How many photons can a quantum computer afford to lose before a calculation falls apart? Fewer than most people outside the lab would guess. Xanadu and HyperLight say they have pushed that number down again. The two companies announced on November 10, 2026, a new photonic switch with just 7 millidecibels of loss, down from the 20 millidecibels they reported in mid-2025. Toronto-based Xanadu made the announcement jointly with HyperLight, a Massachusetts company that builds the switch’s underlying chip material. Nothing here has been checked by an outside lab yet. Everything below comes from the companies’ own announcement and the figures they chose to report.
The Switch Loss Numbers, Explained
- 2025 benchmark: 20 millidecibels of switch loss, announced in July 2025 and described at the time as among the lowest ever reported for the application.
- 2026 benchmark: 7 millidecibels, announced November 10, 2026, about 16 months later.
- The improvement: Xanadu calls this a threefold cut. The company also says it beats “the closest competitor’s most recent published result” by a factor of ten, though it does not name that competitor anywhere in the announcement.
- Where it was measured: on a high-volume wafer manufacturing line, not a one-off lab sample. The companies say that points to repeatable production rather than a cherry-picked chip.
A millidecibel is a sliver of a sliver. Decibels already measure a ratio on a logarithmic scale, so a drop from 20 to 7 sounds modest on its own. It matters more once you remember that a working photonic quantum computer chains hundreds or thousands of these switches together. Small losses at each one add up fast across a full circuit.
Why Light Can’t Simply Be Amplified
Photonic quantum computers carry information on individual particles of light instead of electrical charge. That creates a problem ordinary fiber-optic networks don’t have. A regular internet signal gets boosted partway through its trip by an amplifier. A quantum signal can’t take that shortcut. Any device that tried to strengthen a single photon’s signal would also collapse the fragile quantum state riding on it, destroying the very information the system is trying to process.
That makes every lost photon a dead end, not something to recover later. Lower switch loss means fewer photons disappear on the way through the circuit. Engineers then need less backup hardware, things like extra lasers or duplicate pathways, just to compensate for what the switches lose. Xanadu and HyperLight frame the 7 mdB figure as a step toward needing less of that overhead.
Built on an Existing Line, Not a New One
The switch isn’t a lab curiosity built from scratch. HyperLight makes it from thin-film lithium niobate, a material its factories already use for optical hardware inside AI data centers and telecom networks. Running the quantum switch through that same production line, instead of a dedicated one-off process, is what let the companies test it at manufacturing scale rather than on a handful of hand-picked chips.
The announcement also sits inside a wider pattern. Xanadu built this generation of hardware on the same TFLN platform behind Aurora, the fiber-networked photonic quantum computer it unveiled last year. A month earlier, in October 2026, Xanadu signed a separate manufacturing partnership with GlobalFoundries to industrialize other parts of its hardware stack. Other photonics companies are racing on a related bottleneck: how fast a chip can generate the single photons a switch like this one has to route.
“Xanadu’s quantum-compatible switches are ready to ramp to production,” said Christian Weedbrook, the company’s founder and CEO.
Why It Matters to You
Nothing here reaches a phone, a laptop, or a quantum cloud account today. This is a factory-floor measurement on one component among dozens a working photonic quantum computer still needs. The honest reason to care is about patience and cost, not immediate access. If switch loss keeps falling, and holds up as more of the chip gets manufactured, engineers need less backup hardware to build a working system. That should eventually help make quantum computers cheaper to build and run. Xanadu and HyperLight also point to a second, nearer-term audience: the AI data centers already buying similar TFLN chips for ordinary networking. A single manufacturing line serving both markets could, in theory, spread development costs across more customers, though neither company has published what that might mean for prices or timelines.
What This Announcement Doesn’t Show
- No outside verification. No independent lab has published a matching loss measurement. Everything here traces back to Xanadu and HyperLight’s own press release.
- The competitor comparison is unnamed. Xanadu’s tenfold claim against “the closest competitor’s most recent published result” names no competitor and links to no public figure a reader can check.
- A component, not a computer. This is a hardware benchmark, not a working quantum calculation. No qubit count, error rate, or completed computation came with it.
- No price or ship date. Neither company has said when a system built around this switch will be commercially available, or what it will cost.
- Not peer reviewed. This is an engineering and manufacturing claim from the companies involved, not a result that has gone through independent scientific review.
Watch for two things next. First, whether an outside lab or paying customer ever publishes its own loss measurement on this chip, not just Xanadu’s. Second, whether this switch shows up inside a working, multi-qubit photonic system instead of staying a standalone hardware demo. Until one of those happens, 7 millidecibels is a strong claim, not yet a verified one.
Sources
- GlobeNewswire: Xanadu and HyperLight Achieve Breakthrough of Nearly 3x Reduction in Optical Switch Loss (Nov. 10, 2026)
- The Quantum Insider: Xanadu and HyperLight Reduce Photonic Switch Loss to 7 mdB (Nov. 10, 2026)
- Quantum Zeitgeist: Xanadu and HyperLight Achieve 3x Loss Cut With Quantum Optical Switch (Nov. 10, 2026)
- PR Newswire: Xanadu and HyperLight Unveil Advancements in Photonic Chips (Jul. 29, 2025)
