AI takes control of quantum computers

NVIDIA has announced the launch of Ising, the world’s first open-source family of artificial intelligence models specifically designed to accelerate the development of practical quantum computers. By addressing two of the biggest bottlenecks – quantum processor calibration and quantum error correction – NVIDIA Ising represents a major step toward making quantum systems more reliable and scalable.

Quantum computing has long promised transformative capabilities, but scaling systems to real-world usability remains constrained by fragile qubits and high error rates. NVIDIA positions AI as the essential control layer that can stabilize and optimize these systems.

According to NVIDIA founder and CEO Jensen Huang, AI is essential to making quantum computing practical. He described the Ising models as turning AI into the “control plane – the operating system of quantum machines,” enabling more reliable and scalable quantum-GPU systems.

The Ising model family introduces two core components:

1. Ising Calibration: A 35-billion-parameter vision-language model (VLM) capable of interpreting quantum processor measurement data (including experimental plots) and automating calibration workflows, often in conjunction with AI agents. Tasks that previously required days can now be completed in hours. It outperforms other models on a dedicated quantum calibration benchmark (QCalEval).
2. Ising Decoding: A pair of 3D convolutional neural networks (approximately 0.9M and 1.8M parameters) optimized for real-time quantum error correction as a pre-decoder for surface codes. These models deliver up to 2.5× faster performance and up to 3× greater accuracy (lower logical error rates) compared to traditional approaches such as pyMatching (when used in a hybrid setup).

Together, these tools address two of the most critical bottlenecks in quantum system development – precision calibration and error mitigation.

NVIDIA Ising is released under a permissive open-source license (NVIDIA Open Model License), allowing researchers and enterprises to train, fine-tune, and deploy the models using proprietary data while maintaining full control over their infrastructure.

The models are already being adopted by a wide ecosystem of organizations, including leading research institutions, national laboratories, and quantum technology companies (e.g., Fermi National Accelerator Laboratory, Lawrence Berkeley National Laboratory, IonQ, Atom Computing, Harvard, and others).

Ising integrates seamlessly into NVIDIA’s broader quantum computing platform, including the CUDA-Q software environment and NVQLink hardware interconnect. The company is also providing pre-trained models, training datasets, documented workflows (“cookbooks”), and NVIDIA NIM microservices to simplify deployment and customization.

This full-stack approach enables hybrid quantum-classical computing systems, where GPUs and quantum processors operate in tandem for real-time control and computation.