Business Wire·Sep 16, 2026, 9:54 AM ET

IonQ, ORNL, NVIDIA, and the University of Tennessee, Knoxville Show AI Method Reduces Quantum Optimization Trade Off

IONQ

IonQ, Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee, Knoxville, demonstrated that a trained generative AI model can directly write quantum optimization circuits to eliminate costly trial-and-error parameter tuning.
The research paper, presented at IEEE Quantum Week 2026 in Toronto, won a best paper award and details how the model replaces iterative tuning with fixed candidate evaluations.
The team trained a transformer model on near-optimal circuits generated by conventional methods, enabling it to sample ten candidates per subproblem and select the best score for global solution updates.
On a benchmark problem with 100 decision variables, the AI-generated approach doubled answer quality as subproblems grew while maintaining a constant runtime of approximately 28 seconds.
In contrast, the traditional state-of-the-art method saw circuit-finding time rise from 34 seconds on 4 qubits to over 11 minutes on 12 qubits.
All circuits in the study were simulated using the NVIDIA cuQuantum SDK on a single NVIDIA H200 GPU within the Defiant2 supercomputer system, ensuring a controlled comparison between workflows.
The study focuses on the distributed quantum approximate optimization algorithm (DQAOA) and its integration with generative AI to enable automatic circuit synthesis for increasingly complex problems.
Researchers plan to extend the framework to real-world scientific applications 8 scale it across larger high-performance computing systems.
The work was led by ORNL with co-authors from IonQ, NVIDIA, 8 UT, including Abhinav Rijal 8 Dr. Martin Roetteler.
IonQ's Superion quantum computers 8 previous systems have previously delivered 20x performance increases for partners including Amazon Web Services, AstraZeneca, 8 NVIDIA.