A German startup called SAXON Q is trying to put a quantum chip in a standard server rack, no freezer required. On October 9, The Quantum Insider ran a guest post from SAXON Q’s CEO and physicist, Marius Grundmann, describing a new test on the company’s diamond-based quantum processor. The company reports the numbers. No outside lab has repeated them yet.
- Price: none published
- Availability: SAXON Q says SXQ128 deliveries come within three months of an order, and SXQ512 shipments start in the second quarter of 2027
- Works with: a standard server rack and routine power, according to Interesting Engineering
- Status: company-reported results from a preprint and a CEO guest post, with no outside replication or peer-reviewed paper yet
What the test showed
The team ran Grover’s search algorithm, a classic quantum method for finding one marked item in an unsorted pile. They used three qubits built from nuclear spins inside a diamond. The post says the machine ran at about 296 kelvin and normal air pressure, with no vacuum chamber and no cryostat, on roughly 600 watts.
The underlying preprint searched a space of eight items. It found one marked answer 77.3 percent of the time, while a classical computer making the same number of guesses tops out at 37.5 percent. With two marked answers, the chip scored 87.0 percent against a classical best of 46.4 percent.
| Grover test (8 items, 3 qubits) | SAXON Q reported | Classical best, same queries | Source |
|---|---|---|---|
| One marked item found | 77.3% | 37.5% | arXiv preprint, Quantum Insider guest post |
| Either of two marked items found | 87.0% | 46.4% | arXiv preprint, Quantum Insider guest post |
| Trapped-ion runs of the same search | 44% to 69% (cited by SAXON Q) | Not applicable | Quantum Insider guest post |
The preprint reports gate fidelity, a measure of how cleanly each operation runs, of up to 99.98 percent, and the authors say their results beat published superconducting results. The CEO’s own post is more careful. It gives an average single-qubit fidelity of 99.90 percent and an average two-qubit fidelity of 95.7 percent, and says ion traps still hold the best raw fidelities of any approach.
The trick lives in a nitrogen-vacancy center, a tiny defect in the diamond that acts as an optical link to the nuclear spins next to it. Grundmann puts the bigger picture bluntly: “Fabricating the qubit is not the bottleneck. Housing it is.”
Why skipping the freezer matters
Many quantum computers use dilution refrigerators that cool chips to tens of millikelvin, close to absolute zero. They also lean on helium-3, a rare gas. PostQuantum reports that bulk helium-3 now costs $1,900 to $2,600 a liter. It says one Bluefors XLD1000sl fridge holds about 40 liters of it, worth roughly $100,000 at $2,500 a liter. Almost all of it comes from the radioactive decay of tritium.
Grundmann’s post says the diamond system used about 600 watts, not the tens of kilowatts a dilution refrigerator and its supporting plant need. SAXON Q’s commercial boxes follow the same idea. The company says the SXQ128 has 128 qubits and the SXQ512 has 512, and that both fit a standard server rack and run on routine power. No price has been published.
A crowded race to room temperature
SAXON Q isn’t running this race alone. In a recent story, researchers claim three room-temperature records for neutral-atom qubits using floating atoms instead of diamond defects. Other labs take a different route. Google’s simulations suggest it could cut error-correction overhead by linking smaller, cold chips together rather than warming the hardware up.
A fair caveat: Grundmann runs the company selling this chip. The demo solved a tiny eight-item search, not a real chemistry problem. Independent coverage of SAXON Q’s commercial launch notes that the fidelity, yield, and energy figures come from the company. Independent tests on the first SXQ128 systems to reach customers will say far more than one company’s demo.
Why It Matters to You
Picture a chemistry teacher at a community college in Ohio, Lena Ortiz, an illustrative example and not a real person, who wants students to model a new battery material at the atomic level. Today’s quantum machines mostly sit behind locked doors at national labs, so her department will never see that budget. If SAXON Q’s prices turn out low, a community college, a hospital lab, or a small manufacturer might one day rent or own a quantum co-processor. That is a possibility, not a promise: the chip has solved only a tiny search so far, and there is no price yet.
Open questions
- The search covered only eight items and three qubits. It shows the method works at tiny scale, not that the chip can solve a useful problem.
- The figures come from a preprint co-written by SAXON Q’s CEO and from a guest post by the same person. A peer-reviewed paper with outside replication is still pending.
- The 99.98 percent figure is a maximum. The average two-qubit gate fidelity in the same post is 95.7 percent.
- PostQuantum notes that the “128 qubits” are spread across 16 cores of eight qubits each, run independently and joined by circuit cutting. It calls the quantum link between cores “a key open issue.”
- PostQuantum also says no full-core or inter-core two-qubit benchmark has been published for the SXQ128, and no third-party data exists for the new systems.
- The 6x to 10x energy advantage over GPU clusters has no stated benchmark, workload, or test method.
Quick answers
Does this chip really work at room temperature?
SAXON Q says it ran the test at about 296 kelvin with no vacuum chamber or cryostat. Outside labs haven’t confirmed that. A customer, Fraunhofer IWU, told Interesting Engineering its system has run at room temperature since June 2025.
What is Grover’s algorithm?
It is a quantum method for finding a marked item in an unsorted list. Classical computers must check items one by one. Grover’s method needs fewer checks as the list grows, though the gain here is small because the list had only eight items.
Is this a 128-qubit quantum computer?
SAXON Q sells the SXQ128 as 128 qubits. PostQuantum says that figure is packaged capacity across 16 independent cores of eight qubits each. The test in this story used three qubits.
Sources
- The Quantum Insider: Guest Post, Quantum Computing’s Transistor Moment (Oct 09, 2026)
- arXiv: Grover Search with Semiconductor Spin Qubits at Ambient Conditions (Sep 11, 2026)
- Interesting Engineering: First 100-Qubit Diamond Quantum Processors (Jul 21, 2026)
- PostQuantum: SAXON Q 128-Qubit Diamond (Jul 22, 2026)
- PostQuantum: Helium-3 and Quantum Computing’s Coldest Problem (Apr 01, 2026)
