Who will control the classical side of quantum computing? NVIDIA’s CUDA‑Q/NVQLink vs AMD and Xanadu’s Backline
The companies that usually come to mind as the stars of quantum computing are QPUQPUQuantum Processor / Quantum Processor / Quantum Processing Unit / QPUThe central part of the hardware that houses qubits and performs quantum computational operations such as quantum gates and measurements.QI NoteThe performance of a QPU cannot be judged by the number of qubits alone. Gate fidelity, connectivity, speed, error rates, and other factors must be considered together. makers such as IBM, Google, Quantinuum, IonQ, QuEra, and Xanadu.
However, a practical quantum computer cannot operate on QPUs alone.
QECQEC量子誤り訂正 / Quantum Error Correction / QECA technique that distributes information across multiple physical qubits and detects and corrects errors without directly disturbing the quantum state.QI NoteSimply implementing it does not automatically provide practical fault tolerance. What matters is whether the logical error rate is improved relative to the physical error rate., decoding, calibration, hybrid computation. As we move toward FTQCFTQCFault-Tolerant Quantum Computing / FTQCA method for future large-scale quantum computing that uses quantum error correction to allow correct computation to continue even when physical errors occur.QI NoteA demonstration of quantum error correction is not the same as realizing FTQC. Logical error rates, the number of physical qubits required, logical gate performance, and so on are important., the importance of low‑latency feedback loops between QPUs and classical machines grows.
A prominent company aiming to capture this “classical side” of quantum computing is NVIDIA (see examples: IonQ, Infleqtion, Quantum Machines, and others).
On the other hand, in September 2026 AMD and Xanadu announced Backline, a low‑latency interconnect for quantum and classical hardware. While the goals are similar, the approaches differ significantly.
NVIDIA isn’t building QPUs
NVIDIA itself does not build QPUs. What it is targeting is the surrounding infrastructure.
The company’s concept of “accelerated quantum supercomputing” treats QPUs not as external devices to GPU supercomputing but as part of the same computational platform.
At the center of this vision is the quantum‑classical hybrid platform CUDA‑Q.
In October 2025 NVIDIA also announced NVQLink, a low‑latency connection between GPUs and QPUs. NVQLink is not merely a link for sending quantum jobs to a QPU; it targets calibration, QEC decoding, and orchestration of logical quantum computation as well. The reference platform currently specifies up to 400 Gbps between GPU and QPU and sub‑4‑microsecond latency for FPGA→GPU→FPGA paths.
As QPUs scale, classical compute becomes essential
This vision already extends into QEC.
On Quantinuum’s Helios, a configuration has been demonstrated that connects NVIDIA GPUs via NVQLink to perform real‑time QEC decoding. NVQLink adoption is also spreading to supercomputing centers including U.S. national laboratories.
In April 2026 NVIDIA also unveiled a suite of AI models for QPU calibration and QEC called “NVIDIA Ising.” NVIDIA is not merely offering GPUs as compute resources but is incorporating algorithms to control and maintain QPUs into its own stack.
NVIDIA’s reach goes beyond GPU‑QPU connectivity. With the addition of CUDA‑Q Logical, the stack now spans from applications down through QEC, real‑time processing, connectivity, and classical compute.
| Layer | NVIDIA’s role |
|---|---|
| Quantum application | CUDA‑Q |
| Logical quantum computation | CUDA‑Q Logical |
| Quantum error correction | CUDA‑Q QEC, GPU decoding |
| Real‑time processing | cudaq‑realtime |
| GPU‑QPU connectivity | NVQLink |
| Classical compute | NVIDIA GPUs, Grace Blackwell, etc. |
| QPU | Various quantum hardware companies |
A different answer from AMD and Xanadu
NVIDIA is not the only player aiming for the classical side of quantum computing.
On September 10, 2026, AMD and Xanadu announced Backline.
Backline is a low‑latency communication and execution framework between quantum and classical hardware that is integrated into Xanadu’s open‑source quantum software PennyLanePennyLanePennyLaneAn open-source quantum software library developed by Xanadu, used for building and executing quantum circuits and for quantum machine learning.QI NoteNot specific to any particular quantum hardware; it can interface with multiple devices and simulators. For demonstrations, check which backend was used..
AMD combines EPYC and Threadripper CPUs, Instinct GPUs, Versal FPGAs, Pensando networking, and other components with QPUs, routing workloads to the most appropriate classical compute resource.
Functionally, it resembles CUDA‑Q plus NVQLink, but the design philosophy differs. Xanadu positions Backline as an open, hardware‑agnostic design and emphasizes avoiding vendor lock‑in.
| NVIDIA | AMD + Xanadu | |
|---|---|---|
| Core software | CUDA‑Q | PennyLane / Backline |
| Real‑time connectivity | NVQLink | Backline |
| Classical compute resources | Centered on GPUs | CPUs, GPUs, FPGAs, etc. |
| QEC | Actively uses GPU decoding | Routes to the most appropriate compute resource per workload |
| Logical quantum computation | CUDA‑Q Logical | PennyLane‑centric architecture |
| Design philosophy | NVIDIA‑centric integrated stack | Open heterogeneous compute platform |
| Hardware | Tightly coupled with NVIDIA | Emphasizes hardware‑agnosticism |
In internal benchmarks, Backline reports end‑to‑end latencies of 2.3 microseconds for a CPU path using AMD Threadripper Pro plus FPGA, and 4.4 microseconds for a path using an AMD Instinct MI210 GPU.
In other words, depending on the workload, a general‑purpose CPU can sometimes yield lower latency than a GPU. This shows that the classical side of quantum computing is not simply a “GPU = classical compute” story.
Will QPU vendors become accelerator suppliers?
If NVIDIA’s strategy succeeds, an interesting industry structure could emerge.
At the time NVQLink was announced, NVIDIA had already brought 17 quantum technology companies and 9 research institutions into its ecosystem. Since then, adoption has expanded to supercomputing centers worldwide.
In the future,
CUDA‑Q
↓
NVIDIA GPU / NVQLink
↓
Various QPUs
could become a common architecture.
If developers program against CUDA‑Q instead of QPU‑specific environments, data centers build systems assuming NVQLink, and QEC and real‑time processing gravitate toward NVIDIA, QPU vendors could be positioned more as suppliers of “quantum accelerators” that attach to NVIDIA’s compute platform.
While partnering with NVIDIA could bring major benefits to QPU vendors, it could also create future risks of platform dependence.
Can AMD + Xanadu prevent a quantum CUDA‑style lock‑in?
In this respect, it is noteworthy that Backline emphasizes being “open,” “hardware‑agnostic,” and avoiding “vendor lock‑in.”
PennyLane originally evolved as quantum software that does not depend on any specific QPU.
Backline pushes that abstraction further down the stack to treat CPUs, GPUs, FPGAs, networks, and QPUs as a single heterogeneous compute system.
That said, it is premature to evaluate NVIDIA and AMD + Xanadu as equals today.
CUDA has an enormous developer base and a dominant GPU market position, and NVIDIA has already brought QPU vendors, supercomputing centers, and national labs into its fold. Backline was released only in September 2026, and it remains to be seen whether it can build an ecosystem comparable to NVIDIA’s.
Still, it is no longer the case that NVIDIA is the only company eyeing this layer.
Players beyond NVIDIA and AMD
As AMD and Xanadu’s Backline demonstrate, it can be rational to choose between CPUs, GPUs, and FPGAs based on the workload.
For example, PFN’s MN‑Core is not designed specifically for quantum tasks, but for QEC decoding and noise estimation dominated by AI inference and matrix operations, such specialized accelerators could find a role.
Conversely, MN‑Core‑style hardware may not be suitable for real‑time QPU control that requires branching and heavy I/O.
At minimum, it is too early to conclude that the classical side of quantum computing equals the GPU market.
NVIDIA is aiming to seize leadership in quantum computing not by building QPUs themselves but by controlling the surrounding compute platform. AMD and Xanadu’s Backline present an alternative: an open heterogeneous compute platform.
When quantum computers scale up, competitiveness will be determined by more than standalone QPU performance. The question is who will bind QPUs, classical compute, QEC, and control software into a single system. The moves by NVIDIA and AMD + Xanadu show that this competition has already begun.
NETWORKSee the industry network around companies tracked by QI →Related articles
- IonQ installs Superion 256 at an NVIDIA research site — competition shifts from standalone QPUs to GPU/HPC integration
- Infleqtion partners with Riverlane on fault‑tolerant quantum computing — integrating qLDPC, Sqale, and real‑time QEC
- QuEra and HPE pursue integration of fault‑tolerant quantum computers and HPC — considering on‑premises use on a Cray platform
- Quantum Machines integrates physical qubits, PPUs, GPUs, and CPUs under CUDA‑Q — NVQLink achieves ~1 microsecond round‑trip
- What Infleqtion’s “5× improvement” in QEC means — reading recent low‑latency, low‑overhead, high‑rate code advances
- Xanadu and AMD unveil Backline — low‑latency quantum‑classical integration and the move from standalone QPUs to system‑level competition
