"Using your time now benefits the entire NERSC community and spreads demand more evenly throughout the year, so to encourage usage now, we are ..."
Sounds like it's heavily directed at grad students. I'll bet there's a huge queue for getting some runtime, toward the end of each semester.
The discount must be due to having some idle time on their A100 nodes, right now. If you have a limited budget of compute hours, you're going to save them until you need them, which is probably going to be skewed towards the latter half of the semester.
... why the organization does not backfill its idle capacity with commercial workloads.
If this HPC GPU supply crunch were an ongoing phenomenon, then you might see something like that. However, the agency is tasked with providing resources to scientific researchers, not with serving the commercial sector or trying to turn a profit.
In offering a commercial service, there could come additional risks, such as more exposure to hackers, which is an added problem when you consider those jobs are running along side some classified government research jobs. You'd also have to do more clearance work to ensure that your commercial customers aren't from sanctioned entities, etc.
All in all, it sounds like more of a headache than it might be worth, especially if your existing users
have the allocation of hours to use. Encouraging them to use their time now not only reduces node idle time, but perhaps more importantly reduces contention during those critical high-demand periods.