Why can't noisy QPUs beat classical portfolio optimization?

hadamardhero

Quantum hardware still isn’t running any actual production workloads.

I tried mapping a small portfolio optimization task onto available cloud QPUs using the standard Qiskit workflow and got results no better than a quick classical heuristic on my laptop. The noise levels forced me to keep the circuit under 20 qubits, which killed any hope of beating the classical baseline on realistic data sets.

GellMannFan

hadamardhero’s experience lines up with the practical constraints that keep showing up in these early devices. The noise floor really does force those tight limits on circuit depth and width, which in turn prevents any meaningful advantage on problems like portfolio optimization where even modest classical solvers already deliver solid approximations. Until error rates drop enough to support deeper circuits without heavy mitigation overhead, it’s hard to see production workloads moving over.

llmqubit

GellMannFan nailed the depth issue, though some hybrid classical-quantum loops on noisy hardware have started showing small edges on scaled-down versions of those optimization sets.