Quantum-Centric Supercomputing (QCSC)

qc/hpc qc/hardware

IBM’s stated three-stage roadmap (as of 2026 — a target, not a guarantee) for integrating quantum processors into classical HPC infrastructure, tracking the same underlying timeline as the hardware roadmap but describing how the systems get wired together rather than what the QPU itself can do.

The three stages

1. Quantum as co-processor to HPC (~2026–2028) — specialized offload engines, quantum hardware co-located with existing HPC systems, post-selected error correction.
2. Heterogeneous Quantum+HPC Systems (~2028–2031) — tight coupling between quantum and classical resources, advanced middleware purpose-built for hybrid algorithms, unified scheduling across both, near-time conditional error correction.
3. Tightly Integrated Quantum+HPC Systems (~2031–2034) — co-designed heterogeneous systems from the ground up, a single unified programming model spanning quantum and classical resources, interoperable high-performance fabric, multi-tenant execution, real-time error correction.

Key insight: this maps directly onto the hardware side’s progression from noisy-and-mitigated to fault-tolerant — Stage 1 is today’s Error Mitigation section reality, Stage 3 assumes the Starling/Blue Jay fault-tolerant hardware already described there. Two parallel roadmaps (hardware capability, integration architecture) converging on the same fault-tolerant endpoint.

Qiskit Runtime’s role

Positioned specifically as the layer that reduces classical-quantum latency — the round-trip cost of a classical optimizer calling back into a QPU repeatedly (as in VQE’s loop) is itself a bottleneck QCSC’s middleware stages are designed to shrink.

Related

• Error Correction (EC) — the parallel hardware-capability roadmap
• What is HPC
• Why Quantum Needs HPC
• The Quantum+HPC Software Stack

Self-Check

• Could you name the three QCSC stages and, for each, say what actually changes about how a QPU is used?
• Why does this roadmap track the same timeline as the hardware fault-tolerance roadmap in Error Correction (EC)?
• Why does classical-quantum latency matter enough to be a named design goal for Qiskit Runtime?