Marin Ivezic
In June 2017, a group led by Jian-Wei Pan used the Micius satellite to distribute entangled photon pairs to two ground stations 1,203 km apart. The result was a genuine first. It was also, in engineering terms, a trickle: the overwhelming majority of what the satellite sent never arrived, and the pairs that did arrive came in at a rate no network operator would recognise as a link.
Today, the same physics is being packaged and sold. Vendors and carriers now use the phrase Entanglement-as-a-Service, or EaaS, to describe an offer where a provider distributes entangled qubits between two or more of your sites on request. Architects are starting to see it in roadmap decks and, occasionally, in procurement documents.
The phrase borrows its grammar from cloud computing, and that borrowing does most of the damage. Compute, storage, and bandwidth are all things a provider can produce in bulk, hold, copy, and resell. Entanglement is none of those. Before anyone can evaluate an EaaS proposal, they need a working picture of what the resource is and what handling it costs.
What the service delivers
Entanglement is a correlation between two quantum systems that is stronger than any correlation classical physics allows. Two photons prepared together can be described only as a joint state: measuring one fixes the outcome you will see on the other, in a way that cannot be reproduced by any shared instruction list agreed in advance. The standard unit is the Bell pair, one entangled pair of qubits, one half held at each end.
A Bell pair is not a message. Nothing is transmitted by measuring it, and it carries no data of its own. What it gives you is a shared resource that two parties can spend. Spend it one way and you get a secret key that no third party can hold a copy of. Spend it another way and you can teleport a qubit’s state from one node to the other, consuming the pair and two classical bits in the process. Spend enough of them together and two separate quantum processors can run operations on each other’s qubits.
So the product an EaaS provider sells is a delivery rate of Bell pairs of a stated quality between named endpoints. Everything else in the offer is scaffolding around that one line item.
Three properties that break the bandwidth analogy
Three facts about quantum states govern every design decision in these networks, and they are the reason the service looks so unlike a leased line.
Quantum states cannot be copied. The no-cloning theorem says there is no operation that takes an unknown quantum state and produces two identical copies of it. A classical network is built on the assumption that a signal can be read, buffered, duplicated, and forwarded. Strip that assumption out and most of the equipment in a carrier’s optical path becomes unusable for the quantum channel. No optical amplifiers. No optical-electrical-optical regeneration. No monitoring tap that reads the traffic in passing, because reading it destroys it.
Loss cannot be compensated by boosting power. In a classical fibre run, attenuation is answered with amplification every 80 km or so. A single photon cannot be amplified. It either arrives or it does not, and the probability of arrival falls exponentially with distance. Over a few hundred kilometres of fibre, direct transmission stops being slow and starts being effectively zero.
Entanglement decays. Interaction with the environment degrades the correlation between the two halves of a pair, a process called decoherence. Quality is measured as fidelity, a number between 0 and 1 describing how close the delivered state is to a perfect Bell pair. In many setups a Bell-state fidelity above roughly 0.5 indicates the pair is still entangled, and the threshold a given application needs is usually well above that. A pair that sits idle in a poor memory is a pair that expires.
Put together: the resource cannot be duplicated, cannot be regenerated the way a classical signal can, and has a shelf life. That is why nobody can build a warehouse of entanglement and ship it out. Pairs are made close to when they are used.
What a deployment contains
Given those constraints, the workaround is to never send a photon very far. The technique is entanglement swapping. Suppose node A shares a pair with node B, and node B separately shares a pair with node C. A joint measurement performed at B on its two halves leaves A and C entangled with each other, and consumes the two original pairs. A and C never exchanged a photon.
Chain that operation and you get a quantum repeater: a station that holds one half of a link in a quantum memory, a physical system able to store a qubit’s state coherently until its partner link is ready, then performs the swap. Memory is what turns swapping from a lucky coincidence into a scheduled operation. Both links have to exist at the same moment, and without storage, the odds of that happening on a long chain collapse.
Around the quantum path sits a second network that gets less attention and causes more integration work. Every entanglement distribution needs classical messaging alongside it: heralding signals announcing that a photon arrived, measurement bases, timing synchronisation, swap outcomes, and the ordinary business of provisioning and monitoring. In practice a deployment carries three distinct channel types. One quantum channel for the photons, one low-latency classical channel tightly coupled to the quantum operations, and one ordinary control channel for management traffic.
The quantum channel can run on standard telecom fibre, and it can share a strand with live classical traffic through wavelength division multiplexing, where different wavelengths run independently down the same glass. One common approach places quantum signals in the O-band around 1310 nm while classical data occupies the C-band around 1550 nm, which keeps scattered light from high-power classical channels out of the single-photon detectors. Plenty of systems run the quantum channel in the C-band instead, with heavy filtering and tight control of the classical launch power. It works, and it is fragile. Raman scattering from a strong neighbour will swamp a detector across a surprisingly wide band, so most current deployments still fall back on dark fibre for the quantum path.
Apologies in advance, but this is the unglamorous part that determines whether a pilot happens: the fibre plant matters more than the physics. Splices, connectors, patch panels, and the number of passive components between endpoints all show up directly in the delivered pair rate.
What has been demonstrated, and when
Four results mark out the current boundary. Read them as capability markers rather than as products.
Satellite distribution, 2017. The Micius experiment shared entangled pairs across 1,203 km of free space. It proved that entanglement survives an atmospheric path and that global coverage does not depend on fibre. It also showed the cost: link losses in free space are severe, and the delivered rate was orders of magnitude below anything an application would want.
Metropolitan fibre, 2020. Argonne National Laboratory and its partners distributed entanglement around an 84 km fibre loop in the Chicago suburbs, on infrastructure outside a laboratory building. This established that deployed fibre with real temperature swings and real vibration can carry the quantum channel.
Multi-node swapping, 2021 and 2024. Ronald Hanson’s group at Delft built a three-node network in which two end nodes became entangled through swapping at a middle node, with no photon passing between the ends. The 2021 version lived inside a single laboratory. In 2024 the group reported entanglement between processor nodes in Delft and The Hague across deployed fibre, roughly 25 km apart. That step, from bench to municipal fibre, is the one that matters for network design.
Memory-based repeater nodes, 2024. A Harvard team working with Amazon Web Services ran two nodes based on silicon-vacancy centres in diamond over a 35 km loop of deployed fibre under Boston. Nuclear-spin memory in those nodes held entanglement at the second scale. Light circles the earth roughly seven times in a second, so memory lifetime has stopped being the binding constraint on distance. Rate and fidelity are now the binding constraints, and both need orders of magnitude of improvement before a service level agreement is writable.
Two things follow. The physics of an end-to-end entanglement network has been demonstrated in the field, node by node. Nobody has yet demonstrated the throughput, the node count, or the uptime that a commercial offer implies.
Where the money is today
Most of what is currently sold under a quantum networking heading is quantum key distribution, or QKD. Two parties measure correlated quantum states and derive a shared secret key from the results; the physics guarantees that an eavesdropper cannot copy the transmission without disturbing it and revealing their presence. Artur Ekert’s 1991 protocol, E91, does this with entangled pairs directly.
QKD is a genuine product with a genuine caveat. Fibre reach without repeaters is limited, so long links are built from segments joined by trusted nodes, intermediate sites that receive a key on one segment, decrypt it, and re-encrypt it onto the next. The operator of that site can read the key. The security argument for the link as a whole then rests on physical and organisational control of every intermediate site, not on quantum mechanics. This is precisely the dependency that working repeaters remove, and it is the most useful question to ask about any deployed network described as quantum-secured.
A large share of what carries an EaaS label today is trusted-node key distribution with newer wording on the datasheet. That is not a fraud, and the underlying links do useful work. It is a different architecture from on-demand entanglement between arbitrary endpoints, and the gap between them is roughly a decade of hardware development.
Meanwhile the standards work has started in earnest. The Internet Engineering Task Force published RFC 9340, Architectural Principles for a Quantum Internet, which names entanglement distribution as the core service every application layer builds on and sketches the layering that quantum networks will need. The US Department of Energy set out a blueprint for a national quantum network in 2020, and the European Quantum Internet Alliance is building a multi-node demonstrator under the Quantum Flagship. Standards bodies at the European Telecommunications Standards Institute (ETSI) and the International Telecommunication Union (ITU) are working the interoperability problem. None of this is a product either, and all of it is where the vocabulary of the eventual products is being fixed.
Six questions for an EaaS proposal
If a proposal reaches your desk in the next few years, these six questions separate an engineering offer from a positioning statement.
- What is the delivered Bell pair rate, at what fidelity, measured at my endpoints? A rate quoted without a fidelity is meaningless, and a fidelity quoted at the source rather than the endpoint tells you nothing about the link. Ask for both, measured at the demarcation point, over a stated observation window.
- Are there trusted nodes in the path, and who operates them? If the answer is yes, the security of the service depends on the physical security of those sites. That is an auditable control, and it belongs in the contract rather than in the physics section of the deck.
- What happens to fidelity as it degrades, and at what cost? Providers may use entanglement distillation, consuming several low-fidelity pairs to produce one better pair. The technique works and the exchange rate is brutal. Find out how many raw pairs are spent per delivered pair at your target fidelity.
- What does the quantum path do to my fibre plant? Dark fibre or shared strand. Which band. What passive components are permitted between endpoints. Whether existing amplifiers must be bypassed, and what that does to the classical services already running on that route.
- How is the link monitored? You cannot ping a quantum channel without destroying the state you are testing. Health is inferred statistically from heralding rates and from sacrificial pairs spent on fidelity estimation. Ask what fraction of capacity monitoring consumes and what the alarm thresholds are.
- What application is this for, and would post-quantum cryptography do the job? For most confidentiality requirements it will, at a fraction of the cost and with no new fibre constraints. The cases where entanglement is the right answer are specific: distributed quantum computing across sites, entanglement-enhanced sensing and timing, and a small set of security requirements where key material must demonstrably never exist in a copyable form. If the proposal cannot name one of those, the answer is a migration project rather than a network project.
What to learn first
The teams that will evaluate these offers well are not quantum physicists. They are network architects who have added enough quantum information to read a datasheet honestly. Concretely, that means being able to reason about the loss budget of a quantum path, to say what a fidelity number implies for a given protocol, to distinguish a repeater from a trusted node in a topology diagram, and to design the classical control plane that any entanglement service depends on.
We built the Certified Quantum Network Architect (CQNA) program for exactly that reader: practising network professionals who need working competence in quantum network architecture rather than a research background. It covers entanglement distribution, repeater and memory technologies, the hybrid quantum-classical control plane, and the evaluation discipline sketched above.
Two useful companions. For the cryptographic migration work that sits alongside this, and that most organisations should be doing first, the methodology at pqcframework.org is the practical starting point. For deeper technical treatment of the underlying research, PostQuantum.com covers the entanglement service literature in more detail.
Entanglement is real, deliverable, and demonstrably survivable over deployed municipal fibre. It is also slow, fragile, and expensive, and it will stay that way for a while yet.