Full-stack quantum networks: what they solve, and what still stands in the way.

What a quantum network actually carries

A classical network moves bits. A quantum network's main job is different: it distributes entanglement between qubits in different places. Once two nodes share an entangled pair, they can use it to teleport a quantum state, run a key distribution protocol, or apply a gate between qubits that sit in separate machines. Entanglement is the resource, and everything else is built on top of it.

That one difference rules out most of what makes the classical internet work. Quantum states cannot be copied, so there is no amplifier and no resending the same packet. A photon lost in a fibre is gone, along with whatever it carried. Standard telecom fibre loses about 0.2 dB per kilometre at 1550 nm. After 100 km roughly one photon in a hundred survives. After 500 km it is about one in ten billion. Pirandola and colleagues showed in 2017 that without repeaters there is a hard ceiling on how fast entanglement or secret key can cross a lossy channel, and that the ceiling falls exponentially with distance.

A long-distance network therefore needs quantum repeaters: nodes that hold entanglement in memory, wait until neighbouring links succeed, and then join them by entanglement swapping. This is where the networking problem starts, because a repeater chain is a distributed system whose resources decay while they wait.

What “full stack” means

Early experiments were built end to end for one purpose: one source, one fibre, one detector, one protocol. A full-stack network splits the job into layers, the way TCP/IP does, so that each layer can be built and improved without rewriting the others. An application should not need to know whether the link underneath uses trapped ions, colour centres in diamond or clouds of cold atoms.

• Physical layer. Generating entanglement between neighbouring nodes. Photon sources, quantum memories, frequency conversion, fibre or free-space channels, detectors, and the timing that tells each node whether an attempt worked.
• Link layer. Turning unreliable physical attempts into a service. A node asks for “an entangled pair with this neighbour, at this fidelity, within this time” and gets back either a pair or a clear failure. Dahlberg and colleagues proposed the first such protocol in 2019, and Pompili and colleagues ran it on hardware in 2022.
• Network layer. Extending entanglement beyond neighbours through swapping, which means deciding which links to build, in what order and over which route.
• Transport layer. Delivering qubits reliably from end to end, for example by teleportation, with whatever distillation or error correction the application needs.
• Application layer. Key distribution, blind computation, distributed algorithms and sensing.

The layer boundaries matter more than they look. A classical router can hold a packet for as long as it likes. In a quantum network every stored qubit is losing fidelity, so one layer's delay becomes the next layer's noise. Designing the interfaces is part of the physics problem.

The software side has started to catch up. In 2025 a team led from Delft published QNodeOS, an operating system for quantum network nodes. Applications written once in platform-independent code ran on nodes built from nitrogen-vacancy centres in diamond, and the same design was shown on a trapped-ion node. In parallel, the IRTF's Quantum Internet Research Group has written down architectural principles (RFC 9340) and application scenarios (RFC 9583). Both are early work, but they are the first signs of a network being designed as a network.

Why it matters

The honest answer is narrower than most headlines suggest. It is also more interesting.

Key distribution that does not rest on hard mathematics

Public-key cryptography depends on problems we believe are hard to solve. Post-quantum algorithms swap out the problems that Shor's algorithm breaks, and they are the right answer for almost every system today. Quantum key distribution offers something different: security that follows from physics, where an eavesdropper shows up as errors. Device-independent QKD, demonstrated in 2022 with trapped ions and with neutral atoms, goes further and does not even require trusting the devices. Those experiments ran at very low rates over short distances. Entanglement-based networks are what would let them scale.

QKD does not replace post-quantum cryptography. It still needs an authenticated classical channel. Most deployed QKD networks rely on trusted relay nodes that see the key. Several national security agencies currently advise PQC over QKD for general use. The case for QKD is strongest for long-lived secrets on a few links where the hardware cost is justified.

Using a remote quantum computer without revealing the computation

Blind quantum computing, proposed by Broadbent, Fitzsimons and Kashefi in 2009, lets a client with very modest quantum capability run a computation on a remote server while hiding the input, the algorithm and the output. Verifiable versions also let the client check that the server did the work honestly. In 2024 an Oxford team demonstrated verifiable blind computing with a trapped-ion server and a client connected by optical fibre. If quantum computers end up in the cloud, as classical ones did, this is how users with sensitive workloads would use them.

Scaling quantum computers by linking them

Building one ever-larger processor runs into limits on wiring, cooling and crosstalk. Linking smaller modules through entanglement is the alternative. In 2025 an Oxford team connected two trapped-ion modules about two metres apart with a photonic link, teleported a controlled-Z gate between them and ran Grover's search across both. The distance is small, but it is the mechanism a datacentre-scale machine would use. The first serious customer for quantum networking may well be quantum computing itself.

Measurement beyond what one site can do

Shared entanglement can synchronise clocks more precisely than classical methods allow, and could in principle let separate telescopes combine light as if they were one larger instrument. These are longer-term goals, but no classical network can offer them at any price.

What a quantum network will not do is make the internet faster or carry ordinary data. It sits beside the classical internet and depends on it.

Where things stand

• 2017 The Micius satellite distributed entanglement between ground stations 1,200 km apart.
• 2021 A three-node network of diamond spin qubits in Delft shared entanglement among all three nodes.
• 2021 A space-to-ground QKD network spanning 4,600 km went into operation, using trusted relay nodes.
• 2022 Teleportation between two Delft nodes that are not directly connected, passing through the third.
• 2023 Twin-field QKD over 1,002 km of fibre, without repeaters or trusted relays.
• 2024 Heralded entanglement between memory nodes over deployed city fibre, reported by three groups: silicon-vacancy centres over a 35 km loop in Boston, diamond nodes 10 km apart between Delft and The Hague, and atomic ensembles up to 12.5 km apart in Hefei.
• 2025 QNodeOS runs network applications on real nodes. Distributed quantum computing across an optical link.

The pattern is clear. Links work. Small networks work, in the lab and on city fibre. A software stack exists as a prototype. What does not exist yet is a network that delivers high-fidelity entanglement at useful rates over long distances to applications that pay for it.

The open problems

These are the problems research groups and standards bodies are working on now. Most come down to one tension: entanglement is slow to make and quick to decay.

1. Memories that outlast the network's response time

Every heralded attempt needs a classical signal to say whether it worked. Light in fibre travels about 200 km per millisecond, so a node waiting on a neighbour 100 km away spends at least a millisecond per round trip, and a multi-hop chain far longer. The memory has to hold entanglement through all of that while staying efficient to write and read. Each platform sits at a different point on this trade-off: some store well but couple to light poorly, others the reverse. Two main lines of attack are multiplexed memories, which run many attempts in parallel, and long-lived nuclear-spin registers sitting next to the communication qubit.

2. Repeaters that beat bare fibre at a useful rate

A repeater only helps if the chain delivers more than the fibre would on its own. Memory-assisted links have beaten the repeaterless limit in controlled experiments, notably at Harvard in 2020. A chain that does so over hundreds of kilometres, at rates an application can use, has not been shown. Repeater designs are usually grouped into generations by how they handle loss and errors: heralded entanglement with distillation, then error correction on each link, then fully encoded one-way schemes. Each generation needs better hardware than the one before.

3. Connecting different kinds of qubits

The qubits that store information best rarely emit light at telecom wavelengths, so most nodes need quantum frequency conversion, which adds loss and noise. Superconducting processors face a harder version of the problem. Their qubits work at microwave frequencies, and converting a single microwave photon to an optical one without destroying it, known as transduction, is still far below the efficiency a network needs. Until transduction improves, the most mature quantum computers cannot easily join an optical network.

4. Fidelity budgets and distillation

Each swap compounds the errors of the links it joins. Distillation turns several noisy pairs into fewer good ones, but it uses up pairs that took a long time to make. How much fidelity each application really needs, and where in the stack to spend effort to reach it, is still being worked out. Key distribution tolerates fairly noisy pairs. Distributed computing needs pairs good enough to sit inside an error-corrected algorithm.

5. Routing with resources that decay

Classical routing assumes a packet can wait without harm. A quantum router must decide which links to build, which pairs to swap and which to throw away, knowing that every stored pair is getting worse. This is a real-time optimization problem with partial information. The classical control plane that coordinates it, with its own latency and clock synchronisation, is part of the network's quantum performance. Simulators such as NetSquid and SeQUeNCe exist largely to explore these protocols before hardware can run them.

6. Sharing fibre with classical traffic

Operators will not lay dedicated fibre for every quantum link. Single photons sharing a fibre with bright classical channels pick up noise, mostly from Raman scattering, and that raises error rates. Wavelength planning, narrow filtering and tight timing windows are active engineering problems rather than solved ones.

7. Standards and interoperability

A network built by one lab on one hardware platform is an experiment. A network that different vendors can join needs agreed interfaces: what a link-layer request looks like, how fidelity is reported, how nodes authenticate each other. The IRTF documents, ETSI's QKD specifications and ITU-T's work on QKD networks are a start. Most of the hard questions are still open.

8. Security of the network itself

Physics-based security covers the quantum channel, not the devices or the software that controls them. Commercial QKD systems have been broken through detector side channels. Trusted relays are only as trustworthy as whoever runs them. The control plane needs authentication, which today means classical or post-quantum cryptography. A secure quantum network still needs conventional security engineering done well.

9. Applications worth the cost

The last open problem is economic. Entanglement distribution is expensive and will stay that way for a while. The first networks to be built will be the ones attached to a clear use: linking processors inside one facility, protecting a handful of very high-value links, or serving a scientific instrument. Working out which uses justify the cost, and which are better served by post-quantum cryptography and ordinary networks, matters as much as the physics.

What this means if you build systems

For most organisations the practical steps today are classical. Migrate to post-quantum cryptography, because it protects data now and does not depend on any network being built. If a use case might need quantum networking later, model it before buying anything. Link distances, loss budgets, memory lifetimes and rate requirements show very quickly whether an idea can work.

That kind of modelling is what our simulator is for. It runs entanglement distribution over fibre with loss, detector noise and memory decoherence, and models distributed computing across QPUs with dedicated communication qubits, all in the browser. It is a quick way to see why a 500 km link without repeaters does not close, or how much a single repeater buys you. The docs explain the models and where they stop being accurate.

Further reading

1. S. Wehner, D. Elkouss, R. Hanson. Quantum internet: a vision for the road ahead. Science, 2018.
2. S. Pirandola, R. Laurenza, C. Ottaviani, L. Banchi. Fundamental limits of repeaterless quantum communications. Nature Communications, 2017.
3. A. Dahlberg et al. A link layer protocol for quantum networks. ACM SIGCOMM, 2019.
4. M. Pompili et al. Experimental demonstration of entanglement delivery using a quantum network stack. npj Quantum Information, 2022.
5. C. Delle Donne et al. An operating system for executing applications on quantum network nodes. Nature, 2025.
6. W. Kozlowski et al. Architectural principles for a quantum internet. RFC 9340, IRTF, 2023.
7. C. Wang et al. Application scenarios for the quantum internet. RFC 9583, IRTF, 2024.
8. J. Yin et al. Satellite-based entanglement distribution over 1200 kilometers. Science, 2017.
9. M. Pompili et al. Realization of a multinode quantum network of remote solid-state qubits. Science, 2021.
10. Y.-A. Chen et al. An integrated space-to-ground quantum communication network over 4,600 kilometres. Nature, 2021.
11. S. L. N. Hermans et al. Qubit teleportation between non-neighbouring nodes in a quantum network. Nature, 2022.
12. Y. Liu et al. Experimental twin-field quantum key distribution over 1000 km fiber distance. Physical Review Letters, 2023.
13. C. M. Knaut et al. Entanglement of nanophotonic quantum memory nodes in a telecom network. Nature, 2024.
14. A. J. Stolk et al. Metropolitan-scale heralded entanglement of solid-state qubits. Science Advances, 2024.
15. J.-L. Liu et al. Creation of memory–memory entanglement in a metropolitan quantum network. Nature, 2024.
16. D. Main et al. Distributed quantum computing across an optical network link. Nature, 2025.
17. D. P. Nadlinger et al. Experimental quantum key distribution certified by Bell's theorem. Nature, 2022.
18. W. Zhang et al. A device-independent quantum key distribution system for distant users. Nature, 2022.
19. A. Broadbent, J. Fitzsimons, E. Kashefi. Universal blind quantum computation. IEEE FOCS, 2009.
20. P. Drmota et al. Verifiable blind quantum computing with trapped ions and single photons. Physical Review Letters, 2024.
21. M. K. Bhaskar et al. Experimental demonstration of memory-enhanced quantum communication. Nature, 2020.
22. S. Muralidharan et al. Optimal architectures for long distance quantum communication. Scientific Reports, 2016.
23. P. Kómár et al. A quantum network of clocks. Nature Physics, 2014.
24. D. Gottesman, T. Jennewein, S. Croke. Longer-baseline telescopes using quantum repeaters. Physical Review Letters, 2012.
25. T. Coopmans et al. NetSquid, a network simulator for quantum information using discrete events. Communications Physics, 2021.
26. X. Wu et al. SeQUeNCe: a customizable discrete-event simulator of quantum networks. Quantum Science and Technology, 2021.

← All posts