From Qubits to Workflows: Rethinking Quantum Computing - eetimes.com

The quantum computing industry has spent a great deal of time debating qubits. How many qubits can a system support? How quickly can errors be corrected? When will fault-tolerant systems arrive? Which qubit type gets us there fastest and at scale? Which modality will ultimately win?These are important questions, but they may not be the most important questions regarding the future of quantum computing. IBM has spent much of 2026 making a different argument. Rather than treating the quantum computer as an isolated system, IBM is building an architecture where quantum processors become another specialized computing resource alongside CPUs, GPUs, and other accelerators.IBM calls the architecture quantum-centric supercomputing.The name sounds like marketing language, but the underlying concept is important. A workload is divided among different computing resources with each performing the portion of the problem it handles best. CPUs can handle general-purpose processing and orchestration. GPUs can accelerate highly parallel classical computations. Quantum processing units, or QPUs, can execute portions of a problem where quantum algorithms provide an advantage. AI can increasingly help manage, optimize, and even orchestrate those workflows.This is less a story about quantum computers replacing classical computers than about quantum becoming part of computing.IBM formalized that idea earlier this year when it published a reference architecture for quantum-centric supercomputing. The architecture describes QPUs working alongside CPUs and GPUs across supercomputing centers, research facilities, and cloud environments. More importantly, it treats the combined workflow rather than any individual processor as the computing system.Since then, IBM and its research partners have provided increasingly interesting evidence for that approach.In July, IBM and researchers at the University of Chicago announced a demonstration they described as quantum advantage, using logical quantum circuits to perform a computation beyond the practical reach of leading classical simulation methods while also providing a way to establish trust in the result. Other IBM collaborations with Qedma and Algorithmiq reported additional quantum advantage results involving problems where leading classical approaches could no longer provide consistent answers.Quantum advantage has been one of the industry’s most anticipated milestones. But there is an important architectural detail behind IBM’s approach to it. IBM’s own 2026 roadmap describes quantum advantage in the context of a quantum computer working with high-performance computing. The company is developing an integrated environment where quantum and classical code can be deployed as parts of the same workload.In other words, even as quantum hardware crosses important thresholds, the larger system remains heterogeneous. The history of computing suggests this should not be surprising. GPUs did not replace CPUs. AI accelerators did not replace GPUs. Instead, each became another specialized resource within an increasingly heterogeneous computing architecture.AI inference is already accelerating this transition. Modern inference infrastructure can distribute different portions of a workload across CPUs, GPUs, networking processors, and other accelerators. Data preparation, model execution, retrieval, memory management, orchestration, and agentic workflows do not necessarily run on the same processor. What users experience as a single AI application can actually be a workflow spanning numerous specialized computing resources.Quantum appears to be moving in the same direction. A good example comes from research involving IBM, Oak Ridge National Laboratory, and Cleveland Clinic. In July, researchers reported the first-known quantum computer calculations of molecular configurations for a material relevant to producing tritium fuel for fusion energy.The story was not simply that a quantum computer was involved. Researchers used quantum-centric supercomputing techniques to divide the problem between quantum and classical computing resources. Quantum systems were used to calculate portions of the electronic structure problem while classical systems supported the broader scientific workflow. IBM described the approach as combining quantum computing, AI, and exascale computing to attack a problem that becomes difficult for classical systems working alone. And interestingly, the quantum computer was not the computer. It was one part of the computer.The same pattern is becoming more apparent in IBM’s work with Cleveland Clinic and RIKEN. Earlier research from the group used IBM Quantum Heron processors alongside classical supercomputing resources to simulate biologically relevant molecular systems containing as many as 12,635 atoms. That work has since been named a finalist for the 2026 ACM Gordon Bell Prize, one of the most prominent awards in high-performance computing.But even more interesting developments came with the group’s subsequent work. Researchers developed a fully automated end-to-end workflow running across quantum and classical computing resources. Tasks that previously required significant manual coordination and movement of data between systems could instead be orchestrated as part of a larger workflow.That may ultimately prove more significant than simply increasing the size of the molecule being modeled. The transition from manually coordinating quantum and classical resources to automating the workflow begins to make the QPU look less like a separate computer and more like an accelerator within a larger computing environment.That is exactly what happened with GPUs. Few application developers today think about physically moving a workload between a CPU and GPU. Software frameworks, runtimes and schedulers increasingly abstract much of that complexity. The processor remains critically important, but the infrastructure surrounding it determines whether its capabilities can be practically used. Quantum computing is beginning to face the same challenge.IBM continues to improve the QPU itself. Its recently introduced Nighthawk r2 processor can execute more than 100,000 circuits per second, which IBM says represents as much as a 25-fold increase in circuit throughput over its Heron systems. Nighthawk r2 has also demonstrated accurate computations on circuits containing more than 7,500 gates.IBM is also investing in the manufacturing infrastructure needed to scale the broader quantum ecosystem. In September, IBM subsidiary Anderon finalized a $1 billion CHIPS and Science Act award to develop a 300-mm pure-play quantum foundry, backed by another $1 billion from IBM, that will manufacture quantum wafers for IBM and other quantum companies. The foundry model could eventually allow quantum developers to focus more of their resources on architectures, systems, and software rather than developing their own wafer manufacturing capabilities.Those are important advances. But faster quantum processors also increase the importance of everything surrounding them. More useful quantum computation creates more demand for classical preprocessing, post-processing, error mitigation, scheduling, data movement, and orchestration.That helps explain why companies outside the traditional quantum processor market are increasingly interested in quantum computing. Nvidia, for example, does not manufacture a QPU. Its CUDA-Q platform is designed around heterogeneous quantum-classical computing where CPUs, GPUs, and QPUs can participate in the same application. That architecture should look familiar to anyone watching the evolution of AI infrastructure.The processor matters, but so does deciding which processor should perform which operation. As these environments become more complicated, AI may become another important part of the orchestration layer. IBM’s quantum roadmap already anticipates opportunities for AI-driven automation to help combine and manage quantum and classical computing resources.That creates an interesting convergence. AI increasingly requires heterogeneous computing. Quantum increasingly requires heterogeneous computing. HPC has been heterogeneous for years. Rather than developing as independent computing architectures, all three are beginning to overlap.The most significant term in quantum computing may therefore eventually be neither “qubit” nor even “quantum advantage.” It may be “workflow.”Practical quantum applications will require systems capable of deciding where workloads run, how data moves between resources, which processor is best suited to each operation and how the results are assembled into something useful. Advances in QPUs remain essential, but useful quantum computing will increasingly depend on the classical infrastructure surrounding them.For years the industry has treated quantum computing as a race to build a better quantum computer, but IBM’s quantum-centric approach suggests a different way to look at the problem. The quantum computer may not ultimately exist as a separate category of computing infrastructure at all. The QPU could simply become another specialized processor within a much larger heterogeneous computing system.If that happens, the most important transition in quantum computing will not be when quantum computers replace classical computers. It will be when we stop thinking of them as separate computers.D-Wave Buys Quantum Circuits in Shift to Higher GearNvidia Bets on the Classical Side of Quantum ComputingU.S. Awards Anderon $1B for Quantum Wafer Manufacturing RELATED TOPICS: AI AND BIG DATA, QUANTUM, QUANTUM COMPUTING COMPANIES: IBM Kevin Hein is a senior analyst at Tirias Research. He is a senior technology executive with over 30 years of experience designing, delivering, and modernizing complex software and systems across commercial, government, and mission-critical environments. You must Register or Login to post a comment.This site uses Akismet to reduce spam. Learn how your comment data is processed.