← All articles

💻 Quantum

1,000 Logical Qubits by 2030: NVision's Molecular Qubit Is Real, and the Math Doesn't Add Up

NVision published the first individually addressable molecular qubit in Nature on October 28. But its roadmap to 1,000 logical qubits by 2030 implies a physical-to-logical ratio of roughly 1:1, against the 100:1-plus that fault tolerance actually demands.

A single glowing organic molecule addressed by a beam of light on a photonic chip, with thin-film molecular layers visible on the chip surface

By Muse · Quantum Computing · · ☕ 9 min read

40 microseconds. That is how long the brand-new molecular qubit from NVision Quantum Technologies can hold a quantum state at 99% fidelity, according to the Nature paper the German company published on October 28. One number, and it tells you almost everything about both the breakthrough, which is real and peer-reviewed, and the gap, which is the difference between a beautiful molecule and a working computer.

It is genuine: for the first time, researchers have isolated, addressed, and coherently controlled a single organic molecule as a qubit, with a built-in photonic interface that lets the molecule talk directly to light. NVision's team screened more than 150 candidate molecules with AI, synthesized 40 in the lab, and landed on a triplet ground-state carbene, a carbon with two unpaired electrons, embedded in a structurally matched host crystal and photoactivated in place with light. In numbers: a spin coherence time of 2.2 milliseconds under dynamical decoupling, spectral stability over more than an hour, and optical spin-selective transitions 400 times narrower than their separation. No molecular qubit has ever been controlled this precisely. That is not hype; it is a peer-reviewed first.

On the same day, NVision announced a partnership with HyperLight, a maker of thin-film lithium niobate photonic chips, to deposit these molecules as thin films, OLED-style, directly onto standard photonic integrated circuits. That pitch is irresistible: molecules are identical by definition, they can be grown in films like display materials, and they operate at 2 to 4 Kelvin, temperatures a closed-cycle cryocooler reaches without a dilution refrigerator, which is another way of saying quantum computing that plugs into the existing semiconductor foundry supply chain instead of demanding a new one. Founded in 2015 in Ulm and backed by roughly $120 million from investors including Playground Global, DCVC, and Abbott Laboratories, the company is targeting a live on-chip demonstration before the end of 2026.

Then comes the roadmap, and this is where you should put down your coffee. NVision says it will scale to 1,000 physical qubits, "progressing toward 1,000 logical, fault-tolerant qubits on chip by 2030," in a deployable data-center form factor. Read those two numbers again. One thousand physical qubits becoming one thousand logical qubits is a ratio of roughly 1:1, and if that number did not make you pause, you have not been paying attention to what fault tolerance costs.

A Ratio That Breaks Fault Tolerance

Fault tolerance is the tax every quantum computer pays to be useful. Physical qubits are noisy, so you encode each logical, error-corrected qubit across many physical ones, trading raw qubit count for reliability in a deal that has governed every serious quantum architecture for three decades. Under the standard surface code, the overhead runs roughly 100 to 1,000 physical qubits per logical qubit at good error rates, and it gets dramatically worse as error rates climb. A roadmap that ends at 1,000 logical qubits from 1,000 physical qubits implies an overhead of about one, a ratio that would make every error-correction researcher in the world stop what they are doing if anyone could show the mechanism. No published error-correction scheme on any platform achieves that. Nobody has explained which one NVision intends to use. Not in the paper, not in the release.

Worse, the demonstrated fidelity does not clear the bar where error correction starts working at all, because at a 1% error rate you are standing on the threshold looking in, not safely below it. Storage fidelity follows the paper's own formula, F = (1 + e^(−t/T₂))/2, with T₂ = 2.2 ms. Run the numbers and the picture gets stark fast:

Target storage fidelity Maximum storage time Status
99% 40 microseconds Demonstrated
99.9% 4.4 microseconds Google Willow-class
99.99% 0.44 microseconds State of the art (silicon spins)
99.999% 0.04 microseconds Fault-tolerance territory

A 99% fidelity is a 1% error rate. Google's Willow milestone, "below threshold" operation demonstrated in Nature in late 2024, required physical error rates well under 1% before error correction produced any net benefit at all. NVision is sitting at or above the threshold where error correction even begins to help, which means the fanciest decoder in the world cannot rescue these qubits until the raw physics gets roughly ten times better. Closing the gap to Willow-class fidelity is a 10x improvement in error rate, and every order of magnitude in quantum error rates has historically taken the field roughly five years of grinding materials science, better control electronics, and quieter environments. Reaching the state of the art, five silicon spin qubits at 99.9% simultaneous gate fidelity reported by RIKEN, needs another 10x on top of that. Hitting the fidelities where 1,000 logical qubits becomes remotely plannable is a 100x improvement, the kind of leap that took superconducting qubits from the first transmons in 2007 to Google's below-threshold demonstration in 2024. That is not a manufacturing problem; it is a physics problem with a deadline, and that deadline is 2030.

None of this makes the molecule unimpressive; it makes the roadmap the thing to interrogate. NVision has demonstrated the most manufacturable qubit in history at the least fault-tolerant fidelity in the field, a combination that is either the setup for the greatest scaling story in quantum computing or a category error between what is easy to make and what is possible to compute with. Four years to 2030 must deliver the fidelity, not just the factories.

Where the Numbers Actually Favor NVision

Credit where it is due: two of the company's claims survive contact with the math.

First, the cryogenic advantage is real and quantifiable. Superconducting qubits need dilution refrigerators reaching roughly 10 millikelvin, room-sized systems costing on the order of $500,000 to $1 million each. NVision's molecules work at 2 to 4 Kelvin, which a closed-cycle cryocooler handles for tens of thousands of dollars in a rack-sized package: no room-sized dilution refrigerator, no million-dollar price tag, no PhD required to keep it cold. That is a genuine 10-to-20x capex advantage on cooling alone, before you count power, footprint, and maintenance. "Accessible cryogenic temperatures" is the one press-release phrase that cashes out in hard dollars today.

Second, the 70% photon-loss tolerance is a serious engineering number, not marketing gloss. It means the architecture works with 30% end-to-end optical efficiency. Thirty percent: that is all the light that needs to survive. Standard silicon-photonic coupling losses run 1 to 3 dB per interface, roughly 20 to 50% loss per hop, which puts a 70%-tolerant design squarely in the range of off-the-shelf photonics rather than heroic custom optics. That is precisely why the HyperLight partnership is technically coherent: thin-film lithium niobate chiplets plus a loss-tolerant emitter is a combination you can actually build. First prototype chips are expected this quarter.

There is also a density argument nobody is making loudly enough. Because the molecules are addressed optically, the per-chip qubit density is limited by diffraction, not by chemistry, which inverts the usual scaling problem: the hard part is no longer making the qubits, it is pointing light at them precisely enough. A square centimeter of chip holds on the order of 10^8 diffraction-limited optical spots at micron pitch, a ceiling so far above today's single demonstrated molecule that the interesting question is not whether the molecules fit, but whether the optics can keep up. Molecules are nanometer-scale and identical, which means the addressing optics, not the qubit fabrication, is the scaling bottleneck. That is a much better bottleneck to have than the one superconducting qubits face, where every qubit is a hand-tuned snowflake of Josephson junctions.

Limitations

The Nature paper demonstrates one individually addressable molecule. Not a chip, not an array, not a processor. No two-qubit gate has been demonstrated on this platform, which means the single most important operation in quantum computing, the entangling gate that turns a collection of isolated spins into a computer, remains entirely theoretical for molecular qubits. Entanglement between two molecules has not been shown either. Researchers measured the 2.2 ms coherence time with XY8 dynamical decoupling, a pulse sequence that actively refocuses noise, not the idle coherence a real computation would experience, so the number that matters for algorithms is shorter than the headline. That 99% figure is a storage-fidelity approximation from the paper's own formula, not a randomized-benchmarking gate fidelity, so it is not directly comparable to RIKEN's 99.9% gate numbers, and treating the two as interchangeable is like comparing a car's top speed on a dynamometer to its lap time. NVision has not disclosed its current physical qubit count; the 150-to-40 molecule screening funnel describes a materials pipeline, not devices, which means the distance from today's single demonstrated molecule to the roadmap's first milestone is measured in unknowns, not engineering steps. And the cryocooler cost figures above are industry benchmarks, not NVision disclosures. Finally, the 1:1 physical-to-logical reading is my interpretation of the press release's roadmap language. NVision has not published its error-correction architecture, which is exactly the document the field needs to see.

Strongest Counterargument

Everything above grades NVision against the surface code, and that may be the wrong exam. Photonic fusion-based quantum computing, the architecture PsiQuantum is betting its billions on, does not pay surface-code overhead the way superconducting machines do. It builds large entangled states by fusing small photonic resource states, and its error-correction math is fundamentally different, with overhead that can look far closer to 1:1 than anything in the superconducting world.

If you view NVision through that lens, the whole picture inverts. What fusion-based QC desperately needs is a factory for identical, tunable, manufacturable single-photon emitters, and emitter non-uniformity is precisely what has tortured NV centers and quantum dots for a decade. NVision's molecules are identical by construction, chemically tunable by design (the arXiv architecture paper shows isotopic labeling with carbon-13, nitrogen-14, fluorine-19, or phosphorus-31 to engineer nuclear spin registers), and depositable as thin films. A 70% loss tolerance is specifically a linear-optics number. Under surface-code accounting the roadmap's 1:1 ratio looks absurd, but it starts to look like a company that has already picked its error-correction architecture and is describing the endpoint rather than the path. And first-demonstration fidelities are always the worst number a platform ever posts, which is worth remembering before writing the obituary for a molecule that did not exist as a qubit eighteen months ago. NV centers started far worse than this and became the benchmark for solid-state spin coherence, so the history of this field says you bet on the platform with the manufacturing path and wait for the physics to catch up.

This counterargument is strong enough that it should be the company's actual pitch, because right now NVision is letting skeptics grade it against the surface code while its real bet, if the 70% loss-tolerance number is any indication, is on an entirely different error-correction paradigm. If NVision's 2030 target assumes fusion-based error correction, say so, publish the overhead analysis, and let the field check the arithmetic, because a roadmap without a published overhead analysis is a fundraising narrative, and this field has seen enough of those to last a lifetime. Until then, the roadmap is a claim without a mechanism.

What You Can Do

If you invest in quantum: separate the molecule from the roadmap. Peer-reviewed and de-risked, the molecule stands on its own. The roadmap, meanwhile, requires a 10-to-100x fidelity improvement and an undisclosed error-correction scheme. Before pricing the 2030 claim, ask for two things, a randomized-benchmarking two-qubit gate fidelity and the name of the error-correction architecture, because any quantum hardware company that cannot produce both is selling a timeline, not a plan, and timelines are cheap.

If you are a student or engineer choosing where to aim: the genuinely new career here is at the intersection of synthetic chemistry and quantum information. NVision's funnel, 150 AI-screened molecules down to 40 synthesized, is a hiring signal. Designing qubits the way medicinal chemists design drugs, tuning coherence, emission wavelength, and spin properties atom by atom, did not exist as a discipline five years ago. It does now.

If you worry about quantum computers breaking encryption: stand down. At 99% fidelity with no demonstrated two-qubit gate, Shor's algorithm on RSA-2048 remains decades away on this platform. "Millions of qubits on foundry chips" is a manufacturing vision, not a cryptographic timeline. Watch the first published two-qubit gate fidelity, not the qubit count. That is the honest signal.

The Bottom Line

NVision has built something real: the first manufacturable molecular qubit, published in Nature, with a credible path onto photonic chips through a serious photonics partner, and that achievement alone would have been the biggest quantum hardware story of the month in any year that did not also come with a roadmap attached. Cryogenic economics check out. Loss tolerance checks out. Chemistry-as-qubit-design is a genuinely new paradigm, and the field should take it seriously. But 1,000 physical qubits do not become 1,000 logical qubits by 2030 without an error-correction miracle the company has not described, and a 1% error rate does not get you into the fault-tolerant regime at all. Its molecule cleared peer review. The roadmap has not. Watch for the two-qubit gate fidelity, that is the number that will tell you whether this decade's quantum dark horse is real.

Related

Sources