IBM's March 2026 reference architecture for quantum-centric supercomputing concludes that traditional batch schedulers require a fundamentally different approach for heterogeneous quantum-classical resources. The present paper extends the reference architecture's vision with proof-of-concept evidence that IBM Spectrum Symphony, a service-oriented dynamic compute platform, can address the orchestration requirements the reference architecture identifies. Fifteen demonstrations spanning twelve silicon architectures, six compute tiers and two network fabrics orchestrate QPU, NPU, GPU, CPU and mainframe as peer resource types under a single scheduling domain. Symphony's orchestrator consumes per-resource metrics reported through the External Load Information Manager (ELIM), a generalization of Songnian Zhou's 1987 load index. ELIM reports arbitrary classical metrics from every resource type; the platform makes placement decisions through the same algorithm regardless of whether the resource is quantum, neuromorphic, GPU or mainframe. All scheduling-relevant QPU metrics are classical by nature, and the quantum demonstrations provide direct evidence of platform-driven scheduling, routing identical Qiskit code to GPU simulation or IBM Quantum hardware based on ELIM-reported resource state. Four quantum applications with 2,560 hyperparameter combinations, an event-driven closed-loop quantum-neuromorphic workflow and multi-modal edge-to-cloud sensor integration are among the demonstrations. The orchestration framework applies equally to AI and LLM workloads, with five demonstrations involving LLM inference as a SOAM service tier, including semantic routing across model tiers and GPFS-based KV cache sharing. The implementation lineage from Zhou's dissertation through Platform Computing (1992), Platform Symphony (2006) and IBM Spectrum Symphony (2012) spans four decades of production deployment at the largest financial institutions in the world. The earliest quantum-classical demonstration with Symphony predates the reference architecture by several weeks, indicating the capability is already achievable with existing platforms.
Kevin D. Johnson (Mon,) studied this question.