Nvidia Merges Supercomputers and Quantum Computers with NVQLink

Metaverse Planet November 19, 2025Last Updated: November 19, 2025

3 minutes read

Nvidia has introduced the NVQLink platform, which could revolutionize quantum computers. Thanks to NVQLink, classical supercomputers and quantum computers will be able to work together.

Although quantum computers have made great progress in the last decade, their practical use cases are still quite limited today. On the other hand, classical supercomputers, despite their power to process massive data sets, struggle with certain problems that require quantum advantage. This creates a need for hybrid architectures that combine the power of both. It seems Nvidia will be the company to meet this need. The new NVQLink platform, announced by NVIDIA at the International High Performance Computer Fair, aims to be the solution to this search for hybrid architecture.

Described by Nvidia CEO Jensen Huang as the “Rosetta Stone connecting quantum and classical supercomputers,” NVQLink is designed as an open system architecture that enables quantum processors to communicate directly with high-performance GPUs. More than a dozen supercomputing centers worldwide are already preparing to adopt this architecture. One of the most notable integrations of the platform will be with Quantinuum‘s newly announced Helios quantum computer. Quantinuum announced that it will support hybrid workflows via NVQLink with NVIDIA GPUs in all its future systems.

Nvidia-Quantinuum Collaboration Achieves Record Speed in Error Correction Nvidia and Quantinuum have not only signed a new partnership; they also shared one of the most impressive technical results in quantum error correction to date. By connecting NVIDIA GH200 Grace Hopper processors to the Helios system, the two teams achieved real-time decoding of scalable quantum LDPC (low-density parity-check) codes. During this process, a response time of 67 microseconds was achieved, falling well below the 2-millisecond threshold required in the error correction loop. This was recorded as the fastest real-time error correction performance the field has reached so far. Additionally, the NVIDIA GPU-based decoder integrated into the Helios control engine increased the logical accuracy of operations by over 3% in the experiments performed. For example, in a scenario where eight logical qubits were encoded into 30 physical qubits using Bring’s code, the error rate dropped from 4.95% to 0.925% after three correction rounds. This signifies an approximate 5.4-fold improvement and is considered an important threshold for quantum computers to approach practical applications.

Behind the performance gains lies the technical infrastructure offered by NVQLink. The system offers a total of 40 petaflops of processing power at FP4 precision used in AI operations; this means 40 quadrillion operations per second, allowing GPUs to process data from the quantum computer almost instantly. Furthermore, the data line between the GPU and the quantum processor (QPU) has a very high bandwidth of 400 gigabits/second. This makes it possible to transfer massive amounts of data between the two systems very quickly. The latency in data transmission is under 4 microseconds, meaning less than four millionths of a second. This level of low latency is of critical importance for applications requiring millisecond precision, such as real-time error correction.

Additionally, NVQLink‘s use of RDMA over Converged Ethernet (a type of advanced network technology providing direct memory access) allows researchers to easily scale classical GPU resources as the system grows. In other words, as quantum processors expand with more qubits and complex circuits, the GPU power on the classical computer side can be increased at the same rate. Thus, the hybrid quantum-classical architecture continues to work without losing performance as the hardware scales.

The results demonstrated by Nvidia and Quantinuum show that this hybrid approach, which remained in the theoretical stage for a long time, is now ready to move out of laboratory experiments and integrate into enterprise and scientific infrastructures. If this trend continues at the same pace, the process of quantum computers generating practical value may happen much sooner than expected.