On November 10, 2026, Xanadu and HyperLight announced that they had reduced the insertion loss of an optical switch used in photonic quantum computing to 7 mdB (0.007 dB). An optical switch is a component that redirects the path of the photons that carry quantum information. Compared with the roughly 20 mdB the companies reported in 2025, the loss figure has fallen to about one-third. Converted to transmission, 7 mdB corresponds to about 99.84%, meaning only about 0.16% of light is lost in a single pass. Lower loss of this kind could be an important step toward scaling up photonic quantum computers, which use large numbers of optical switches. However, what was announced is a measurement result for a single component, not a demonstration that the performance of an entire quantum computer has improved.
Optical switch loss reduced from about 20 mdB to 7 mdB
The companies developed the switch using thin-film lithium niobate (TFLN). In their November 10 announcement, Xanadu and HyperLight explained that they used HyperLight's "TFLN Chiplet" platform to demonstrate the 7 mdB loss on a wafer production line compatible with high-volume manufacturing. This figure was reported as a measurement from actually fabricated components, not as a simulated prediction.
In the July 29, 2025 announcement used for comparison, the switch's insertion loss was given as about 20 mdB. That result was also presented as having been obtained at a facility compatible with mass production, so the new progress lies not in using a volume-capable manufacturing base for the first time, but in achieving a lower loss figure on a similar base. It has not been made clear, however, whether the measurement conditions in 2025 and 2026 were identical. The difference between the published values therefore does not necessarily represent an improvement measured under the same conditions. In addition, the announcement materials do not point to an academic paper corresponding to the 7 mdB measurement, so for now this is a report based on a corporate press release.
Why photon loss matters in photonic quantum computing
Insertion loss is a metric for the loss that occurs when a component is placed in a light transmission path. For optical switches, it is used to assess how much light can pass through without being lost when the photon's path is switched. The smaller the number, the less loss there is and the more efficiently light is transmitted.
In photonic quantum computing, photons themselves carry the quantum information, which creates constraints that differ from ordinary optical communications. In conventional optical communications, signals weakened in transit can be boosted with devices such as optical amplifiers, but applying the same method directly to photons carrying quantum information risks destroying the quantum state. Photonic quantum computers therefore need to guide photons accurately to their intended paths while minimizing the photons lost along the way. This quantum-specific constraint is why Xanadu and HyperLight place such emphasis on lowering switch loss.
Furthermore, when a photon passes through multiple switches, the loss from each element accumulates. Even a slight difference per switch can lead to a large difference in the fraction of light that ultimately remains as the number of elements traversed grows. For scaling up photonic quantum computers, what matters is not only the performance of individual elements but also keeping the cumulative loss low when many elements are combined.
7 mdB equals 99.84% transmission, and the gap widens over 100 passes
mdB stands for millidecibel, and 1 mdB equals 0.001 dB. Last year's roughly 20 mdB is therefore about 0.020 dB, and the new 7 mdB is 0.007 dB. Because the decibel is a logarithmic scale, 7 mdB does not mean 7% of the light is lost. If the loss in dB is and the light transmittance is , the relationship is given by the following formula.
\[ T = 10^{-L/10} \]
Using this formula, the transmission at 7 mdB is about 99.84%, and at 20 mdB it is about 99.54%. The table below was calculated by the XenoSpectrum editorial team based on the 2025 published value and the 2026 published value. The figures for 100 passes assume light passes through 100 switches with the same loss and that no other losses occur. This is not a performance comparison test premised on identical measurement conditions in both years.
| Date announced | Switch insertion loss | Transmission after one pass | Transmission after 100 passes (assumed calculation) |
|---|---|---|---|
| July 29, 2025 | About 20 mdB (about 0.020 dB) | About 99.54% | About 63.1% |
| November 10, 2026 | 7 mdB (0.007 dB) | About 99.84% | About 85.1% |
The fraction of light lost in one pass is about 0.46% at 20 mdB and about 0.16% at 7 mdB. The difference, about 0.30 percentage points, is small, but it becomes significant when the same loss occurs repeatedly. Over 100 passes, the cumulative loss equals about 2.0 dB and 0.7 dB respectively, and the transmission can be calculated with the following formula.
\[ T_{100} = 10^{-100L/10} \]
Under this assumption, about 63.1% of the light remains after passing through 100 switches at 20 mdB, but 85.1% is retained at 7 mdB. In other words, a transmission difference of about 0.3 percentage points for a single switch grows to about 22 percentage points after 100 passes. The significance of the reported loss reduction lies in this suppression of cumulative loss. However, these are simple conversion examples using published insertion loss values, not experimental results or simulations of quantum computer operation. They do not indicate that an actual device would use 100 stages of switches, nor do they predict improvements in photon count, computation speed, or accuracy.
In its 2025 announcement, Xanadu and HyperLight also reported waveguide propagation loss of under 2 dB/m. That is the loss that occurs as light travels a given distance through a waveguide, a different metric from the insertion loss of an optical switch. The new waveguide propagation loss in dB/m cannot be back-calculated from the 7 mdB value, so it is important not to confuse the two.
Challenges remain for mass production; effect on whole quantum computers untested
Xanadu and HyperLight also assessed that they achieved high levels of manufacturing yield and reproducibility this time. Yield refers to the proportion of manufactured elements that meet defined quality standards. However, no specific yield figures or distribution of insertion loss across multiple elements have been published. Using a production line compatible with high-volume manufacturing needs to be evaluated separately from being able to ensure stable quality across many mass-produced elements.
It is also unclear whether the reported 7 mdB is an average of the measured elements, a representative value, or the best result. The number of elements measured and the measurement error are not given, nor are the wavelength of light used, the operating temperature, or the switch drive conditions. To judge how low a loss can be expected in mass production, data linking such conditions to measurement results from multiple elements is needed. As for the companies' assessment of reproducibility, the materials currently available show no results of independent replication by third parties.
Xanadu CEO Christian Weedbrook has expressed expectations that the achievement will improve the economics of manufacturing costs and allow device footprints to shrink. However, this is an outlook for future effects, not a result demonstrated in this experiment. What was measured is a characteristic of a single component, the optical switch's insertion loss, and how much that improvement affects the manufacturing cost, footprint, or computational error rate of a whole quantum computer remains unclear.
Suppressing photon loss is one of the key technical challenges for building complex optical circuits and scaling up photonic quantum computers. The 7 mdB figure shows component-level progress on that challenge. For practical use, however, it will be necessary to confirm whether the same performance can be stably reproduced under the required operating conditions and whether low loss can be maintained when many elements are combined. If performance distributions across elements and verification results for entire quantum computers are published, it should become possible to assess more concretely how far this result translates into actual computing performance and manufacturing advantages.