accuracy (6.29% average R² improvement from 5 to 25 stocks), and high-quality sentiment (ChatGPT) provides a modest 0.2% improvement for transformer models, while lower-quality sentiment (TextBlob) degrades performance. The Transformer model achieves the highest accuracy (R²=0.988). FNSPID is publicly available with reproducible code, offering a valuable resource for financial sentiment analysis, LLM fine-tuning, and time-series forecasting research.