A team of researchers from IonQ, Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee has demonstrated that generative AI can design quantum circuits directly, eliminating the time-consuming trial-and-error process that has limited quantum computing's practical use. The breakthrough could accelerate the path toward commercially viable quantum optimization applications.

What Problem Does This Solve in Quantum Computing?

Quantum optimization works by breaking large problems into smaller pieces, solving each one with a quantum circuit, and recombining the results. Traditionally, each circuit required extensive parameter tuning: researchers would run a circuit, measure the results, adjust settings, and repeat this cycle hundreds of times. As problems grew larger, this tuning process became prohibitively expensive in terms of computational time.

The researchers trained a transformer model, the same type of neural network architecture that powers large language models (LLMs), to learn what high-quality quantum circuits look like. Instead of showing the model how to tune circuits through trial and error, they first ran the conventional method across many sample problems, kept only the best-performing circuits, and used those examples to train the AI.

How Much Faster Is the New Approach?

The performance difference is dramatic. Using the traditional method, circuit-generation time increased from approximately 34 seconds on 4 qubits to more than 11 minutes on 12 qubits. The generative AI approach maintained circuit-generation time at roughly 28 seconds across all tested problem sizes. This represents a significant speedup, particularly as problem complexity increases.

On a benchmark problem with 100 decision variables, the AI-generated circuits roughly doubled in solution quality as subproblem size increased, demonstrating that the approach maintains effectiveness while dramatically reducing computational overhead.

How to Understand the Technical Breakthrough

• Generative Circuit Synthesis: Instead of iteratively tuning parameters through repeated runs and measurements, the trained AI model directly generates candidate quantum circuits without the repetitive loop, sampling ten candidate circuits for each subproblem and selecting the best-scoring option.
• Transformer Architecture Application: The researchers adapted transformer models, the neural network class behind ChatGPT and other large language models, to work with quantum circuit instructions instead of text, showing that the architecture's pattern-recognition capabilities transfer to quantum optimization.
• Controlled Comparison Environment: All experiments were simulated on NVIDIA H200 GPUs using identical infrastructure for both methods, ensuring a fair comparison that isolates the benefit of generative synthesis from other variables.

"Better answers in hybrid quantum optimization have traditionally come with a steep tuning tax. In this benchmark, generative AI replaced the iterative tuning loop, and as the quantum subproblems grew the solution quality improved," said Dr. Martin Roetteler, Vice President of Quantum Applications Research and Development at IonQ.

Dr. Martin Roetteler, Vice President of Quantum Applications R&D at IonQ

Why Does This Matter for Quantum Computing's Future?

Quantum computing has long promised transformative capabilities for optimization problems in finance, drug discovery, materials science, and logistics. However, the practical barrier has been the time and cost required to prepare quantum circuits for real-world problems. By replacing this bottleneck with a fast, AI-driven approach, researchers are removing a major obstacle to scaling quantum optimization to commercially meaningful problem sizes.

The study was conducted entirely through simulation rather than on actual quantum hardware, which means the results represent benchmark-scale validation. However, the researchers emphasized that this work provides a potential path toward scaling hybrid quantum optimization and unlocking capabilities that align with current and future quantum computing hardware generations.

"This work brings generative AI, quantum computing, and high-performance computing together to tackle large-scale, complex optimization problems. AI can become a new computational layer for quantum circuit synthesis, enabling the automatic design and optimization of quantum circuits for increasingly complex problems," explained Dr. In-Saeng Suh and Dr. Seongmin Kim from the National Center for Computational Sciences at Oak Ridge National Laboratory.

Dr. In-Saeng Suh and Dr. Seongmin Kim, National Center for Computational Sciences, Oak Ridge National Laboratory

The paper was presented at IEEE Quantum Week 2026 in Toronto and won a best paper award. It represents one of nine IonQ papers accepted to the conference, reflecting the growing momentum in quantum-AI hybrid approaches. The research demonstrates that generative AI is not limited to language and image tasks but can accelerate scientific and engineering workflows where traditional optimization methods have hit computational walls.