Accumulating evidence indicates that microorganisms respond to the ubiquitous plastic pollution by evolving plastic-degrading enzymes. However, the functional diversity of these enzymes and their distribution across the ocean, including the deep sea, remain poorly understood. By integrating bioinformatics and artificial intelligence-based structure prediction, we developed a structure- and function-informed algorithm to computationally distinguish functional polyethylene terephthalate-degrading enzymes (PETases) from variants lacking PETase activity (pseudo-PETase), either due to alternative substrate specificity or pseudogene origin. Through in vitro functional screening and in vivo microcosm experiments, we verified that this algorithm identified a high-confidence, searchable sequence motif for functional PETases capable of degrading PET. Metagenomic analysis of 415 ocean samples revealed 23 PETase variants, detected in nearly 80% of the samples. These PETases mainly occur between 1,000 and 2,000 m deep and at the surface in regions with high plastic pollution. Metatranscriptomic analysis further identified PETase variants that were actively transcribed by marine microorganisms. In contrast to their terrestrial counterparts—where PETases are taxonomically diverse—those in marine ecosystems were predominantly encoded and transcribed by members of the Pseudomonadales order. Our study underscores the widespread distribution of PETase-containing bacteria across carbon-limited marine ecosystems, identifying and distinguishing the PETase motif that underpins the functionality of these specialized cutinases.