NVIDIA Ising: The AI Platform That Could Make Quantum Computers Actually Useful

Quantum computing has spent decades promising to reshape the world. NVIDIA just gave it a new operating system.

When NVIDIA CEO Jensen Huang told a CES audience in 2023 that useful quantum computers were still 15 to 30 years away, the quantum industry winced. Three years later, the same company has launched what may be the most important piece of quantum infrastructure to date — and it's not a quantum computer at all.

Meet NVIDIA Ising: the world's first family of open-source AI models built specifically to make quantum hardware work better, faster, and at scale. Announced on April 14, 2026, Ising doesn't compete with the quantum machines built by IonQ, IBM, or Google. Instead, it sits at the control layer, using artificial intelligence to solve the two engineering problems that have kept quantum computers stuck in the lab.

Two Problems, Two Models

Every quantum processor faces the same pair of bottlenecks. First, the qubits — the fundamental units of quantum information — need constant, painstaking calibration. A process that can consume days of expert labor every time a system drifts out of tune. Second, quantum computers make errors. Lots of them. Correcting those errors in real time, fast enough to keep calculations on track, has been one of the field's hardest unsolved challenges.

Ising attacks both problems simultaneously with two purpose-built AI models.

Ising Calibration is a 35-billion-parameter vision-language model trained on real measurement data from quantum processors spanning every major hardware type — superconducting qubits, trapped ions, neutral atoms, quantum dots, and even electrons on helium. Feed it the multi-modal output of a quantum processor, and it autonomously diagnoses what needs adjusting and recommends calibration steps. Work that once took days now takes hours. The model integrates with NVIDIA's NeMo Agent Toolkit, enabling fully autonomous calibration workflows that run without human intervention.

Ising Decoding tackles quantum error correction with a pair of compact 3D convolutional neural networks. The fast variant, at roughly 912,000 parameters, delivers 2.5 times the speed of pyMatching — the current industry standard decoder — while simultaneously improving accuracy. The more precise variant, at 1.79 million parameters, achieves a 1.53-times improvement in logical error rate. At projected scale — 13 NVIDIA GB300 GPUs running FP8 precision across 1,000 error correction rounds — the system reaches 0.11 microseconds per round. That's potentially fast enough for real-time fault-tolerant quantum computing, a milestone the field has chased for years.

Benchmarks That Turn Heads

NVIDIA didn't just release the models — it introduced a new benchmark to prove they work. QCalEval, a semantic scoring framework for quantum calibration, measures performance across six dimensions. On this benchmark, Ising Calibration outperforms Gemini 3.1 Pro by 3.27 percent, Claude Opus 4.6 by 9.68 percent, and GPT 5.4 by 14.5 percent. It's a purpose-built model outrunning the world's most capable general AI systems on a task none of them were designed for — a pattern that has become familiar in the age of specialized AI.

The decoding benchmarks are equally compelling. NVQLink, NVIDIA's dedicated QPU-to-GPU interconnect announced in late 2025, delivers a maximum round-trip latency of 3.96 microseconds over standard networking hardware, giving the decoder the speed link it needs to keep pace with real quantum error rates.

Open Source, Strategic Intent

In a move that mirrors its approach with Nemotron, Cosmos, and BioNeMo, NVIDIA is releasing Ising under permissive open licenses. Model weights are available on Hugging Face and NVIDIA's NGC Catalog under the NVIDIA Open Model License. The training framework ships under Apache 2.0. Deployment runs through NVIDIA NIM microservices. Cookbooks, benchmark datasets, and quantization recipes are all included.

The strategy is familiar: give away the software to lock in the hardware ecosystem. If every quantum lab in the world calibrates and decodes with Ising, they'll need NVIDIA GPUs — Grace Blackwell, Vera Rubin, or DGX Spark — to run them. It's the CUDA playbook applied to quantum computing.

The Industry Takes Notice

The launch roster reads like a who's-who of quantum research. On the calibration side: Atom Computing, IonQ, IQM Quantum Computers, Infleqtion, Q-CTRL, Harvard SEAS, Fermi National Accelerator Laboratory, and Lawrence Berkeley National Laboratory, among others. For decoding: Cornell University, Sandia National Laboratories, UC Santa Barbara, the University of Chicago, and more. Over 20 institutions are already building on the platform.

The stock market responded accordingly. IonQ shares jumped 20 percent and D-Wave climbed 22 percent following the announcement — a signal that investors see Ising not as a threat to quantum hardware companies but as a catalyst for the entire sector. Analysts at Resonance project the quantum computing market will exceed $11 billion by 2030, and platforms like Ising could accelerate that timeline.

Why It Matters

"AI is essential to making quantum computing practical," Jensen Huang said at the launch. "With Ising, AI becomes the control plane — the operating system of quantum machines."

That framing is deliberate. NVIDIA is not building quantum processors. It's building the intelligence layer that every quantum processor will need to function at scale. The analogy to CUDA — which quietly became the default substrate for all AI compute — is hard to miss.

Ising's hardware-agnostic design is perhaps its most underappreciated feature. By training on data from superconducting, trapped ion, neutral atom, and quantum dot systems, NVIDIA has built a tool that works across the fragmented quantum landscape rather than betting on a single winner. Whichever quantum modality eventually dominates, Ising is positioned to be part of the stack.

For quantum computing, the path from laboratory curiosity to practical technology has always run through two obstacles: keeping qubits calibrated and keeping errors in check. NVIDIA Ising doesn't eliminate those obstacles, but it brings AI to bear on them at a scale and speed that wasn't previously possible. If quantum computing's moment is approaching, NVIDIA just made sure it will arrive running on GPUs.

---

NVIDIA Ising is available now on Hugging Face, NVIDIA NGC Catalog, and build.nvidia.com. Model weights, training frameworks, and benchmark tools are open-source.