Even if you can calculate the layout for a very large DAG in a reasonable time, the zooming and scrolling required as a user to make sense of it and follow interesting paths makes it unwieldy in my experience. One approach to use with very large DAGs is to interactively expand / collapse nodes - for a simple tree example imagine something like a treeview control for browsing a file system, expanding folders of interest to drill down rather than viewing all expanded. For non-tree DAGs you need the layout algorithm and you might need to automatically collapse unrelated sections as you expand nodes along a path to keep the time reasonable.

This approach can be combined with lazy loading data as you expand a node as another performance enhancement.

With some effort you can add this interactivity to vis.js or d3. I have previously used a commercial product called KeyLines [1] which has this feature, uses WebGL to run the layout in gpu.

[1] https://cambridge-intelligence.com/keylines/layouts/