Nvidia launches CUDA-Q Logical for fault-tolerant quantum processors

CUDA-Q is a software environment used to build hybrid applications that span classical and quantum architectures. Nvidia said developers use it to orchestrate workloads across traditional processors such as CPUs and GPUs, as well as quantum processing units, or QPUs. The platform is one of the most widely used quantum programming environments, but Nvidia said the industry's move toward fault-tolerant QPUs has created significant problems for developers.

Fault-tolerant systems rely on logical qubits, which are groups of physical qubits that work together to correct errors and prevent calculations from being corrupted. Nvidia said writing software for logical qubits is extremely difficult because the error-correction code they use changes the underlying physical resources required to execute an application, disrupting system planning and execution.

CUDA-Q Logical is intended to address that by giving researchers a programmable and verifiable environment to simulate and test fault-tolerant quantum computing systems. It lets developers model different algorithms, error-correction techniques and QPU architectures side by side, so they can identify optimal configurations before committing to hardware.

Timothy Costa, Nvidia's vice president and general manager of quantum, said quantum computing is maturing rapidly with the 'era of logical qubits,' and researchers need an open, customizable programming platform that represents all aspects of fault-tolerant systems. He said the addition of CUDA-Q Logical provides power and flexibility to explore fully integrated, co-optimized systems regardless of qubit type and architecture, and 'drastically shortens the timeline to useful quantum-GPU supercomputing.'

Nvidia cited early work by Iceberg Quantum, an Australian quantum startup that designs fault-tolerant architectures. Iceberg Quantum used CUDA-Q Logical to model a new architecture for silicon-based qubits developed by Diraq Pty Ltd. Its model showed it should be possible to create 1,000 logical qubits from just 150,000 physical qubits, 10 times less than Diraq originally estimated.

Fermi National Accelerator Laboratory also used CUDA-Q Logical to evaluate new error-correction strategies, runtime requirements and algorithms across multiple quantum computing architectures. Nvidia said Fermilab researchers reduced the average development cycle of fault-tolerant algorithms from five months to three weeks, roughly seven times faster than before.

Anna Grassellino, Fermilab's chief technology officer, said fault-tolerant quantum computing is the path to unlocking new scientific discovery, but reaching it will require researchers to co-design algorithms, error correction, architectures and hardware together. She said the team used CUDA-Q Logical to explore combinations of those resources in just three weeks, compared with what would typically take about five months of building specialized infrastructure.

Nvidia also highlighted broader adoption of its Quantum-GPU Supercomputing Platform, a cloud service that combines high-performance GPUs with quantum processors. The service is designed for experimental applications and workloads best solved by a combination of classical and quantum computers, according to SiliconANGLE.

Diraq, which designs quantum computers using spin qubits based on modified silicon transistors, is using Nvidia's open-source Ising models to calibrate its processors. Anyon Computing and Quantum Machines are using Nvidia's NVQLink networking technology to connect QPU clusters directly to GPU supercomputers. BlueQubit, IonQ Inc. and Qedema Quantum Computing have integrated their tools with CUDA-Q to deploy hybrid-quantum workloads in production, Nvidia said.

Editor's Summary Nvidia's CUDA-Q Logical adds simulation and testing capabilities for fault-tolerant quantum systems, with early users reporting faster development cycles. The launch at IEEE Quantum Week 2026 deepens Nvidia's role in hybrid quantum-classical computing as quantum hardware vendors and research labs adopt its GPU-QPU software and networking tools.