Nvidia unveils Ising models

Quantum computers are controlled by ordinary computers, and Nvidia said on April 14 it is releasing open artificial intelligence models to handle two of the hardest jobs: tuning the chips and catching errors. (nvidianews.nvidia.com) A quantum processor has to be calibrated, which means adjusting thousands of control settings so fragile quantum bits behave as expected. Nvidia said its new Ising family starts with a calibration model and a decoding model, and it is making them available as pre-trained open models with retraining and fine-tuning tools. (nvidia.com) Decoding is the step where software reads noisy measurement signals and infers what errors happened before they spread. Nvidia said Ising Decoding runs up to 2.5 times faster and is up to 3 times more accurate than traditional approaches, while Ising Calibration is aimed at cutting tuning work from days to hours. (nvidianews.nvidia.com) (markets.businessinsider.com) That focus reflects where the quantum industry is stuck in 2026: hardware exists, but keeping it stable long enough to do useful work remains difficult. Nvidia’s quantum push has centered on the classical side of the stack, where graphics processing units simulate devices, steer experiments and process correction data in real time. (nvidianews.nvidia.com 1) (nvidianews.nvidia.com 2) Nvidia has been building that position for more than a year. In March 2025, it said it would open a Boston research center for quantum computing, and in November 2025 it introduced NVQLink, an architecture for linking quantum processors to graphics processing units in accelerated quantum supercomputers. (nvidianews.nvidia.com 1) (nvidianews.nvidia.com 2) The name Ising comes from the Ising model, a physics model used to describe interacting spins in magnetic materials. Nvidia and industry coverage said the new release is not a chatbot-style system but a set of task-specific models for calibration and quantum error correction. (datacenterdynamics.com) (nvidia.com) Nvidia said the first calibration model has 35 billion parameters, a scale more often associated with large language models than with laboratory control software. Sam Stanwyck, Nvidia’s director of quantum product, said the company picked calibration and decoding first because they are the most immediate bottlenecks in scaling quantum systems. (abit.ee) (siliconangle.com) Nvidia is packaging the models as part of a broader argument that artificial intelligence will be the control layer for future quantum machines. Whether that claim holds will depend on adoption by hardware makers and researchers, but the company is now offering the models, data guidance and deployment tooling rather than only chips and interconnects. (nvidia.com) (msn.com) For now, Nvidia is betting that useful quantum computing will arrive through a hybrid system: quantum chips doing the fragile physics, and artificial intelligence models doing more of the constant tuning and cleanup around them. (nvidianews.nvidia.com)