IonQ and NVIDIA speed up quantum circuit compilation

Researchers from Oak Ridge National Laboratory, IonQ and NVIDIA demonstrated a generative AI approach for designing quantum optimization circuits, targeting one of quantum computing’s persistent software bottlenecks. Instead of relying on repeated variational parameter tuning, the team trained a transformer model on near-optimal circuit configurations and used it to generate candidate circuits directly.

The benchmark ran on a single NVIDIA H200 graphics processing unit inside Oak Ridge’s Defiant2 supercomputer using NVIDIA’s CUDA-Q platform and cuQuantum software development kit. In a dense benchmark problem involving 100 decision variables, traditional circuit-finding runtime increased from approximately 34 seconds to more than 11 minutes as scale rose, while the AI model held runtime at roughly 28 seconds across every qubit scale tested.

The result could improve the economics of quantum optimization for logistics, financial modeling and materials science by reducing compilation costs and making workloads more predictable. IonQ brings commercial access through Amazon’s AWS, Microsoft’s Azure and Alphabet’s Google Cloud, while NVIDIA is positioning CUDA-Q and related software as a bridge between classical data centers and quantum processors.

IonQ reported second-quarter revenue growth of roughly 286.7% year-over-year (YOY), but also posted an annual net loss of approximately $510.38 million. NVIDIA generated annual revenue of around $215.94 billion and annual net income of about $120.07 billion, giving investors a more profitable infrastructure path into the hybrid quantum market.