AI for QC
Five use cases for classical AI to accelerate the development and operation of quantum computers
Classical AI technologies play an important role in facilitating current quantum computing (QC) development and operation. Ongoing technical research is showing how classical AI may play an even more important role in the QC market in the future. As a recent example, NVIDIA launched “Ising” [1] in April 2026 as a family of open-source AI models to address different engineering challenges in quantum computing.
The use of classical AI technologies to enhance QC is typically called “AI for QC” and while some stakeholders have included AI for QC under the quantum machine learning (QML) umbrella, the two concepts are very distinct. QML is the use of QC systems to perform AI-style computation: learning from data, automated model-driven functionality, and the absence of human-coded rules and structures. In contrast, AI for QC is the use of classical computing systems to enhance the development or use of QC technology.
There are five main types of AI for QC use cases at present:
1. Calibration and control
2. Application development
3. QEC decoding
4. Algorithm discovery
5. New material discovery
Calibration and control
The benefit of classical AI in the calibration and control process is in enabling systems to scale up to very large sizes and to improve system performance over what can be achieved manually.
Calibration and control are core functions of “bringing up” (i.e., setting up a QC for operation) and operating a QC on an on-going basis. Calibration involves characterizing the behavior of individual elements in the QC, such as its qubits and couplers. QCs are incredibly sensitive to factors such as manufacturing variability and environmental conditions. Individual qubits, for example, may respond differently to the same microwave control pulse signal. Calibration determines these behavioral differences and how to optimize the QC program to account for this variability.
Control involves shaping the microwave (or optical) signal to optimally implement the ongoing set of gates needed to turn a software program into a physical set of actions on the QC. Importantly, QCs suffer from “drift” in operating performance over time, due to changing environmental conditions. Control systems need to account for this drift and re-calibrate control signals in real time.
Traditionally, calibration and control are done manually, with PhD-level staff running experiments to characterize QC behavior and plan control pulse sequences. Brandon Severin, the co-founder and CEO of Conductor Quantum, has described [2] calibration for a single qubit taking up to months during his doctoral studies at Oxford. Scaling up to hundreds of thousands or millions of qubits will require finding automated ways to calibrate and control QCs.
Beyond scaling, companies such as Q-CTRL are using AI to calibrate and control QCs at a level of robust operations beyond what has been achieved even by expert manual operation. For example, the company claims [3] that in a “tuneup challenge” conducted at an APS Global Physics Summit, Q-CTRL’s autonomous calibration and control software, in a competition against an expert audience of physicists, provided a calibration that was 42x more error resistant than the average “manual” gate tuneup, and was completed 3x faster.
Application development
Similarly to the growing use of classical AI in classical software development (i.e., “vibe coding”), classical AI is also being integrated into QC-specific SDKs (software development kits), including:
IBM Quantum’s Qiskit (Qiskit Code Assistant)
Xanadu’s Pennylane (PennyCoder)
Microsoft Quantum’s QDK (GitHub Copilot)
NVIDIA’s CUDA-Q (from partners, such as Conductor Quantum’s Coda MCP, which enables third-party AI agents to interact with CUDA-Q)
Notably among the big five SDKs, Google Quantum AI does not seem to have developed AI-assistance capabilities to the same extent as the other four vendors.
The primary benefits of AI coding assistance are captured well in quantum software vendor Haiqu’s press release announcing [4] the introduction of their Agentic Quantum Operating System (OS): “…help enterprise R&D teams get usable results faster, spend less money per experiment, train new researchers more easily, and turn early ideas into testable prototypes faster.”
In addition to the big five platforms, innovative private vendors also provide AI assistance for QC software development. In addition to Haiqu, Classiq announced [5] the Classiq Quantum Agent that sits on top of the company’s hardware-agnostic platform. The company’s focus with the Quantum Agent is to move past “free-form code generation” to enterprise-ready development. It’s notable that vendors emphasize differing strategies for their SDK offerings. For example, comparing Classiq to IBM Quantum, the focus of Classiq is on higher-level abstraction for the developer and tighter integration of the SDK with the development environment, while IBM Quantum’s focus is more on direct gate-level exposure and control.
QEC decoding
Classical AI can help in the development of quantum error correction (QEC) decoders that are potentially faster, more efficient, and more adaptable than traditional, hand-coded decoders. QEC decoders are algorithms running on classical computers that enable fault-tolerant quantum computers (FTQCs) to combine some number of physical qubits together to form a logical qubit. Physical qubits are extremely error-prone, but spreading the informational value of a logical qubit across many physical qubits means that errors on the component physical qubits comprising the logical qubit can be detected and corrected before they destroy the information contained in the overall logical qubit.
Decoders work by taking measurements from special physical qubits (i.e., “ancilla qubits”) that are set aside in the logical qubit just for use in the QEC decoding process and using those measurements to detect and correct errors in the data qubits that contain computational information. This process may seem laborious but is necessary because the data qubits can’t be measured directly without destroying their quantum information; the decoder instead infers errors in the data qubits by taking “syndrome measurements” from ancilla qubits that have been briefly entangled with the data qubits.
A key requirement for QEC decoding is speed. The decoding cycle must happen quickly enough for errors to be corrected in real time rather than accumulating into a backlog that eventually overwhelms and destroys the overall computation. This requirement is particularly challenging because FTQC cycle times are very fast. For example, superconducting qubit FTQC cycle times must happen on the order of a single microsecond [6].
Research on classical AI-developed QEC decoders is happening both in the vendor community and at academic labs:
NVIDIA: As part of its “Ising” launch in April 2026, NVIDIA introduced [7] “Ising Decoding,” two variants of a 3D convolutional neural network (CNN) that are optimized either for speed or accuracy. The company claims that Ising Decoding models are up to 2.5x faster and 3x more accurate than current open-source industry alternatives. Partners using Ising Decoding include Cornell University, EdenCode, Infleqtion, IQM Quantum Computers, Quantum Elements, Sandia National Laboratories, SEEQC, University of California San Diego, UC Santa Barbara, University of Chicago, University of Southern California and Yonsei University.
Google Quantum AI: The company first introduced [8] AlphaQubit in 2024 to use transformer-based neural networks to perform decoding that was more accurate than at least two other, traditional methods, but at the time was too slow for practical use. More recently, Google Quantum AI announced [9] AlphaQubit 2 in December 2025. AlphaQubit 2 comes in two variants, the first of which can decode in less than a microsecond, which is fast enough for superconducting qubit-based QCs. The second decodes in about 100 microseconds, which is too slow for superconducting qubit-based QCs, but could be used for neutral atom qubit-based QCs or trapped ion qubit-based QCs.
Harvard: The university announced [10] a CNN-based decoder they call “Cascade” based on numerical simulation (i.e., it hasn’t been demonstrated on an actual QC) that the researchers claim exhibits a strong efficiency gain that they call a “waterfall” effect. The importance is that when used to process one type of QEC code called the [[144,12,12]] Gross code, Cascade required 40% less physical qubits under specific conditions than other types of decoders. This efficiency gain could mean that FTQC algorithms could be useful sooner than expected.
It’s interesting to note that AI for QC in the QEC process is not limited to decoding. For example, a new QC startup called MagiQware is developing [11] an AI approach to “magic state” production. Magic states are essentially small-angled gates needed for universal FTQC calculations that are extremely hard to produce. In practice, the useful magic state gates are distilled in the computer in a process called “magic state distillation”. In some cases, up to 90% of gate operations in a calculation are consumed just producing magic state gates (i.e., non-Clifford gates like the T gate.)
Algorithm discovery
Finding new QC algorithms, and especially FTQC algorithms, is a very important industry requirement to expand the usefulness of QC. QCs don’t process classical computing algorithms but rather require QC-specific algorithms that can take advantage of the quantum mechanical properties of QC information processing. QCs aren’t faster than classical computers. Rather, QCs offer the possibility of using quantum algorithms that need many fewer (ideally, exponentially fewer) steps to complete a given computation compared to a classical algorithm. Therefore, finding the right algorithm for the right problem is crucial to achieving “quantum advantage”.
Unfortunately, finding algorithms that can take full advantage of QCs has proven difficult. While many quantum algorithms have been discovered, only two (Shor’s algorithm for factoring, and quantum simulation) have a proven exponential speedup. Algorithm research is a deep area of focus for the quantum industry.
It may be that classical AI can speed up the search for new quantum algorithms and expand the search in directions human researchers might not have considered on their own. This remains speculative. However, an adjacent development from IBM Quantum provides a glimpse of the possibilities.
The IBM Quantum example [12] is specifically focused on the search for QEC codes (the codes that are decoded as described in the previous section). To be clear, QEC codes are classical algorithms, not quantum algorithms. However, IBM Quantum’s use of classical AI to search for members of a type of QEC code called bivariate bicycle (BB) codes, hints at how researchers might use AI more broadly for algorithm discovery.
New material discovery
Finally, classical AI can optimize the physical side of QC development, by enabling discovery of new materials that are better-suited than current options for building QCs. New material discovery focuses on screening, simulation, and synthesis. Examples include:
Screening: Mind Success launched [13] an AI-driven “Quantum Materials Discovery Platform” that it claims screens more than 150,000 candidate materials and generates density functional theory simulations to narrow down to the most promising options for a particular search. The company says in a press release that the platform is “intended to reduce the time required to identify materials suitable for applications including quantum processors, sensors and cryogenic components.”
Simulation: Microsoft Quantum in a recent blog post [14] described using agentic AI to run materials simulations that enabled the company to gain a purported 1,000x improvement in reliability comparing their Majorana 2 chip with their Majorana 1 chip by moving from aluminum to lead as a key ingredient of the chip. Zulfi Alam, a corporate vice president for quantum at Microsoft, states in the blog that, ““Finding the exact recipe, the right amount to put to get the desired energy structure, requires a lot of experimentation in the old world order. In the new world order, through simulations, you can see where the highly probable target is. And then with that knowledge, you ideally only have to experiment once.”
Synthesis: Researchers described in a recent ArXiv review paper [15] the use of AI to improve diamond synthesis, resulting in improving nitrogen vacancy (NV) magnetometry sensitivity by 300% over “average” samples. While this example is specific to quantum sensors, rather than quantum computing, it’s easy to see how this approach could translate directly into improvement in material synthesis for NV center in diamond qubit-based QCs.
Key Takeaway
Classical AI is already playing a significant role in the development and operation of quantum computers. I expect use of AI will grow strongly in the future, given the complexity of QC technology, both in its development and operation. Automating key functional processes while learning from data should speed up technical breakthroughs and time to market, produce results that may not have been possible through manual-only approaches, and make both R&D device development as well as operational systems more efficient in resources and time.
Citations
[1] NVIDIA, NVIDIA Launches Ising, the World’s First Open AI Models to Accelerate the Path to Useful Quantum Computers (2026), https://nvidianews.nvidia.com/news/nvidia-launches-ising-the-worlds-first-open-ai-models-to-accelerate-the-path-to-useful-quantum-computers
[2] B. Severin, 007: Scaling Quantum Hardware with AI, Entangled Show (2026),
[3] Q-CTRL, Q-CTRL Delivers Autonomous Calibration Solutions to Enable Rapid Integration with Ecosystem Partners QuantWare & TreQ (2025), https://q-ctrl.com/blog/q-ctrl-delivers-autonomous-calibration-solutions-to-enable-rapid-integration-with-ecosystem-partners-quantware-treq
[4] Haiqu, Haiqu Launches Agentic Quantum Operating System for Enterprise Applications R&D (2026), https://haiqu.ai/insights/haiqu-launches-agentic-quantum-operating-system-for-enterprise-applications-r-d
[5] Classiq, Classiq Certifies Expert-Level Quantum AI Agents for Real-World Applications (2026), https://www.classiq.io/insights/classiq-certifies-expert-level-quantum-ai-agents-for-real-world-applications
[6] U. Akhouri, Active Reset for Qubits: Building Qubits You Can Count On, Quantum Machines (2026), https://www.quantum-machines.co/resources/blog/active-reset-qubits/
[7] NVIDIA, NVIDIA Launches Ising, the World’s First Open AI Models to Accelerate the Path to Useful Quantum Computers (2026), https://nvidianews.nvidia.com/news/nvidia-launches-ising-the-worlds-first-open-ai-models-to-accelerate-the-path-to-useful-quantum-computers
[8] Google DeepMind and Google Quantum AI, AlphaQubit tackles one of quantum computing’s biggest challenges (2024), https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphaqubit-quantum-error-correction/
[9] Google DeepMind and Google Quantum AI, A scalable and real-time neural decoder for topological quantum codes (2025), https://arxiv.org/pdf/2512.07737v1
[10] A. Gu, et al., Scalable Neural Decoders for Practical Fault-Tolerant Quantum Computation, Harvard University and Harvard Quantum Initiative (2026), https://arxiv.org/pdf/2604.08358v1
[11] Quantum Computing Report, MagiQware Secures €575K ($658K USD) Pre-Seed Allocation to Optimize Quantum Magic State Factories via Reinforcement Learning (2026), https://quantumcomputingreport.com/magiqware-secures-e575k-658k-usd-pre-seed-allocation-to-optimize-quantum-magic-state-factories-via-reinforcement-learning/
[12] IBM Quantum, Can LLMs discover quantum error correction codes? (2026), https://research.ibm.com/blog/ai-for-qec
[13] M. Swayne, Mind Success Unveils Quantum Materials Discovery Platform, Digital Twin Software for Hardware Development, Quantum Insider (2026), https://thequantuminsider.com/2026/07/16/mind-success-unveils-quantum-materials-discovery-platform-digital-twin-software-for-hardware-development/
[14] C. Bolgar, Introducing Majorana 2, How Microsoft’s new quantum chip was made 1,000x more reliable with the help of Microsoft Discovery’s agentic AI, Microsoft Blog (Source) (2026), https://news.microsoft.com/source/features/innovation/majorana-2-microsoft-discovery-agentic-ai
[15] M. Chen, et al., When AI meets quantum information: A comprehensive review, ArXiv (2026), https://arxiv.org/pdf/2607.00365