Quantum computing attracts more nonsense per column inch than any other technology, including AI. Let us separate the three things that are true from the many that are not.
What quantum computers are actually good at
Not \”everything, but faster.\” Quantum machines offer advantage on a narrow set of problem structures: simulating quantum systems (chemistry, materials, drug binding), certain optimisation landscapes, and — famously — factoring large numbers. For the overwhelming majority of computing, including training neural networks, classical hardware is and will remain better.
Anyone selling you a quantum AI product today is selling either a research collaboration or a story.
The one genuine breakthrough
Qubits are fragile; noise destroys computation. For years the question was whether error correction could scale — whether adding more physical qubits to protect a logical one made things better or worse. Recent results demonstrated that crossing the threshold works: below-threshold error correction where scaling up genuinely reduces the error rate. Google Quantum AI and IBM Quantum both publish on this, and it is the difference between \”interesting physics\” and \”eventual engineering.\”
That does not mean useful machines tomorrow. It means the path is no longer theoretically blocked.
The deadline that applies to you
Here is the part with a date on it. Most of today\’s public-key cryptography — the padlock on your site, your wallet keys, your VPN — is breakable by a sufficiently large quantum computer. Nobody has one yet. But adversaries can harvest now and decrypt later: capture encrypted traffic today, store it, and open it in a decade.
Consequences for anyone shipping software:
- Post-quantum cryptography standards are finalised and being deployed. Your libraries and vendors should have a migration story — ask them.
- Data with a long secrecy lifetime — medical, legal, financial, state — needs to move first.
- Blockchains face a specific version of this. Long-term key hygiene and chain-level migration plans matter more than any token narrative. If you build dApps, this belongs on your risk register.
The realistic threat is not that a quantum computer breaks your encryption tomorrow. It is that traffic you send today is being stored for a machine that arrives in 2035.
Where quantum and AI genuinely touch
- AI helping quantum. The stronger direction today. Machine learning improves qubit calibration, error decoding and pulse control. Real, deployed, unglamorous.
- Quantum-inspired classical algorithms. Ideas from quantum research producing better classical optimisers. Useful, and available now.
- Quantum machine learning. Active research, no established commercial advantage. Interesting if you are a researcher, irrelevant if you are a founder. PennyLane is the friendliest way in.
What to do this year
If you are technical and curious, run a circuit on real hardware — free access through IBM, and Qiskit is well taught. If you run a business, put post-quantum migration on the roadmap and ask your vendors one pointed question about it. If you are investing, treat 2030s revenue claims with the scepticism you would apply to any pre-revenue physics.
More links and reading on the robotics, 3D printing and quantum page.
