QCNEWS
Quantum Computing's Next Bottleneck Is Reliability, Not Qubits
The hard work has shifted from physics to industrial operations, and most vendors are not yet ready for it.
The hard work has shifted from physics to industrial operations, and most vendors are not yet ready for it.
A quantum machine goes down at a remote customer site, two time zones from the vendor's service hub. The cryogenic module that failed needs a specialist engineer, specialized tooling, and a careful cool-down sequence to bring it back online. The vendor has maybe two people trained to do that swap, and they are already on a flight to a different customer's site.
This is the operational problem the quantum industry is only beginning to confront. For most of the field's history, the hard work was physics: how many qubits, how low an error rate, how to encode logical qubits so that noise stops destroying the computation. The bottleneck has now changed categories, and the new work is older than the field itself.
Industry analyst Doug Finke, writing in Quantum Computing Report this month, frames the shift through a discipline that IBM built into its mainframes in the 1960s: RAS, or reliability, availability, and serviceability. Finke runs a long-running industry tracker, and he argues that quantum computing's move from lab prototypes to commercial production is being blocked less by physics than by industrial operations. The hardware has to run at the roughly 99.9% uptime that enterprise buyers expect, and the field has not yet built the service machinery to deliver it. (Quantum Computing Report)
Most industry coverage still orients around the physics metrics. Qubit counts, gate fidelities, and error-correction codes dominate the headlines, because they are the visible scoreboard of progress. The operational characteristics that decide whether a customer signs a multi-year purchase order are largely absent from that coverage. Finke is one of the more persistent voices pointing this out, and his mainframe reference is deliberate: the discipline already exists in adjacent industries.
What changes, in concrete terms, is the vendor's obligation. Production quantum systems need Service Level Agreements, with penalties for missing uptime targets. They need regional inventories of spare parts, modular Field Replaceable Units that a service technician can swap without dismantling a dilution refrigerator, and remote telemetry that flags a failing pump before a customer notices. The senior people who can do this work are usually not on-site, because customer facilities rarely house a quantum-experienced engineer. Senior technicians are dispatched from regional hubs, which means the design itself has to be serviceable by someone who has never seen that specific machine before.
Finke also flags operational prerequisites that do not show up in the qubit-count race. On-premise quantum installations need substantial electrical supply, liquid or water cooling, and physical clearance through doors and freight elevators sized for a dilution refrigerator. Procurement officers at large enterprises are starting to walk RAS checklists before signing purchase orders, and at least some quantum vendors are hiring classical IT veterans with mainframe-era reliability backgrounds to staff the work.
Cybersecurity is part of the same shift. Finke notes that enterprise buyers will require SOC 2 Type 2 certification, the audit standard that covers data confidentiality and operational controls, before they will hand over sensitive workloads. The certification process is a multi-month project, and most quantum vendors have not yet published a SOC 2 Type 2 report.
The honest version of the picture is that most quantum vendors are still in lab-deployment mode. Production SLAs are rare or non-existent. Remote-service logistics, particularly the calibration, cryogenic handling, and modular repair that a real service organization needs, are largely unsolved. Capability claims have outrun operational discipline, and the gap is wide enough that a serious enterprise customer can usually find reasons to wait.
The path forward is tractable, though. The same engineering that made mainframes reliable, made modern cloud data centers redundant, and made telecom switches field-serviceable by regional crews, is a known body of work. The work is not in inventing new physics. It is in applying discipline from adjacent industries to a new hardware class, and in being honest about how far most vendors still are from doing it. The bottleneck has changed categories. The next round of progress will be measured in uptime, spare-parts inventories, and signed SLAs, not in qubit counts.