I agree Arrow by itself doesn't address any novel/fundamental OLTP challenges, but I'm not arguing that the thing which eventually supplants Postgres will succeed because of best-in-class OLTP performance - anyone needing that today is not choosing Postgres anyway (same as ever).
The real proposition is having a modern, general purpose workhorse underpinned by an ecosystem with strong network effects and polyglot APIs. Assuming analytical systems continue to gravitate towards Arrow I believe the OLTP world will be dragged along also.