Researchers unveil Tiny-TSM, a 23M-parameter foundation model trained on one A100 GPU in under a week that achieves state-of-the-art time series forecasting results. Its synthetic data pipeline and novel normalization technique challenge resource-intensive AI paradigms.