Marine phytoplankton contribute over half of the global oxygen production and carbon sequestration, and shifts in their community structure profoundly affect carbon sinks and biogeochemical cycles. As global warming accelerates the redistribution of phytoplankton communities, satellite remote sensing has become an indispensable tool for their monitoring. Our research team has made a series of advances in this field. Using in situ hyperspectral measurements from the East China Sea and artificial intelligence techniques, we achieved satellite-based retrieval of phytoplankton species in the coastal East China Sea with HICO (Hyperspectral Imager for the Coastal Ocean) data. By systematically integrating in situ observations, multi-sensor ocean color data, and marine big data, we developed a multi-model ensemble retrieval approach and constructed a global, long-term, daily, gap-free dataset of phytoplankton functional type (PFT) chlorophyll-a concentrations and their distributions. The launch of the new generation ocean color mission PACE, carrying the hyperspectral Ocean Color Instrument (OCI) with 5 nm spectral resolution, a 1–2 day revisit frequency, and a high signal-to-noise ratio (>1000 in the visible), may offer unprecedented opportunities for subcategory classification of phytoplankton types. Our team further built the largest synthetic dataset to date that matches HPLC-derived PFTs with hyperspectral reflectance, and proposed HyperPFT—a lightweight Transformer-based deep ensemble model trained with 100 bootstrapped sub-models—which enables high-accuracy PFT retrieval from PACE/OCI observations. Nevertheless, hyperspectral remote sensing of PFTs still faces significant challenges: concurrent in situ measurements of PFTs and hyperspectral optical data are extremely scarce to date, and retrieving individual PFT endmembers from mixed spectra constitutes an ill-posed inverse problem with inherent spectral unmixing difficulties. Sustained efforts are required to overcome these challenges and advance hyperspectral PFT retrieval toward operational application.