A wave of light based quantum computers just started shipping to real data centres, no room sized freezer required. They are still tiny and unproven, but the shape of the machine might turn out to matter more than the qubit count everyone quotes.
Picture the quantum computers you have seen in photos: a column of gold rings hanging upside down like a chandelier, sealed inside a locked chamber and chilled colder than deep space by a machine called a dilution refrigerator. It is often the single most expensive and most fragile part of the whole system, and it can take days just to cool down.
Now picture an ordinary server rack, the kind humming in any data centre on Earth. In July 2026, a small Dutch company slid a working quantum computer into a rack that looked almost exactly like that, shipped it to a customer in Germany, and switched it on without a drop of liquid helium anywhere in the building.
It cannot do anything useful yet. But it is a real clue about what quantum computing might look like once it finally leaves the physics lab.
Here is what happened
The machine is called Carina, built by QuiX Quantum, a company spun out of the Netherlands' University of Twente in 2019.
What shipped. QuiX announced Carina on July 14, 2026: a photonic quantum computer, meaning it computes with particles of light rather than the chilled metal circuits or trapped atoms that rival machines use, built for real deployment rather than a lab bench. It now sits at DLR, Germany's national aerospace research centre, in a government funded project meant to prove a photonic machine can run a full, general purpose set of quantum operations at room temperature, not just one narrow trick.
It is not alone. Canada's Xanadu switched on Aurora in January 2025, wiring four server racks together with 13 kilometers of ordinary fiber optic cable so they worked as one 12 qubit computer. France's Quandela delivered its own 12 qubit machine, Belenos, to a national supercomputing centre. Britain's ORCA Computing has shipped rack mounted, room temperature photonic processors since 2022, tuned for narrower problems.
The giant in the wings. PsiQuantum, in the United States, is skipping the small machines and aiming straight for roughly a million qubits at a site in Australia by the end of 2027.
A research twist. Days before Carina shipped, a team at Imperial College London published a photonic chip called Clavina, no relation despite the similar name, showing that new functions can be added to one reprogrammable light circuit later, rather than etching a fresh chip from scratch each time.
How it works
Every quantum computer needs a physical object to carry its qubit, the basic unit of quantum information, like a bit that can hold a mix of states at once. The object you pick decides almost everything else.
Light does not mind the heat. Superconducting machines, the kind IBM and Google build, use tiny loops of metal that only behave like qubits a few hundredths of a degree above absolute zero, inside that chandelier shaped refrigerator. Photonic machines use photons, the individual particles of light, which are already comfortable at room temperature. A photon just needs a smooth glass channel etched onto a chip to travel through. No chandelier, no liquid helium, no vibration proof clean room.
The maths runs by measuring. QuiX and PsiQuantum both use an approach called measurement based computing. Instead of pushing qubits through a strict sequence of logic steps, they weave photons into a large tangled web called a cluster state, then measure the photons one at a time, and each result decides how the next one is read, a little like a choose your own adventure book where every page changes what the next page says.
Scaling looks like networking. Because photons already travel through fiber optic cable, wiring two photonic computers together looks like plugging in another network switch, not redesigning the machine. That is how Xanadu turned four racks into one computer, and it is the logic behind hopes that these systems can grow by adding boxes rather than building one impossibly large fridge.
Why it matters
Nobody has yet built a quantum computer that reliably beats an ordinary computer at a task anyone actually needs done. That is the real prize: simulating new medicines atom by atom, untangling planning problems too knotty for classical machines, and one day breaking today's encryption.
The finish line is fault tolerance. Nearly every serious approach is racing toward the same milestone: a machine that can catch and fix its own errors fast enough to trust a long calculation. Most researchers think that needs upward of a million physical qubits checking each other's work, a process called error correction.
Four technologies, one race. Google's Willow chip showed in late 2024 that adding more superconducting qubits can shrink errors instead of multiplying them. IBM has pushed past 1,100 physical qubits and, weeks after Carina shipped, claimed a 74 qubit calculation no known classical method can match. Quantinuum turned 98 charged atoms into 48 error corrected logical qubits, while QuEra reached 96 logical qubits from 448 neutral atoms.
Photonics has a record too, and China owns it. In May 2026, Chinese researchers unveiled Jiuzhang 4.0, juggling more than 3,000 photons to solve one narrow maths problem in about 25 microseconds that would tie up the world's fastest supercomputer for an age. It is a genuine, peer reviewed result. It is also a party trick: the problem was chosen because light happens to be good at it, not because the answer matters to anyone.
The quiet bet underneath. What makes QuiX and Xanadu worth watching is that neither is chasing one record breaking chip. Both are betting that many modest, networked, room temperature boxes can add up to more than one heroic refrigerator ever could, the way data centres got powerful by wiring together thousands of ordinary servers instead of building one giant computer.
The honest catch
Every announcement here deserves a second look. This is what the press releases leave out.
These machines are tiny. Carina, Aurora, and Belenos each work with roughly a dozen qubits or fewer, nowhere near the millions that would make them useful beyond a demonstration.
Room temperature does not mean no cooling. Several photonic systems, PsiQuantum among them, still chill their light detectors to a few degrees above absolute zero, using coolers far simpler than a dilution refrigerator, but real cryogenics all the same.
Photons are slippery. They are hard to make exactly on demand and easy to lose in transit, and solving that at scale is one of the field's toughest open problems. It is why PsiQuantum's 2027 target for a million qubit machine is such an aggressive deadline in an industry famous for missing them.
EDITOR'S TAKE
Photonic quantum computing's real pitch is not a bigger chip, it is a different shape: modest, networked, room temperature boxes instead of one giant custom refrigerator. That is a genuinely credible engineering argument, and 2025 and 2026 turned it from a slide deck into hardware you can walk up to. But a box that fits in a server rack is not the same as a computer that does something new. Watch whether QuiX's Dedalo roadmap, Xanadu's next machine, or PsiQuantum's Australian build can multiply their qubit counts a hundredfold without multiplying their error rates just as fast. Until one of them does, this is the most credible infrastructure bet in quantum computing, not yet a quantum computer that changes anything real.
Quick questions
What is a photonic quantum computer, in plain terms?
It is a quantum computer that stores and processes information using photons, the individual particles that make up light, instead of the tiny superconducting circuits or trapped atoms that rival machines use. The photons travel through hair thin glass channels etched onto a chip, much like the fiber optic cable that already carries internet traffic. Because photons barely interact with their surroundings, they can work at room temperature rather than needing extreme cooling. QuiX Quantum, PsiQuantum, Xanadu, and Quandela are all building versions of this approach.
Why does skipping the dilution refrigerator matter so much?
A dilution refrigerator is the chandelier shaped machine that cools superconducting quantum computers to within a hair of absolute zero, and it is one of the most expensive and fragile parts of that whole approach. Skipping it lets a photonic quantum computer sit in an ordinary server rack, draw far less power, and avoid the days long process of cooling down or warming up. It also makes networking several machines together more practical, since photons already travel well through standard fiber. Some photonic systems still need modest cooling for their light detectors, just nowhere near as extreme.
Is any of this actually useful yet?
Not yet, and nobody serious claims otherwise. Today's photonic machines, like QuiX's Carina and Xanadu's Aurora, work with roughly a dozen qubits or fewer, while most researchers believe useful, error corrected quantum computing needs upward of a million. What is new is that these are working, deliverable machines rather than one off lab demonstrations, and their modular, networked design gives photonics a plausible, if unproven, path to that far larger scale. The honest summary right now is early infrastructure, not a finished tool.
Sources
QuiX Quantum Announces Carina, the First Universal Photonic Quantum Computing Architecture for Commercial Deployment: Original announcement coverage, including the DLR deployment in Germany.
With The Carina System, QuiX Pushes Photonic Quantum Computing Forward: Detailed technical account of Carina and the Dedalo scaling roadmap.
Extensible universal photonic quantum computing with nonlinearity: Peer reviewed Nature Photonics paper describing the Clavina reprogrammable architecture from Imperial College London.
Xanadu introduces Aurora, world's first scalable, networked and modular quantum computer: Company announcement of the 12 qubit, four rack Aurora system, later published in Nature.
Chinese Scientists Develop Jiuzhang 4.0, Setting New World Record in Quantum Computing: Chinese Academy of Sciences release on the 3,000 photon boson sampling result.
Frontier Signal explains frontier technology in plain English. Company and agency figures should be independently verified. This is general information, not investment or professional advice.
