Recurrent depth is bullish and represents efficient scaling.
Current workloads and systems are memory-limited.
RD shifts some of that to compute which unlocks incremental serving capacity per node.
This is a relative benefit for compute providers like $NVDA, and helps
