When a quantum startup opens a semiconductor fab, most people read it as marketing. The press release uses words like "groundbreaking" and "milestone", and the stock photos show a clean room. But the real story is buried in three numbers that rarely make the headline: a yield jump from 40 percent to over 70 percent, a target of 10,000 devices per year, and a test step that screens a qubit in minutes. Those are the numbers that tell you whether a company is doing physics or doing manufacturing. They also tell you something important about the state of the whole quantum industry, which has spent the last decade proving qubits can exist and is only now figuring out how to make many of them.
Quandela, the Paris-based quantum computer company, inaugurated its first manufacturing pilot line on September 11, 2026, at the Photovoltaic Institute of le-de-France (IPVF) inside the Paris-Saclay innovation cluster. The facility merges integrated photonics with semiconductor quantum-dot devices, which serve a double role: the quantum dots act as spin qubits for storage, and the same structures emit the single photons that carry information between them. According to the company press release, the pilot line will produce more than 2,000 devices over the next two years before reaching 10,000 per year at full capacity, with qubit device density climbing toward hundreds of devices per square millimeter.
This is not a vague statement about "scaling quantum." It is a specific claim about a production line, and the production line is where every quantum hardware team eventually dies or learns to live.
Why the Fab Matters More Than The Qubit Count
The quantum computing field has a public relations problem. It measures itself in qubit count, and the qubit count is a poor proxy for usefulness. A superconducting processor with 1,000 physical qubits that cannot be manufactured consistently is less useful than a photonic processor with 12 qubits that can be built identically, one after another, forever.
That is the core tension in Quandela's announcement, and it is worth sitting with. The company's systems, Belenos and Canopus, carry a modest number of qubits by headline standards. Belenos, launched in May 2025, is a 12-qubit photonic system. Canopus, its successor, is slated to double that to 16 qubits. By comparison, the leading superconducting chips fielded over a thousand qubits in the same window. If you are scoring the race with a spreadsheet that counts qubits, Quandela is losing badly.
But quantum engineering is not a qubit-counting sport, and anyone who treats it as one is misunderstanding what makes the hardware hard. The hard part is not creating a single qubit. You can create one qubit in a laboratory with a laser and a cryostat. The hard part is creating ten thousand identical qubits, each one performing within the same tolerance, with a coherence time that does not vary wildly from chip to chip. That is a manufacturing problem, and manufacturing problems are solved on production lines, not in research papers.
The reason this matters is that the two dominant qubit platforms, superconducting and photonic, face fundamentally different scaling problems.
| Dimension | Superconducting qubits (transmon) | Semiconductor quantum-dot spin qubits (Quandela approach) |
|---|---|---|
| Qubit mechanism | Superconducting loop with Josephson junction | Electron spin in a quantum dot, read out optically |
| Photon emission | None; requires microwave control lines | Intrinsic single-photon source from the dot |
| Operating temperature | Millikelvin dilution refrigerator (~15 mK) | Qubit near cryogenic; photonics and detection near room temperature |
| Interconnect | Microwave coaxial wiring, dense but lossy | Photonic interconnect, low loss, high bandwidth |
| Yield bottleneck | Wiring and crosstalk at scale | Qubit uniformity and photon indistinguishability |
| Manufacturing leverage | Existing CMOS-ish fabs | Existing semiconductor fabs, proven design history |
The table hides a lot of subtlety, but the shape is right. Superconducting chips are essentially a wiring nightmare at scale, and every qubit needs a bundle of coaxial cables running back to room-temperature electronics. Photonic and spin-qubit platforms like Quandela's use light to connect, which is more naturally compatible with the dense, standardized manufacturing that the semiconductor industry already mastered. The trade-off is that photonics and spin qubits are far more sensitive to fabrication uniformity, which is exactly the problem the pilot line is built to solve.
The Manufacturing Numbers, Read Like An Engineer
Here is where I put on the engineer hat, because the pilot line's specifications are genuinely interesting if you know how to read them.
The headline yield improvement, from 40 percent to over 70 percent, is a large jump on an established process. In semiconductor terms, moving a yield by 30 percentage points on the same line usually means you found and eliminated a dominant failure mode. You stopped throwing away chips for one identifiable reason. That is not incremental; it is the difference between a lab curiosity and a factory. For a qubit device, a 70 percent yield means the fabrication process is now deterministic enough to plan around. You can order 10,000 devices and expect roughly 7,000 to work, instead of gambling on which wafer will be good.
Then there is the qubit identification tool. The press release describes a machine that probes hundreds of nanometer-sized structures, maps their uniformity, and tests coherence time at cryogenic temperatures in a few minutes. This is the part most people will gloss over, but it is arguably the most technically significant sentence in the whole announcement, and here is why.
The fundamental problem with making qubits from quantum dots is that quantum dots are defined by self-assembly and lithography, both of which introduce tiny variations. A quantum dot is supposed to be a precise structure, but in practice the dots come out with slightly different sizes and positions, and those differences change the qubit's properties. Coherence time, the duration for which a qubit keeps its quantum state, is exactly the property you cannot afford to vary. If your qubits have wildly different coherence times, you cannot calibrate a processor, because each qubit behaves differently.
The identification tool turns this from an unsolvable materials problem into a sorting problem. You do not need every dot to be perfect. You need a fast way to measure the dots you have and keep the good ones. That is a classic manufacturing approach: test everything, bin by performance, ship the winners. It is the same reason the entire semiconductor industry works the way it does. The tool that "collects high-volume data while mapping uniformity" is the feedback loop that makes a fab actually a fab. Research happens once. A fab iterates, measuring thousands of devices and feeding that data back to tighten the next run.
Two engineering consequences follow. First, the density target of hundreds of devices per square millimeter is meaningful only if yield and identification hold. Density without yield is just a bigger pile of dead chips. Second, the combination of cryogenic coherence testing plus room-temperature photonics is what makes the photonic path architecturally attractive for the datacenter, which is where the company says its systems are headed. You can keep the cold part small and the hot part cheap, which is an economics argument as much as a physics one.
What This Means For The Quantum Landscape
Stepping back to the data-science side of the lens, the pilot line is a signal about where the industry's bottleneck has moved. For most of the last decade, the constraint on quantum progress was physics: could we even make a qubit that behaves, and could we entangle several of them? That question is largely answered for a handful of platforms. The new constraint is manufacturing: can we make millions of qubits cheaply and consistently?
The data supports that reading. The same semiconductor fabs that build the GPUs and AI accelerators you read about elsewhere on this site can, in principle, build these devices. That is why the memory and chip stories dominate tech news, and why the quantum bottleneck is increasingly a fab story rather than a physics story. When a quantum company's hardest problem becomes yield and uniformity, it has stopped competing with physicists and started competing with TSMC and Samsung, the same supply-chain dynamics driving the HBM4 memory shortage and wafer cost curve right now.
There is also a strategic angle worth flagging. The company's quantum-dot knowledge dates back twenty years in French telecom laboratories and CNRS, transferred to Quandela in 2017 and commercialized since 2018. That twenty-year runway is the moat. A startup that buys a microwave fridge and some lithography cannot replicate this overnight, because the know-how for making quantum dots that emit indistinguishable photons is not in a textbook. It is in the process recipes, the calibration data, and the tacit understanding of how to run a line that produces consistent results. The pilot line is the physical embodiment of that accumulated knowledge.
Honest Gaps And What To Watch
Before this becomes a victory lap, three caveats are in order.
First, the numbers are vendor-reported. The 70 percent yield, the 10,000-device target, and the density figures all come from Quandela's own press release. There is no independent benchmark, and there never will be for a company's internal manufacturing metrics. Treat them as the company's self-assessment, not as audited fact. A yield claim of "over 70 percent" is also oddly precise in a way that suggests it may be the yield for a specific device class, not the whole product line, so read it as a directional signal.
Second, the system scale is still tiny. Twelve to sixteen qubits is a research-grade processor, useful for specific photonic algorithms and quantum machine learning demos, nowhere near the millions of error-corrected qubits needed for problems that classical computers cannot handle. The pilot line is about making those qubits producible, but producibility at a small qubit count is not the same as usefulness at a large qubit count. The physics of error correction and crosstalk at scale is a different mountain.
Third, photonic qubits carry their own trade-offs, notably the difficulty of deterministic two-qubit gates, which is why many photonic schemes lean on measurement-based or linear-quantum-computing models. That is a genuine architectural limitation, not a marketing afterthought, and it is why the photonic path is a bet on a particular style of algorithm rather than a universal solution.
The metrics to watch over the next twelve months are the same ones any builder should track: when Canopus ships, whether the qubit count doubles as promised, what the reported coherence times actually are, and whether independent groups can reproduce the photon indistinguishability numbers. The pilot line tells you the company is serious about manufacturing. It does not yet tell you whether that manufacturing will be enough to win.
References
- Quandela press release, "Quandela's Pilot Line Launch: A New Semiconductor Manufacturing Site" (September 11, 2026). Primary source, company announcement of the IPVF pilot line, yield and throughput figures. See Quandela pilot line announcement.
- Quandela, "Belenos, the world's most powerful photonic quantum computer" (May 22, 2025). Company source on the 12-qubit system and the roadmap to Canopus. See Quandela Belenos launch.
- Nature Light, "Spin-photon qubits for scalable quantum network" (July 3, 2026). Peer-reviewed background on solid-state quantum light sources and the coupling of stationary spin qubits to photonic qubits. See Spin-photon qubits paper.