> there are some use cases that could tolerate it
Yes, and those usecases that "could tolerate it" will be a super-minority of usecases compared with usecases that "prefer moon-local latency".
Specifically for moon-local activities:
For example, the moon will likely host the largest telescopes and other measurement devices. Think JWST but 10^5x bigger! Including other measurement devices important for deep space exploration.
It'll be significantly cheaper to use AWS' moon-region (with 3 AZs of redundancy) for storage and processing, rather than shipping the data back to Earth for initial processing. Earth based users of the data can get local read-replicas of the post-processed data, with the ability to request transfer of the raw files.
It will also likely host the largest hadron collider, fusion research lab, etc. Anything that requires lots of land, few humans, and would otherwise disproportionately "impact the environment" or "cost too much based on land value" will likely be hosted on the moon. Additionally, anything like anti-matter research that might be considered "too risky for Earth".
Including ofcourse the moon wide Iron Dome's processing needs to protect the upcoming infrastructure from asteroids.
Additionally, coordination of large scale robotics on the moon for infrastructure creation will also be cheaper to host storage and processing on the moon.