Deutsche Telekom and Qunnect Teleport Photonic Qubits Over a 30 km Berlin Fiber Loop Using Commercial Hardware

19 Feb 2026 – Deutsche Telekom’s research and development unit, T-Labs, and the Brooklyn-based quantum networking company Qunnect said on Thursday that they had carried out quantum teleportation over a 30 km loop of deployed fiber in Berlin, with an average teleportation fidelity of 90%. The trials ran in January on Qunnect’s commercially available hardware and Deutsche Telekom’s Berlin quantum infrastructure, Deutsche Telekom said.

The partners described the trial as the first practical test of the core components that a future teleportation service would require. Fifteen researchers from Deutsche Telekom, Qunnect, Paderborn University and Orbit GmbH, with Zofia Borowska of Deutsche Telekom as first author, reported what they call preliminary results in a preprint posted to arXiv on 18 February, which has not been peer reviewed. The German Federal Ministry of Research, Technology and Space partly funded the work through its QR.N project, according to the preprint.

Quantum teleportation transfers an unknown quantum state from one particle onto another at a different location, without sending the original particle. The sender and receiver first share an entangled pair. The sender then makes a joint measurement, called a Bell-state measurement, on the particle to be sent and one half of the pair, and a classical message tells the receiver which correction to apply to the other half.

In Berlin, the input qubits came from a 795 nm laser that was locked to a rubidium reference and attenuated to the single-photon level, with a motorized controller setting each photon’s polarization. At T-Labs’ Quantum Lab, each input photon interfered with the 795 nm photon of a polarization-entangled pair from Qunnect’s Qu-SRC, a source that generates photon pairs in warm rubidium vapor. The other photon of each pair has a wavelength of 1,324 nm, in the telecom O-band.

A successful Bell-state measurement transferred the input polarization state, up to a known rotation, onto that O-band photon, which then traveled a 30 km deployed fiber loop from the Quantum Lab to a Deutsche Telekom office on Dottistraße and back. Qunnect’s Qu-APC units corrected polarization drift on the loop, and a tomography station in the same lab reconstructed the received state. The source and the compensators belong to Qunnect’s Carina product line, according to Deutsche Telekom.

The team teleported three input polarization states, according to the preprint. Averaged over the three, and counting only events in which the Bell-state measurement succeeded and the output photon was detected, the conditional fidelity was 92.3 ± 2.2% with the loop bypassed and 90.1 ± 3.3% over the 30 km loop with no other signal in the fiber.

With a 10 Gbit/s C-band data channel sharing the fiber, the authors reported 85.9 ± 4.3%, which they described as consistent with residual crosstalk from the classical band reaching the detectors. They compared all three averages with the two-thirds classical limit for teleporting an unknown single-photon qubit.

Individual states on the deployed loop ranged from 87.2% to 95.7% without the data channel and from 77.2% to 94.2% with it, according to a chart in the preprint.

The data channel was a 10 Gbit/s ADVA signal at 1,561.42 nm with a fixed output power of 1 dBm, sent between a pair of optical transceivers connected directly to the experiment’s multiplexer and demultiplexer, the authors wrote. The team recorded only one of the four possible Bell-state outcomes and applied no corrective operation to the output photon. On the deployed loop, each of the six tomography settings took three to ten minutes of data collection.

Deutsche Telekom’s release cited the 90% average and a peak of 95%. “Our fiber optic network is quantum ready,” Abdu Mudesir, the Deutsche Telekom board member for product and technology, said in the release. He said the transmission had run alongside regular data traffic. Teleportation, he said, would lay the technical groundwork for linking quantum computers at different locations, and for secure communication and European technological sovereignty.

Mael Flament, Qunnect’s chief technology officer, said the trial had moved teleportation’s building blocks out of the physics laboratory and into equipment a telecom operator installs and runs.

Deutsche Telekom said the partners would next extend the demonstration to multi-node teleportation. The company plans a teleportation showcase on its stand at Mobile World Congress in Barcelona, which runs from 2 to 5 March, and a 3 March panel with experts from Deutsche Telekom, Qunnect and TU Dresden.

The BearlinQ Testbed Behind the Trial

The trial ran on BearlinQ, the partners’ name for their quantum testbed in Deutsche Telekom’s Berlin metro fiber. T-Labs operates it from the Quantum Lab it opened in 2023.

At the OFC conference in San Francisco on 31 March 2025, the partners presented results from distributing polarization-entangled photons over 30 km of that fiber for 17 days, at 99% fidelity and 1% network downtime, according to Deutsche Telekom’s April 2025 announcement.

Matheus Sena of Deutsche Telekom and 15 co-authors published a fuller account in the December 2025 issue of the Journal of Optical Communications and Networking, after an arXiv preprint first posted in April 2025. They reported entanglement distribution over selectable looped paths from 10 m to 60 km, with the O-band quantum signal sharing fibers with C-band classical traffic, Bell-state fidelity bounds of 85–99% and Bell-test (CHSH) S-values of 2.36–2.74, where values above 2 indicate entanglement.

The authors counted downtime as time below 85% fidelity or spent recalibrating, and reported less than 1.5% on the 60 km path. A separate 78-hour run over about 100 km, without path switching or the DWDM equipment used to add classical traffic, gave fidelity bounds of 80.5–98.4%.

Qunnect had earlier run its entanglement distribution hardware on GothamQ, the company’s testbed in New York. In 2024 its researchers reported 15 days of automated distribution over 34 km of buried city fiber, with uptime above 99.8%.

Two Related Announcements Within Days

On 13 February, Photonic Inc. and TELUS announced that they had teleported quantum information over 30 km of TELUS commercial fiber into a matter-based quantum processor that can store the received information. The companies did not publish a paper or a fidelity figure with the announcement.

On 18 February, Qunnect said it had performed entanglement swapping with Cisco over deployed New York fiber. In the preprint, posted the day before with co-authors from New York University, the two entanglement sources were in separate rooms of one building at the Brooklyn Navy Yard, each linked by about 8.8 km of fiber to a hub at QTD Systems’ data center at 60 Hudson Street in Manhattan, for 17.6 km in total.

The authors measured swapping rates above 0.65 per second with a CHSH value above 2, or above 1.5 per second after correcting for detector efficiency. Qunnect’s release gave the corrected figure as 5,400 swapped pairs per hour and cited more than 99% polarization fidelity, a figure the preprint does not report for the swapped pairs.

My Analysis

Deutsche Telekom now has a documented progression on one metro testbed, from entanglement distribution with measured downtime to a teleportation interface. In less than eleven months T-Labs and Qunnect have taken BearlinQ from a conference presentation to a peer-reviewed journal paper and now a teleportation preprint. EPB, the Chattanooga utility that runs a commercial quantum network, has published operator results too: Oak Ridge National Laboratory reported more than 30 hours of uninterrupted entanglement distribution on EPB’s network, between nodes about half a mile apart, in January 2025. Berlin’s record covers longer paths and now adds teleportation, which makes it a reference point for other operators.

With the teleportation trial, the partners add an interface. A qubit on a 795 nm photon, a wavelength that rubidium-based devices use, moves onto an O-band photon inside the operator’s building and travels through the operator’s own fiber. Teleportation between two separate quantum devices, which is what a teleportation service would sell, needs a receiving qubit that stays coherent until the Bell-measurement result arrives. In this trial the output photon was measured on arrival.

Where the Bell Measurement Happened and Where the Photon Traveled

One headline had quantum data teleported 19 miles across Berlin. The photon did cover 30 km of Berlin fiber, and it finished the trip in the lab it had left. The Bell-state measurement took place at T-Labs, next to the entanglement source. The O-band photon went out to Dottistraße and came back to a tomography station in the same Quantum Lab.

Tom’s Hardware reported that the qubits never traveled through the fiber. In this configuration the Bell-state measurement ran at the start of the loop, so the O-band photon covered the full 30 km with the input state, up to a known rotation, encoded in its polarization.

Loops are standard practice in field testbeds. Qunnect’s GothamQ is one, and in 2024 Harvard and AWS entangled diamond memories in two separate labs through a 35 km loop that crossed four Boston-area municipalities. With a loop, the photons meet the temperature swings, vibration and polarization drift of deployed fiber, and the team runs a single site. A customer link would join two sites with their own equipment rooms and timing, and the Berlin trial doesn’t test that.

Borowska and her co-authors describe their result as a network’s end-node function, a “device-to-telecom state-transfer interface,” and write that reaching beyond one fiber span requires entanglement swapping between links, which this paper does not attempt. The 90% is a system-level figure for that interface, measured under four conditions:

• The input was laser light attenuated to the single-photon level, not a photon emitted by a qubit or a memory. The authors note that the laser’s residual multi-photon component lowers both the interference visibility and the fidelity.
• The team recorded only the $$|\Psi^-\rangle$$ outcome, one of four possible Bell-state outcomes, and compared each output with the target rotated by that outcome’s known correction instead of applying the correction to the photon.
• The output photon was measured on arrival. Nothing stored it.
• The team tested three input states: horizontal, diagonal and right-circular polarization.

None of this is unusual for a first field trial, and the authors state every item. The 90% therefore describes a wavelength-and-transport interface tested with laser light. It does not transfer directly to a qubit leaving a rubidium memory or atom.

The three-state test also weakens the comparison with two-thirds that the authors draw in Figure 6. Two-thirds is the best average fidelity a classical measure-and-prepare strategy can reach on a uniformly unknown single-photon qubit. For a test of three known states on orthogonal axes, the classical benchmark is higher: a device that ignored its input and always sent one fixed state would score about 79% on each, and the best classical measure-and-prepare strategy averages 5/6, about 83%. On that simplified benchmark the local and dark-fiber averages clear 83% by about four and two times their stated uncertainties, while the 85.9% with the data channel does not clear it by more than its uncertainty. Weak coherent inputs and post-selection complicate even that comparison, and the authors ran neither a decoy-state analysis nor a full process characterization that would settle it.

Deutsche Telekom’s Release Compared With the Preprint

Deutsche Telekom’s headline described a live Berlin network, and in their abstract the authors claim compatibility with multiplexed networks with live traffic. In the setup description they report something narrower: a research and development test network, run as dark fiber for the 90.1% result, with one specified 10 Gbit/s channel added for the 85.9% result from a pair of transceivers connected directly to the experiment’s multiplexer. That is a legitimate coexistence test, though not customer traffic. Mudesir’s statement put regular data traffic and the 90% average in the same sentence, but the authors measured them in separate runs, and the 85.9% is a lower central value whose uncertainty overlaps the dark-fiber result.

Deutsche Telekom also cited a 95% peak. The only deployed-loop value near that figure in the preprint’s Figure 6 is the horizontal input without traffic, at 95.7%, which appears to be the result the release rounded. In the idealized protocol, the Bell-state measurement’s heralding fixes the output polarization for a horizontal or vertical input with or without two-photon interference, so a classical measure-and-prepare strategy in the horizontal–vertical basis would reproduce that state perfectly. The diagonal and circular inputs are the ones that depend on the interference, and on the deployed loop they reached 87.2% and 87.4% without traffic, and 86.3% and 77.2% with the data channel on.

Deutsche Telekom lists quantum cryptography first among the uses of teleportation, and Mudesir tied the result to secure communication. Basic entanglement-based key distribution protocols such as BBM92 consume distributed entangled pairs directly, which BearlinQ already delivered in 2025, so this state-transfer interface adds little to them. Key distribution over longer networks may still rely on related primitives such as entanglement swapping. The Berlin interface is most useful for applications that move quantum states between devices, such as distributed computing and networked sensing.

Earlier Teleportation Experiments on Deployed and Spooled Fiber

Teleportation across several kilometers of a city’s fiber was first shown in 2016. That September, Nature Photonics published two field experiments together. Wolfgang Tittel’s group at the University of Calgary teleported the state of a 1,532 nm telecom photon onto a 795 nm photon across the City of Calgary’s dark fiber, over a straight-line distance of 6.2 km. Jian-Wei Pan’s group built a 30 km fiber network spread over 12.5 km of Hefei and teleported qubits with independent sources, entanglement distributed before the Bell-state measurement, and active feed-forward.

The Berlin team uses an independent input source, but it makes the Bell-state measurement before the entangled pair has been distributed and applies no feed-forward. The Calgary team moved a state from the telecom band onto a 795 nm photon. In 2026 an operator moves one from 795 nm into the O-band, on its own fiber, with rack-mounted commercial sources and polarization compensators and with its own engineers among the authors.

A closer precedent for connecting matter hardware to city fiber came from Saarbrücken in 2024, where Jürgen Eschner’s group at Saarland University teleported the state of a trapped calcium-ion qubit onto a remote telecom photon over a 14.4 km urban dark-fiber link. That experiment started from a real matter qubit, which Berlin’s laser input did not. Berlin’s contribution is the operator setting: commercial components, the operator’s own fiber and engineers, and coexistence measurements.

The closest precedent I know of for the coexistence result comes from Prem Kumar’s group at Northwestern, which in December 2024 teleported qubits over 30.2 km of fiber that also transmitted a 400 Gbit/s C-band channel. Kumar’s team used heralded single photons as inputs and put the Bell-state measurement at the fiber’s midpoint. The team launched the classical channel at 18.7 dBm, about 60 times the optical power of Berlin’s 1 dBm channel, on a spooled quantum link. Northwestern tested coexistence at much higher classical power on spooled fiber, and Berlin tested its interface under deployed-fiber conditions. The two experiments address different parts of the engineering problem, and an operator will eventually need both at once, on deployed fiber with a loaded C-band.

A team reporting in Light: Science & Applications in 2023 teleported weak coherent states at 7.1 events per second over a 64 km fiber channel and used decoy states to estimate a single-photon fidelity of at least 90.6%. The Berlin authors report no teleportation rate and no decoy-state analysis.

Fidelity figures also depend on the input source. In November I covered a Stuttgart-led team that teleported a state between photons from two separate quantum dots at about 72% post-selected fidelity. Its input photons came from an independent solid-state emitter and had to be frequency-converted to match. Berlin’s came from a laser locked to the same rubidium line as the source’s own 795 nm photon. I wouldn’t rank the two experiments on those numbers.

The 795 nm to O-Band Interface and Rubidium Devices

Rubidium’s D1 transition is at 795 nm. Rubidium is the working atom in many atomic-ensemble quantum memories, in atomic clocks and in several neutral-atom processors. Standard telecom fiber attenuates light near rubidium’s wavelengths by several decibels per kilometer, against about 0.34 dB per kilometer at 1,324 nm on the Berlin loop. A rubidium device that needs a metro link benefits strongly from a telecom-band interface.

The University of Science and Technology of China team behind the device-independent quantum key distribution result I covered earlier this month converted rubidium’s 780 nm photons to 1,315 nm with nonlinear frequency conversion, and Harvard and AWS converted photons from their diamond memories to 1,350 nm the same way. Qunnect teleports the state instead, onto a photon that its source creates in the O-band.

For teleportation to work as a wavelength bridge, the input photon has to be indistinguishable from the source’s 795 nm photon. The authors attribute the fidelity drop on the deployed loop to lower two-photon interference visibility, 68.6% against 72.8% locally, although the two values overlap within their stated uncertainties of about five percentage points. They trace the lower visibility to the higher pump power the team used to make up for 18 dB of end-to-end loss on the link, which yields more pairs at a lower single-photon quality.

A photon from an actual rubidium memory or atom would have its own spectrum and timing, and matching it to the source’s photon is a separate engineering problem from the one Borowska’s team solved. That is the Berlin experiment I would most like to see next: the same interface with a rubidium device, not a laser, supplying the input.

What an Operator Teleportation Service Still Requires

Mudesir presented teleportation as the basis for networking quantum computers across locations and pooling their computing power. In February 2025, David Lucas’s group at Oxford teleported controlled-Z gates between two trapped-ion modules about two meters apart, with 86% gate fidelity, and ran Grover’s search algorithm across the pair. An on-demand service linking matter-qubit processors across a city would need more than the interface Berlin demonstrated:

• Heralded entanglement between the two sites, available when the sender’s qubit is ready.
• A receiving qubit, in a memory or a processor, that stays coherent until the Bell-measurement result arrives and the correction is applied or tracked in software.
• Measured rates, latency and availability. The Berlin authors give no teleportation rate: two-fold coincidences between the source’s photons, not teleportation events, ran at about 300 per second on the deployed loop, and each tomography setting took three to ten minutes.
• A way to join elementary links. Entanglement swapping is one building block, and Qunnect’s New York preprint with Cisco this week shows it over deployed fiber at a measured rate above 0.65 swapped pairs per second. A working repeater also needs memories and end-to-end performance that swapping alone doesn’t establish.

Higher Bell-measurement efficiency would raise throughput. A standard passive linear-optical analyzer without extra photons succeeds at most half the time, typically by distinguishing two of the four Bell states, and Berlin’s setup accepted one, which makes the protocol probabilistic rather than unusable. The detectors are a cost question for operators: both Berlin stations used superconducting nanowire detectors, which need cryogenic cooling, while Qunnect’s New York design keeps those at the hub and runs its spoke nodes on room-temperature avalanche detectors.

Photonic and TELUS announced the receiving-qubit step six days before Deutsche Telekom’s release: a teleported state received into a matter qubit that can store it, after 30 km of TELUS commercial fiber. Their announcement included no paper, fidelity or rate, so a numerical comparison with Berlin has to wait for their data. Of the two, Berlin claims less and has published its numbers.

The business model operators are working toward is entanglement as a service: the operator distributes entangled pairs, and customer devices consume them for key distribution, sensing or computation. Deutsche Telekom has now shown it can distribute pairs with measured downtime and hand a rubidium-wavelength qubit into its fiber. The customer devices that would consume teleported states over a metro link are still in research labs.

Deutsche Telekom has not put a date on a teleportation service, and its announced next step is a multi-node test on the same network. For anyone planning around quantum networking, the dated events this spring are demonstrations at Mobile World Congress, not products.

The Barcelona showcase opens on 2 March. I’ll read the multi-node follow-up for two details: what produced the input photon, and what received the output state. If a rubidium device supplies the first and a memory stores the second, Berlin will have moved from a transport test to teleportation between quantum devices on an operator’s network, with the data published.

Update, 26 September 2026: On 22 September the authors posted a revised version of the preprint under a new title, “Quantum Teleportation of Polarization States on Live Metropolitan Fiber.” It keeps the local Bell-state measurement with a weak coherent input and reports an average fidelity of 90.4 ± 2.5% with a live 100 Gbit/s dense wavelength-division multiplexing channel, at a fixed output power of 0.27 dBm, sharing the fiber. The analysis above refers to the version posted on 18 February 2026.