Large language models and vision transformers have benefited from massive, standardized datasets like Common Crawl and ImageNet, which provided the necessary scale for emergent properties to appear. However, temporal data—the time-series information generated by sensors, stock markets, and industrial processes—has historically lacked a unified data infrastructure of comparable scale. Most time-series research relies on fragmented, domain-specific datasets that are too small to support the training of true foundation models. The release of TimeNet, an open-source data infrastructure for Temporal Foundation Models (TSLMs), aims to rectify this by providing a standardized framework for large-scale temporal data curation and benchmarking.…