Many papers claim that it will be needed new approaches to allow genome sequence assemblers to scale with the amount of data generated. But, will it be necessary distributed assembly like Lazer, Spaler, Ray, and SWAP? Actually, a server with moderate resources (500 GB RAM) can deal with very large datasets. For example, the memory required to encode an Iowa prairie soil metagenomics dataset takes 345 GB memory for MEGAHIT and only 29 GB for Minia [MEGAHIT:https://academic.oup.com/bioinformatics/article/31/10/1674/177884].
If I were discussing this over a beer, maybe I could entertain a discussion for hours and hours. As BioStars is better suited for focused questions and answers, I will make just a couple of remarks:
First and foremost, your post ignores the fact that the MEGAHIT assembly is almost one order of magnitude better than the Minia assembly at several metrics.
Many papers claim what they claim so they get published. It may even be justified, but technology and science moves fast, and some 3-year old claims justified at that time may be obsolete today.
As sequencers move to longer and better (less errors) reads, assembly will become an easier problem (some people even think it will be trivial in the near future), and there won't be much need for special hardware.