Quantum technologies use controllable quantum states as computational, communication and sensing resources; the sources show sensing and memory hardware already delivering measurable results while the computing part remains at 6-to-10-qubit demonstrations plus simulator-stage error-correction theory.
Summary
“Quantum technologies” is not one product but three distinct engineering stacks that share a physics base. Computing encodes information in two-level (or multi-level) quantum systems and drives them through interference to solve problems classical machines find hard. The leading solid-state route is superconducting circuits: Josephson junctions embedded in microwave resonators behave as artificial atoms with anharmonic spectra, controlled and read out within circuit quantum electrodynamics, a framework that has underpinned the transmon and fluxonium platforms since the early 2000s. Communication moves quantum states between nodes, which requires memories that store a photon and give it back on demand with high efficiency, at telecom wavelength, ideally on a chip ref. Sensing and metrology exploit the fragility of quantum states as a feature: nitrogen-vacancy centres in diamond act as nanoscale magnetometers, and entanglement can be engineered to push measurement precision past classical limits.
The binding constraint on the computing part is decoherence. Physical qubits fail, so useful computation requires quantum error correction: encode one logical qubit across many physical ones, measure syndromes repeatedly, and infer the error in real time. Decoding is a hard classical inference problem in its own right, and existing decoders (clustering, matching, belief propagation, neural networks) each trade accuracy against cost or generality across the main code families, including surface, toric and bivariate bicycle codes; the stated target is fault tolerance on millions of qubits decoded in real time. Theory is now describing decoherence itself as a phase transition in mixed-state topological order, with the loss of protected information classified by anyon condensation at a critical coupling. That is the intellectual frame; the practical question is whether decoders can be built into local hardware at latency.
The parameters that decide each part are different. For computing: physical error rate, decoder accuracy and latency, and the qubit cost of encoding a real problem, which is why work on compact encodings matters (reducing the travelling salesman problem from the O(n²) qubits of one-hot encoding to O(n log n)). For algorithms: whether speedups survive outside structured problem classes; Decoded Quantum Interferometry has strong evidence of a superpolynomial advantage on the optimal polynomial intersection problem, while broad practical optimisation advantage is still described as challenging. For memory: the four-way trade of efficiency, wavelength, temperature and integration, which no single platform currently satisfies ref. And competitively, quantum is not the only physics-based computing bet: thermodynamic hardware proposes to exploit noise as a resource for AI workloads, positioning itself explicitly against quantum computing with the analogy that thermodynamic chips emulate Langevin’s equations as quantum computers emulate Schrödinger’s ref.
Evidence base
- 4 May 2026: Lottery BP reports 2-8 orders of magnitude decoding accuracy improvement over belief propagation for topological codes, with syndrome voting to mitigate the decoding backlog, targeting fault tolerance on millions of qubits in real time.
- 8 November 2025: integrated Eu:YSO quantum memory achieves 80.3(7)% storage-retrieval efficiency, 69.8(1.6)% for telecom-heralded single photons and 20 temporal modes at 51.3% average efficiency, but requires cryogenic operation; cold-atom (~90%) and warm-vapour (~82%) rivals are more efficient yet near-IR and free-space ref.
- 4 May 2026: NV-centre ensembles fabricated on a diamond anvil surface image iron’s stray magnetic field up to 30 GPa and resolve the α-ε magnetic transition.
- 4 May 2026: a compact-encoding variational TSP framework reduces data qubits from O(n²) to O(n log n) and reports simulated success rates of 100%, 100% and 95.5% for 4, 5 and 6 cities, with hardware validation only as a two-qubit implementation on SpinQ NMR machines.
- 4 May 2026: a review of Decoded Quantum Interferometry states strong evidence of superpolynomial speedup on the optimal polynomial intersection problem while noting that clear, scalable advantage on broadly useful practical optimisation remains challenging.
- 4 May 2026: room-temperature on-chip magnetic levitation of a 6.5 µm-radius ferromagnetic microsphere with librational eigenfrequencies above 10 kHz, avoiding the cryogenic requirement of Meissner levitation.
Open questions
- Does the 2-8 orders of magnitude accuracy gain of Lottery BP survive implementation in the PolyQec architecture at real-time latency on live syndrome data, rather than in simulation?
- Does the DQI superpolynomial speedup extend beyond structured instances such as optimal polynomial intersection to problems with commercial demand, or does it stay confined to the max-LINSAT class?
- Can rare-earth integrated memories retain ~80% efficiency and telecom operation while relaxing the cryogenic requirement, or does the four-way trade force hybrid architectures ref?
- Do compact encodings such as O(n log n) TSP registers combined with divide-and-conquer execution scale past toy instances on hardware, or does the classical post-processing dominate the result?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.