Parallel applications can benefit greatly from massive computational capability, but their performance usually suffers due to large latency in I/O accesses. Conventional I/O prefetching techniques are conservative and are limited by low accuracy and coverage. As the processor performance has been increasing rapidly and the computing power is virtually free, we introduce a novel speculative approach for comprehensive and aggressive parallel I/O prefetching in this study. We present the design of our approach, as well as challenges, solutions, and the prototype implementation. The experiments have shown promising results in reducing I/O access latency.