Quantum's near-term job is structure, not search

The most useful quantum computer in 2026 is not the one that searches faster. It is the one that teaches a classical solver where to look. Near-term gate-based hardware has a job to do before it outcomputes anything, and the job is reading the problem's hidden structure, not solving the problem itself.

Rigetti and Purdue's joint benchmark on hard-constrained graph partitioning makes the new role concrete. A depth-p=1 QAOA circuit pulled two-point correlations out of the problem; the Gurobi branch-and-bound solver reshaped its objective around those correlations and hit a near-optimal threshold (ε = 0.01) on a benchmark class, with the hard constraints preserved inside the classical solver. The 100-times speedup the team reports is scoped to that benchmark, at that depth, on that problem class.

The read that misses the moment is "quantum beats classical." The read that catches it is that the chip is now a preprocessor for commercial MIP solvers, a structure-learner feeding the workhorse. The same recipe transfers anywhere a problem has hidden correlations a branch-and-bound tree ignores: delivery routing, scheduling, network design.

What stays unsettled is scale. The number is on one benchmark class, at one depth, on Rigetti's announced results, and independent reproduction at larger instances is the open question. What is settled is the mental model. The chip's new job is to read the problem, not to solve it.