AT&T expands D-Wave test, but a 15-second workload is not a network rollout

The filing does not identify the workload's objective function, solution-quality threshold, classical baseline, total cost or whether the quoted time includes the surrounding data, software and integration work. Those omissions matter because an operational optimization system must produce a sufficiently good decision as well as produce it quickly.

AT&T says its existing network agentic-AI tools reduced customer downtime by 12 million hours in 2025. The filing says quantum capabilities could increase those tools' benefits, but it does not connect the 15-second workload to that downtime measure or report a quantum-related outage, routing, planning or traffic-management outcome.

D-Wave’s company-reported illustration of hybrid-solver model types and supported variable forms. Source: Scientific Reports.

Why the formulation is the real test

Annealing is specialized for optimization problems; D-Wave's annealers are not universal quantum computers that can run arbitrary algorithms. That makes routing and planning plausible candidates, but it also means an outcome from one formulation cannot establish a general advantage for network operations.

A 2025 Scientific Reports benchmark compared D-Wave's hybrid solver with CPLEX, Gurobi and IPOPT. The study found the hybrid solver most promising for binary quadratic problems, while its competitiveness diminished for non-binary, non-quadratic problems and combinations of them. Its hybrid workflow also combines classical heuristics with quantum-hardware subproblems, so its results cannot be treated as a measure of quantum hardware alone.

The same study gives a useful boundary on the claim, rather than a verdict on AT&T's undisclosed workload. In a full-scale energy unit-commitment mixed-integer linear problem, none of 50 D-Wave runs at the solver's default 20-second setting found a feasible solution. Longer settings produced feasible results, but the reported comparison favored a classical alternative:

That comparison is not transferable to AT&T: it is a power-system problem with its own constraints and a different performance measure. It does show why the 15-second figure needs a like-for-like baseline that includes both feasibility and solution quality. The study's authors also found a D-Wave advantage in their particular 500-variable binary quadratic test, where CPLEX was capped at 1,000 seconds; the evidence is mixed by problem class, not uniformly negative.

The decision AT&T still has to make

The relevant next evidence is operational rather than promotional: which workflow AT&T selects, what it is compared against, the quality and cost of the resulting decisions, and whether those decisions measurably improve outage response, routing, planning or traffic management.

AT&T is separately evaluating D-Wave's forthcoming gate-model systems for quantum security and quantum communications. That work, too, remains prospective. The nearer decision is whether the annealing test can consistently beat the best available classical approach on a disclosed network problem—not merely complete one calculation faster.