The dendritic unit is better on real, published axes — parameter efficiency, robustness to overfitting, continual learning. So why does the dense point neuron win deployment? Sara Hooker's answer, in 2020, was the hardware lottery: an idea succeeds or fails partly on whether the hardware of its era happens to suit it, independent of its merit. Dendritic computation is sparse and conditional — only a few units fire per input — which on paper is far fewer multiply-adds. But GPUs are built for dense matrix multiplies and cannot skip zeros; run a sparse layer on one and it multiplies the zeros anyway, so the theoretical saving never arrives. That is why the parameter-efficient, forgetting-resistant unit is stuck, and the dense one scaled. It is not that the point neuron won the argument. It won the lottery. Here you measure the gap directly, and it names the frontier: the prize both dendritic networks and the push for continually-adapting AI are reaching for is the same, from opposite ends.