A technique for predictive streaming involves receiving a request for a block associated with a streaming application and serving data associated with the block. A block request database is checked to predict what block is likely to be requested next based upon prior block request data. The predicted block may be identified when serving the data associated with the requested block. A system developed according to the technique includes a streaming server, a block request database, and a prediction engine that uses the block request database to predict block requests. The streaming server provides data associated with the predicted block request.