In written responses, CEO Marius Grundmann says the cores are not entangled and details the estimates behind SaxonQ's six-to-ten-times energy claim.
SaxonQ has answered Supercomputing News' fact-check questions about its 128- and 512-qubit room-temperature diamond quantum computers. Asked whether a complete SXQ128 or SXQ512 has been assembled, demonstrated, delivered, or benchmarked, chief executive Marius Grundmann described a dual-core system demonstrated with its multi-core software stack at Hannover Fair 2026, an 8-qubit single-core system with two coupled nitrogen-vacancy centers due for delivery to a paying client before the end of the third quarter, and a demo SXQ128 the company is "currently setting up" while it negotiates orders and system sizes, in written responses to SCN dated July 27.
The responses address the five questions SCN sent on build status, addressability, inter-core entanglement, benchmark provenance, and model-specific data, along with a follow-up on manufacturing yield. They arrive after SaxonQ opened orders for both systems. Its roadmap, investor materials, and company launch release list the SXQ128 as orderable with three-month delivery and the SXQ512 as orderable for delivery from the second quarter of 2027:
Model | Aggregate physical qubits | Cores | Fully entangled qubits per core | Public status |
|---|---|---|---|---|
SXQ128 | 128 | 16 | 8 | Orderable; vendor states three-month delivery |
SXQ512 | 512 | 32 | 16 | Orderable; vendor states delivery from Q2 2027 |
The totals and per-core figures measure different things, and the company has now put the distinction on the record. All qubits are individually addressable and run at the same time, Grundmann wrote, and "only the qubits within a given core can be entangled." What SaxonQ has still not published: model-specific full-core fidelity distributions, shot throughput, or calibration duty-cycle data for either announced system. Grundmann said the company "will provide further benchmarking data as they become available."
A nitrogen-vacancy center places a nitrogen atom beside a vacancy in the diamond lattice; its spin can serve as a room-temperature qubit. Grundmann described the core design in more detail than SaxonQ's public materials have. Each core contains two NV centers fabricated reproducibly 10 nanometers apart, he wrote, close enough that they couple strongly through dipole-dipole interaction and can be entangled. That coupling weakens steeply with distance, which is why placement precision at the 10-nanometer scale is the fabrication problem the company's sulfur-assisted implantation process exists to serve. He called the coupled-pair core, which he said is commercially available, the company's "central progress in the field of NV/diamond quantum computing." Later cores on the roadmap will contain more coupled NV centers per core.
The first delivery vehicle for the design is the single-core, 8-qubit system due this quarter. The SXQ512, per his answers, runs 32 such cores of 16 qubits each.
The cores "work in parallel/simultaneously in a multi-tasking way (useful for circuit cutting) and are not entangled," Grundmann wrote. That settles what SaxonQ's public specifications left open. The aggregate qubit count describes parallel capacity; the per-core count is the bound on native entanglement.
Grundmann defended the design as unexceptional. In many modalities, he argued, the maximum number of entangled qubits or the quantum volume sits well below the total qubit count, and he said most commercially available spin-qubit processors, including silicon spin qubits at millikelvin temperatures, contain far fewer than 16 qubits. Gaps of that kind do appear across the industry; SCN found the same definitional problem between headline qubit numbers and usable capacity in vendor roadmaps toward DOE's 2028 fault-tolerance target. Cleanly separating entangled registers from registers that cannot be entangled reduces crosstalk and simplifies scaling, he wrote, and the multi-core layout lets multiple users run jobs simultaneously. "Virtually any number of cores is feasible if larger systems are demanded."
Circuit cutting is the route SaxonQ offers for circuits larger than a core. Its QC2026 whitepaper says cutting divides a circuit into subcircuits, runs the pieces on smaller cores, and recombines the results classically, while carrying significant overhead and scaling poorly for highly entangled problems. Whether cutting pays off depends on circuit structure, the number of cuts, sampling cost, and the classical reconstruction step. The written responses did not include a measured cutting overhead for either announced system. For batchable jobs, independent cores can instead raise throughput through parallel circuits or faster shot collection.
The distinction still matters when comparing machines. Quantinuum's H2 data sheet specifies 56 physical qubits with all-to-all connectivity; SCN covered a 56-qubit H2 many-body experiment. That connected register is a different computational object from 512 qubits divided among 32 disclosed 16-qubit registers. Quantum chemistry exposes the same boundary: IBM and Oak Ridge National Laboratory mapped a 33-orbital fragment to a 66-qubit circuit in a FLiBe study, part of a hybrid workflow SCN covered.
The 99.92 percent figure associated with the launch comes from SaxonQ's QC2026 whitepaper, which reports one-qubit gate fidelity up to 99.92 percent for the prior-generation Gen3 platform, alongside 97 percent two-qubit fidelity and 600 watts of power draw. Grundmann's responses add the method. The figure refers to an electron-nuclear conditional rotation, an SX gate, determined from randomized benchmarking at room temperature, he wrote. The whitepaper lists the number as a one-qubit gate fidelity; the responses describe the underlying operation as a conditional rotation between an electron and a nuclear spin. Both attributions are SaxonQ's, and neither is an SXQ128 or SXQ512 measurement.
A second fidelity number appears only in the responses. SaxonQ's best nuclear qubit was recently measured at 99.98 percent, plus or minus 0.02, on a client's four-qubit register, Grundmann wrote. The company gave SCN unpublished randomized-benchmarking data behind that figure on condition it not be reproduced. The number describes a single best-performing qubit on a small register, not a distribution across a core. No third party has verified it, and SaxonQ has not published it.
On stability, the responses stake out vendor claims no third party has yet tested. NV electrons are the shortest-lived qubits in the system, Grundmann wrote, with Hahn-echo T2 typically above 400 microseconds at room temperature, while nuclear spin qubits hold coherence much longer. Recalibration is unnecessary, he said, because the NV and its nuclear spin system are atomically defined and stable; the qubit resonance frequency is checked and fine-tuned roughly once per hour or on request, and temperature shifts across the specified 18 to 27 degrees Celsius are compensated automatically.
Independent evidence still stops at smaller hardware. The German Aerospace Center accepted four-qubit demonstrators after verifying lower bounds above 95 percent for one-qubit operations and 90 percent for two-qubit operations. Fraunhofer IWU installed a four-qubit SaxonQ system in 2025. Neither result benchmarks the announced products.
SaxonQ's launch release says the systems use six to ten times less energy than GPU clusters on equivalent workloads. Grundmann called the figure "an estimate" and laid out its basis. IBM and IonQ have estimated that their quantum computers with 33 to 35 qubits perform similarly to a single Nvidia GPU on certain problems, he wrote. He added that IBM's 156-qubit system, running an undisclosed optimization algorithm from Kipu Quantum, works similarly to eight GPUs drawing about 5 kilowatts, and that an IBM system with more than 100 qubits presently runs on 45 to 50 kilowatts. Against that, SaxonQ projects the SXQ128 "will need less than 2 kW" at full load and the SXQ512 less than 4 kilowatts, with further per-core reductions on its roadmap.
The GPU-equivalence figures are third-party comparisons as characterized in SaxonQ's responses, the algorithm behind the IBM optimization comparison is undisclosed, and the SaxonQ wattages are projections for machines the company has not finished building. Without a named workload, accuracy target, runtime, and measured full-system power, the comparison is still not reproducible.
SaxonQ's launch materials say sulfur-assisted implantation converts more than 85 percent of implanted nitrogen into usable centers. Grundmann's answers define the figure: the conversion yield of implanted nitrogen atoms into negatively charged NV centers at implantation energies around 10 kiloelectronvolts, the low energies that placement accuracy for coupled-NV registers demands. Without sulfur, the yield at those energies is below 1 percent, he wrote, a baseline corroborated independently by a deterministic single-ion implantation study in New Journal of Physics.
The published record on the sulfur effect remains narrower than the model-scale claim. A 2019 Nature Communications experiment reported 75.3 percent conversion with sulfur co-implantation, against 6 to 8 percent for intrinsic diamond under that experiment's conditions. A 2025 first-principles study, now published in Physical Review Research, proposed an atomistic mechanism for the effect; it is theoretical and does not estimate yield.
The greater-than-85-percent figure itself still has no published independent verification. Grundmann said SaxonQ has corroborated its single-ion implantation results with low-energy implantation of nitrogen molecules, which yields the coupled NV pairs used in its cores, and pointed to routine fabrication of those pairs as evidence of high conversion, though the supporting data behind both corroborations remain unpublished.
Leipzig hosts a small diamond-quantum cluster. XeedQ lists its XQ1 product at four or more qubits and holds a German Aerospace Center project targeting more than 32 qubits, while Bechtle describes itself as SaxonQ's first certified partner. For supercomputing centers weighing an on-premises quantum accelerator, rack installation at 18 to 27 degrees Celsius from a standard outlet removes runtime cryogenics. It could also shorten the control loop between quantum jobs and the classical systems that prepare circuits and reconstruct cut results. The trade remains compact integration against disclosed registers of eight or 16 qubits.
The next evidence is functional. On Grundmann's timeline, the single-core system due before the end of September will put a coupled-NV-pair core in a paying client's hands. Full-core one-qubit, two-qubit, and readout distributions, shot throughput, calibration duty cycle, wall power measured under a named workload, and measured circuit-cutting overhead remain unpublished for both announced models. Until they appear, 128 and 512 are announced aggregate capacities, while eight and 16 are the disclosed fully entangled register sizes.
The SCN Staff is a small AI editorial squad working under human direction. Each agent owns one job.
Scout does the research. It runs down primary sources and checks what's already been published, on SCN and everywhere else, before a story gets written. If a claim can't be traced back to a real document, Scout flags it.
Forge writes. It takes what Scout found and turns it into a draft, argument and sentences and all. Every SCN piece starts here, then gets sharpened.
Cipher handles search: the titles, descriptions, and keyphrase work that decides whether a good article ever gets found. Least glamorous job on the squad. Also one that matters more than it looks.
Pixel makes the visuals. Images, charts, the occasional diagram, all built to SCN's brand instead of pulled from a stock library. When something's easier to see than to read, it goes to Pixel.
Editorial judgment and the final call stay with the humans. So does the fact-checking.
