Here's the thing that keeps scaling problems interesting: the bottleneck almost always ends up being something you didn't plan for. This researcher pushed last-mile routing into the tens of thousands of stops and found the algorithm wasn't the weak link, the system architecture around it was. That's a finding worth paying attention to, because it reframes where the real work happens when you scale.
What this means in practice is that throwing a better solver at a big routing problem won't fix the friction points. Clustering by geometry alone breaks down when constraints pile up. Route optimization can become a runaway cost if you don't bound it as a discrete step. And the boundary between clusters? That's where inconsistencies live, and they multiply. The researcher also flagged the hidden tax of recomputing distances repeatedly, wasted cycles that add up fast. The surprise was near-linear scaling once these structural issues were handled. That's not how these problems usually behave, and it suggests the ceiling on large-scale logistics isn't algorithmic, it's systemic.
For anyone running logistics at scale, the takeaway is direct: pay attention to how your system is built around the solver, not just the solver itself. The clustering logic, the cost controls, the boundary handling, the caching of distance calculations, these are the levers that matter when you push past typical sizes. The researcher's experience shows that a well-structured system can behave more predictably than theory would predict. That's not a claim that every VRP problem will scale linearly. It's evidence that the system design, not the algorithm, is the decisive constraint. If you're building for scale, that's where you should look first.