The COWT‑SAR dataset provides a vectorized inventory of offshore wind turbines (OWTs) across China’s coastal waters for the period 2020–2025. Derived from Sentinel‑1 SAR imagery using the structure‑aware deep learning framework SA‑OWTDet, the dataset includes 8,593 turbines with attributes such as geographic location, installation year, distance to shore, water depth, and associated sea area. This dataset supports the consistent identification, mapping, and monitoring of offshore wind energy infrastructure, and facilitates applications in energy system analysis, marine spatial planning, infrastructure management, and a wide range of geospatial studies.