Quantum is back in the business conversation, but usually through two very different doors. One is post-quantum cryptography, where the NCSC is clear that organisations should already be preparing for migration. The other is operational quantum sensing and navigation, where work such as Network Rail’s March 2026 mainline railway trial and SandboxAQ’s AQNav offering show that some quantum-adjacent capabilities are moving into real-world testing and deployment.
What gets talked about much less is quantum machine learning. I have been digging into that recently through IBM Quantum’s Machine Learning certificate, and then trying to turn the learning into something practical. I built a small workflow (with Anthropic Claude Code) that benchmarked data classically, used a local model to decide whether quantum was even worth pursuing, generated Qiskit code only when the case looked reasonable, and then submitted jobs to IBM Quantum hardware.
That exercise left me more positive about the space, but not because quantum machine learning is ready for everyday business use, but because it is becoming much easier to explore intelligently.
What is it?
Quantum machine learning is not a wholesale replacement for machine learning as most organisations know it today.
It is better understood as a set of hybrid approaches where classical data is prepared, but selected parts of the workflow use quantum circuits. IBM’s own learning material makes that very clear. The workflow still starts with classical inputs, then maps them into a quantum problem, optimises for execution, runs the circuit, and post-processes the result. The same course also explicitly calls out that QML is promising in some cases and not in others.
In other words, this is not “machine learning, but faster”. At the moment, QML is mostly about specific methods and training on specific datasets rather than presenting QML as a general-purpose enterprise ML replacement.
That distinction matters because a lot of business readers hear “quantum” and understandably assume general breakthrough. My takeaway is more restrained, that quantum computing is still best viewed as suited to certain classes of complex problems rather than as a substitute for classical computing overall. IBM says this directly, noting that current hardware is expensive, large, and error-prone, and not expected to replace traditional computing any time soon.
At the same time, the tooling is getting more accessible. IBM’s QML course is designed to let learners implement using Qiskit, adjust examples to their own datasets, and make recommendations about where QML might help an organisation. That does not mean the field is mature. It does mean the barrier to hands-on learning is starting to come down.
That was the real lesson for me. Not mature. Not simple. Not yet industrialised. But more accessible.
What does it mean from a business perspective?
The business direction is not “rush into production”. It is “be more deliberate about where and how you explore”.
The cost of experimentation is falling. In my own work, GenAI (Claude Code) made the quantum world feel much more approachable. It helped close some of the gap between concept and implementation, especially around Qiskit syntax, code structure, and workflow assembly. That does not remove technical complexity, but it does reduce friction enough that small teams can learn faster.
Suitability matters more than excitement. The right question is not whether quantum is interesting (it is very, very interesting). It is whether classical methods are struggling enough to justify extra complexity.
QML is a workflow design problem as much as an algorithm problem. The practical pattern is hybrid: classical preparation, quantum execution, post-processing. That means business value depends not just on the model, but on orchestration, validation, data handling, and who is accountable for interpreting results.
Different quantum topics are at very different levels of maturity. PQC is already a planning issue for mainstream organisations. Quantum navigation is showing operational trials and deployment-oriented products. QML, by comparison, is still earlier and more experimental. Treating all “quantum” as one market signal leads to poor decisions.
The skills challenge is shifting. The differentiator is becoming less about hand-coding everything from scratch and more about problem framing, benchmarking, validation, and governance. As tooling improves, judgement matters more, not less.
There is a leadership angle here. Risk-aware organisations, public sector teams, and consultants do not need to become quantum shops overnight. But they do need enough literacy to separate real opportunity from premature enthusiasm and to recognise when a small exploratory effort is justified.
What do I do with it?
This is not a “buy quantum” moment. It is a “build enough understanding to make better decisions” moment.
Start with the business problem, not the technology. Look for hard classification, optimisation, or simulation problems where classical methods are expensive, flattening out, or structurally limited. Do not start with “we need a quantum use case.”
Put a classical baseline first. Benchmark conventionally and make quantum earn its place. That was the most useful design choice in my own pipeline, and it seems to be the right instinct from both a strategy and governance perspective.
Keep early work hybrid and contained. Use proofs of concept, sandboxes, and explicit decision gates. Small, bounded experiments create learning without pretending the technology is more mature than it is. (Also, quantum computing time is expensive - $96USD/minute.)
Use GenAI to reduce learning friction, not to bypass judgement. It can accelerate code generation, explain strange syntax, and help teams get further, faster (see my GitHub repo for inline code generation by Claude - Quantum Agent).
Separate your quantum conversations. PQC readiness, quantum sensing/navigation, and quantum computing applications such as QML may share a headline, but they do not share the same investment logic, urgency, or operating model.
Add light governance early. Decide who can experiment, what data can be used, how results will be validated, and what would count as enough evidence to continue or stop.
Build literacy before urgency arrives. The organisations that will handle this space best are unlikely to be the ones that wait for a fully polished market. They will be the ones that develop a practical understanding early, while expectations are still manageable.
My main takeaway from learning and testing quantum machine learning is quite simple.
Quantum ML is not ready to replace mainstream machine learning in business. But the path to testing, understanding, and experimenting with it is getting much more approachable.
That matters, because lowering the cost of exploration changes who can learn, how quickly they can learn, and how prepared they will be when the technology does become more useful in selected areas.
For now, I would treat QML as a space for disciplined curiosity. Strategy first. Quantum second.
GitHub Repo: https://github.com/steveh250/IBMQuantum-Agent