Rule of thumb: if the data fits on an ordinary PC, it's not Big Data. The right tool is probably an Ordinary Database.
I mean if that was the case then ELK stacks would not be running in cluster mode. There many 3-Node ELKs handling less than 4 TB of data efficiently where one could not. By efficiently I mean, not having time for Coffee while waiting for Kibana to load the graphs...
Your point is simply not true. Many users of big data platforms e.g. Spark aren't doing so because of the volumes of data. It's because they want to do machine/deep learning on a proven and popular platform. With technologies like Caffe, Sparkling Water, Tensorflow all available on the one platform.