It didn’t take long for the trading desks to respond when the announcement was made on a Monday morning—World Quantum Day, of all days. A new front has been established by Nvidia, the company that currently supplies the picks and shovels of the AI gold rush.
The project, which is the first family of open-source AI models created especially for quantum computers, is called Ising, after the outdated physics model that previously assisted scientists in understanding complex systems. Analysts were rewriting their notes in a matter of hours. It appears that investors think this is more than a side project.
| Company | Nvidia Corporation |
| Founded | April 1993 |
| Founders | Jensen Huang, Chris Malachowsky, Curtis Priem |
| Headquarters | Santa Clara, California, USA |
| CEO | Jensen Huang |
| Industry | Semiconductors, AI, Accelerated Computing |
| Product Launched | NVIDIA Ising — Open Source Quantum AI Model Family |
| Launch Date | World Quantum Day, April 14 |
| Key Models | Ising Calibration (35B-parameter vision-language model), Ising Decoding (3D CNN, two variants) |
| Performance Claim | Up to 2.5x faster and 3x more accurate than pyMatching for error correction |
| Early Adopters | IonQ, IQM, Atom Computing, Harvard, Fermilab, Lawrence Berkeley National Lab, UK NPL |
| Related Platforms | NVIDIA CUDA-Q, NVQLink, NIM Microservices |
| Quantum Market Forecast (2030) | Over $11 billion (Resonance) |
Observing Jensen Huang discuss it gives the impression that he has been anticipating this. He referred to AI as the “control plane” of quantum machines, which is essentially their operating system. When you consider it, that line is quite remarkable. The fragility of quantum processors is well known. Qubits move around. They lose knowledge. Before a single useful run, engineers must manually calibrate these devices for days or even weeks. Theoretically, Ising condenses that into hours. Perhaps less.
At the center of the launch are two models. A 35-billion-parameter vision-language model called Ising Calibration receives measurements directly from the quantum hardware and makes real-time adjustments. With up to 2.5x speed and 3x accuracy improvements over pyMatching, the open source standard that the majority of the field has been relying on, Ising Decoding, which is based on a 3D convolutional neural network, manages mistake correction in real time. According to Nvidia, the decoders operate at microsecond timeframes, which is the level of detail that matters to quantum researchers.
To be honest, it’s difficult to ignore the list of early adopters. Paulson School at Harvard. Berkeley’s Advanced Quantum Testbed, Fermilab. Information, Atom Computing, IonQ, and IQM. even the National Physical Laboratory in the United Kingdom. Getting so many significant players on board at launch indicates that Nvidia has been discreetly developing something for some time. Quantum is a small, opinionated universe. It’s difficult to ignore the fact that some of these organizations seldom ever work together on anything.
The way that Nvidia’s director of quantum product, Sam Stanwyck, presented the issue resonated with me. Approximately for per thousand processes, the finest quantum processors available today malfunction. That must decrease to one in a trillion in order to be truly helpful. A huge disparity. At this point, gap is nearly impossible to close it without AI, and Stanwyck referred to calibration and error correction as “AI-shaped problems”—high-throughput, data-intensive tasks that are precisely what GPUs were designed for. Of course, convenient framing. However, it’s also true.
The market’s response was swift, if not spectacular. Early trading saw a little increase in Nvidia shares. Particularly the smaller quantum-adjacent names moved more abruptly. The question of how much of this is reflex and how much is sincere conviction is constantly present. Investors have previously been let down by quantum. Everyone who experienced the IonQ SPAC era may recall. However, this launch has a unique texture because of the connection with CUDA-Q and NVQLink, open models, and actual clients.
Whether Ising will fulfill the larger commitments is still up in the air. In a lab, decoding at code distance 31 is impressive. It is quite another matter to scale it to a fault-tolerant computer with thousands of logical qubits. Adoption will be unequal, according to Stanwyck himself, with calibration coming first and mistake correction coming later, depending on each hardware team’s actual location. In product briefings, that kind of candor is uncommon.
The openness, however, feels different here. Frameworks, training data, and refined recipes were all made public. For years, quantum teams have been developing this technology in secret silos. There’s a common floor now. Over the following few quarters, it will be interesting to see if that floor holds or if it turns into the foundation that everyone silently builds upon. Tesla once encountered similar skepticism, and you can see how it turned out.