Room Temperature, 600 Watts: Three Qubits Run Grover's Search in a Customer's Machine Room

Forum topic · QianXun · 2026-10-18

Summary

A commercial quantum computer from Leipzig-based SAXON Q, working with Fraunhofer IWU in Dresden, has run a complete three-qubit Grover's search algorithm at room temperature and ambient pressure using only about 600 watts of power. The machine uses nitrogen-vacancy (NV) center diamond, but unlike typical NV setups, all three computing qubits are nuclear spins—a nitrogen-14 nucleus plus two carbon-13 nuclei—with coherence times of 1.4–4.3 ms, roughly 1000× longer than NV electron spins. The electron spin serves only as an optical read/write interface. The single-marked-state search achieved 77.3% ± 3.4% success versus a classical limit of 37.5%; two marked states reached 87.0% ± 4.2%. This is reportedly the best published result for this algorithm on a room-temperature platform, only slightly below the leading superconducting result that requires cooling to 15 millikelvin. The experiment was performed in June 2025 on a previous-generation machine at the customer's site, not on the newer SXQ128/SXQ512 systems. The result is an arXiv preprint and remains unreviewed, and the three-qubit scale demonstrates feasibility rather than practical quantum advantage.

A commercial quantum computer sat in a customer's facility—no vacuum, no cryostat, drawing roughly 600 watts—and completed a three-qubit Grover's search in ambient air at an average of 23.1 °C. One marked state, 77.3% success; the classical upper bound for the same problem is 37.5%. This is the first time an algorithm of this scale has been run end-to-end on a room-temperature quantum computer, and the best comparable superconducting result beat it by less than one percentage point—while requiring 15 millikelvin.

💎 The Computation Happens in Atomic Nuclei

The machine comes from SAXON Q (Leipzig), in collaboration with Fraunhofer IWU (Dresden); the paper was released on October 14. The hardware is nitrogen-vacancy (NV) center diamond, but the computation does not use the famous electron spin: all three computing qubits are nuclear spins—the nitrogen-14 nucleus plus two neighboring carbon-13 nuclei. The carbon lattice shields these nuclear spins well, with measured coherence times of 1.4 to 4.3 milliseconds, roughly one thousand times longer than the NV electron spin. The electron spin is demoted to a supporting role in this architecture, serving only as the optical read/write interface for initialization, entanglement, and readout.

> Why is Grover's algorithm often used as a stress test? It requires running a full gate sequence: superposition, oracle, diffusion, iteration, and final readout—not just benchmarking a single metric. Completing it end-to-end shows the entire stack is working. It is also the standard template for quantum search, with a theoretical quadratic speedup over classical brute force.

📊 The Scorecard and Comparisons

One marked state: 77.3% ± 3.4%, classical limit 37.5%.
Two marked states: 87.0% ± 4.2%, classical limit 46.4%.
It is the highest published room-temperature result; the only higher figure, 89.4%, comes from a silicon spin-qubit processor at the cost of 15 millikelvin plus post-selection on readout.
For similar algorithms, superconducting processors have reported 49.2%–78.0%; trapped ions, 43.7%–69.0%.
Gate metrics: average single-qubit Clifford fidelity of 99.90% (best 99.98%); the two-qubit subspace containing the three-qubit gate averaged 95.7%—both measured via randomized benchmarking.

The contrast is easy to anchor: this machine runs at 296.3 K, ~600 W, in open air; the dilution-refrigerator route needs below 20 mK, over 25 kW, plus facility-scale chilled water. The company puts the temperature gap at roughly 15,000×. Fraunhofer values the other end: the machine was in their machine room, they supervised the measurement methodology themselves. The head of the institute's digital production twin department said that seeing this level of performance on-site gives them confidence for industrial optimization and robotics work.

| Platform | Success rate (same algorithm) | Operating conditions | |---|---|---| | SAXON Q diamond | 77.3% single / 87.0% double marked | 296.3 K, ~600 W, ambient air | | Superconducting (published range) | 49.2%–78.0% | below 20 mK, over 25 kW | | Trapped ions (published range) | 43.7%–69.0% | ultra-high vacuum, laser cooling | | Silicon spin (single highest) | 89.4% | 15 mK plus post-selection |

🧊 The Cold Side of the Story

Three qubits, an eight-element search space: this is a small hardware benchmark—enough to prove the algorithm runs on a room-temperature machine, not enough for practical quantum advantage. Grover's quadratic speedup only emerges at large scale; an eight-element venue cannot demonstrate scale. The experiment ran on the previous-generation machine, not the SXQ128 and SXQ512 announced in July, whose performance this result does not endorse. Per the company roadmap, the SXQ128 is now orderable and the SXQ512 is slated for delivery starting Q2 2027. The paper is an arXiv preprint, not peer reviewed, and the numbers are per the official release. CEO Marius Grundmann's summary: the quantum community has long assumed you trade performance for convenience; these results show that trade-off is not mandatory. Fair enough—but read that at three-qubit scale.

What deserves attention is the roadmap itself: room temperature and pressure, semiconductor-compatible fabrication, over 220 patents and applications, and a team from the University of Leipzig group that has studied NV centers for over three decades. If full algorithms can run at customer sites, quantum computing's deployment model shifts from large instrument to purchasable equipment. What that "if" is worth is more interesting than the 77.3% figure.

Sources + caveats: Primary sources are the SAXON Q newsroom and the arXiv preprint (SAXON Q with Fraunhofer IWU); figures cross-checked as consistent via Quantum Computing Report, EEJournal, QuantumWire, and Science.Report. The experiment was completed in June 2025 on a previous-generation commercial machine, unrelated to the SXQ128/SXQ512 released in July; the 3.4 and 4.2 percentage points are the paper's stated uncertainties; ~600 W and 25+ kW are system-level power figures per the company and press coverage; the paper is not peer reviewed and full tables were not verified.