Three Atoms, One State: The Promise and the Price of Networked Qubits

I went looking for one paper to begin this notebook, and the one that caught me came fresh out of Christopher Monroe's group at Duke.¹ The name matters. Monroe co-founded IonQ, and three decades of his trapped-ion physics are the bedrock the company is built on.² The claim that pulled me in was both narrow and large: the first fully distributed three-node GHZ state shared across single atomic qubits, joined together by light, with a clean violation of Mermin's inequality to prove it was genuinely quantum.¹ This is not a near-term revenue story. It is something rarer and, for the long arc, more important: a clear look at how separate quantum machines might one day be wired into one. So it is worth understanding what was done, what it cost, and where the cost goes from here. It is not a valuation and not financial advice. Do your own research.

The claim, and why three is hard

A GHZ state, named for Greenberger, Horne and Zeilinger, is one of the cleanest ways quantum mechanics shows its strange, non-local side.³ Where two entangled qubits are a handshake, a GHZ state is a whole room agreeing at once. The three atomic qubits here land in the state

$$ \lvert \Psi_{\mathrm{GHZ}}^{\pm} \rangle = \frac{\lvert \downarrow\downarrow\downarrow \rangle \pm e^{i\Phi}\,\lvert \uparrow\uparrow\uparrow \rangle}{\sqrt{2}} . $$

That notation is less forbidding than it looks. The bracket $\lvert \cdots \rangle$ is just physicists' shorthand for "the state in which." Each arrow is one atom, set either down or up, the quantum stand-ins for 0 and 1. So $\lvert \downarrow\downarrow\downarrow \rangle$ means "all three atoms down" and $\lvert \uparrow\uparrow\uparrow \rangle$ means "all three up." The state above is both at once, all-down and all-up together, with nothing in between. The symbol $\Phi$ is a shared phase, one quantum bookkeeping angle that ties the two halves together. Keeping that phase coherent is one of the delicate things the experiment has to get right.

The difference between three sparks and three lightbulbs you can keep lit.

Here is what nagged at me. Three-photon GHZ entanglement was already observed back in 1999.⁴ So why, in 2026, is doing it across three atoms still news? The answer is the whole point. Photons are wonderful messengers and hopeless filing cabinets: they arrive, you measure them, they are gone. A single atom is the opposite. It can be held, controlled, read out, and reused as a memory. Producing a certified GHZ state across three atoms, sitting in separate traps about two meters apart and linked only by optical fiber, is far harder than producing one across three photons in flight. It is the difference between three sparks and three lightbulbs you can keep lit.

How you entangle atoms that never meet

The apparatus is a small work of art, and it matters to anyone thinking about scale. Start with one isolated barium ion. A magnetic field splits its lowest energy level into two, an effect named for Pieter Zeeman, and those two levels become the qubit, written $\lvert\downarrow\rangle$ and $\lvert\uparrow\rangle$. Build three of these, each in its own trap, each linked by a three-meter optical fiber to a central box the authors call the GHZ-state generator.¹

A fast laser pulse lifts each ion to a higher energy level. When it falls back, it emits a single particle of light, a photon, and here is the trick: that photon comes out entangled with its own local ion,⁵

$$ \lvert \Psi \rangle_i = \frac{\lvert H \rangle_i \lvert \downarrow \rangle_i + e^{i\,\delta k_i x_i}\,\lvert V \rangle_i \lvert \uparrow \rangle_i}{\sqrt{2}} . $$

In words: the photon's polarization, the direction its light wave wiggles, horizontal ($H$) or vertical ($V$), is locked to the atom's state. A horizontal photon means the atom is down, a vertical photon means it is up, and because the atom is in both, so is the pair. (The exponential factor is simply a phase the photon picks up on its way out, a delicate angle the experiment must keep steady. Hold that thought, it returns in the error budget.) On its own this atom-photon link is excellent, with an average fidelity of 0.983, where fidelity is simply how close the real state is to the ideal one, 1.000 being perfect.¹

Then the three photons travel down their fibers and meet at the generator, where they are interfered and detected together. When one particular pattern of clicks fires, one photon caught in each of three detectors, the photons are measured away and the three distant atoms are left entangled with each other. The light is the matchmaker; once the match is made, it leaves. Nothing ever passes between the atoms except the photons they emitted, yet they end up sharing one state.

This is why atomic platforms have a natural appeal for this kind of networking: the atoms are the memory, the photons carry the entanglement between them. And networking is the quiet half of the scaling problem. A useful quantum computer will need more than good local qubits. It will need many separate modules joined into one larger machine.

Hand-drawn schematic: three trap nodes A, B and C, each holding one ion that emits a photon down an optical fiber to a central GHZ generator. When all three photons are detected together at the hub the photons vanish and a single red loop ties the three distant ions into one shared state. — Figure 1: How three atoms that never meet end up sharing one state. Each ion emits a photon entangled with it; when the three photons are jointly detected at the hub, the photons are measured away and the entanglement is left behind on the three distant atomic memories.

The price, part one: the error budget

Nothing is free. Conditioned on a successful heralding event, that is, the three photons all being detected, the authors do not cherry-pick the atomic measurement outcomes that follow. On those heralded events, the fidelity of the distributed GHZ state is bounded between 0.841 and 0.881.¹ That discipline matters: post-selection, quietly keeping only the runs that happen to come out well, can flatter a result, and these numbers are not flattered that way. The authors bracket the fidelity with two ingredients,

$$ \frac{\mathcal{P}+\mathcal{C}}{2} - \mathcal{Q} \;\le\; \mathcal{F} \;\le\; \frac{\mathcal{P}+\mathcal{C}}{2} + \mathcal{Q}, $$

where $\mathcal{P}$ is how often all three atoms are genuinely found together as all-down or all-up (here 0.955), $\mathcal{C}$ is the parity contrast, a measure of how cleanly the shared phase $\Phi$ shows up when you rotate and read the atoms (0.77), and $\mathcal{Q}$ is a small correction for stray error populations (0.020).¹

Where does the missing fidelity go? Three culprits dominate, and they are worth naming plainly, because they are exactly the things a network makes harder.

Polarization mixing (a fidelity cost of about 0.037). The whole scheme depends on every node emitting photons in the same clean horizontal-or-vertical channel. If one node leaks a little of the wrong polarization, the joint three-photon measurement no longer projects the atoms onto the ideal GHZ state, and the entanglement is diluted.
Spatial mode mismatch (under 0.03). For the photons to interfere correctly they must be, as far as the detector can tell, identical: same shape, same timing, same color. If their spatial profiles differ, the interference is imperfect, and imperfect interference means imperfect entanglement.
Recoil error (about 0.03). When an atom emits a photon it recoils, like a tiny rifle firing a tiny bullet. That kick can betray which way the photon went, and any information about the photon's origin leaking into the atom's motion blurs the shared state.

Add a longer tail of smaller imperfections, state preparation and readout, phase tracking, wave-plate calibration, and the total error comes to about 0.12.¹ Each piece is small. That is precisely the lesson. A network turns every small imperfection into a synchronization problem, one more thing to hold calibrated, in lockstep, at every node at once. Two years after the same group's two-node barium experiment,⁵ adding the third node was not a tidy case of two becoming three. It was a fight to keep all of those small errors small at the same time.

And the result is genuinely quantum, not a clever classical imitation. The team measured Mermin's inequality,⁶ a score built from a fixed set of measurements,

$$ M = \bigl\lvert \langle YYX\rangle + \langle YXY\rangle + \langle XYY\rangle - \langle XXX\rangle \bigr\rvert \le 2. $$

Here $X$ and $Y$ are two particular measurement settings applied to each atom, and the angle brackets $\langle \cdots \rangle$ mean the averaged result over many runs. Any classical world, one where the atoms quietly carry secret pre-agreed answers, is capped at a score of 2; quantum mechanics can reach 4. The team measured 3.203, clearing the classical ceiling by 27 standard deviations.¹ Better still, because trapped ions can be read out with near-perfect efficiency, they closed the "detection loophole" for the first time in a fully distributed system. The loophole is the escape hatch left open when you only catch a fraction of your events: a classical model can mimic a quantum result by hiding in the ones you missed. Catch almost all of them, and the hatch closes. There is no fair-sampling excuse here; the violation is real.¹

The price, part two: the rate

Fidelity and rate are two different questions. Fidelity asks: once the state is made, how good is it? Rate asks: how often do you actually get one? The fidelity is the delicate price; the rate is the brutal one, and it is the number an investor should sit with.

The protocol needs a usable photon from all three nodes in the same attempt. Each node delivers one only about 1% of the time. Because all three are required together, the probabilities multiply, and the rate of finished entangled states scales as

$$ r_{\mathrm{ent}} = \tfrac{1}{4}\,R\,D\,p_A\,p_B\,p_C \;\propto\; p_i^{3}, $$

where $R$ is how fast the machine makes attempts, $D$ is the duty cycle (the fraction of time it is actually trying rather than pausing to cool the ions), and each $p$ is the chance that a given node yields a collectible photon on a given try.¹ With each $p$ near 1%, the triple product is about one in a million. So even though the underlying atom-photon physics worked beautifully, it took billions of attempts: the measured output was 0.095 entangled states per second, one good three-atom GHZ state roughly every ten seconds.¹

My takeaway: this paper does not show practical modular quantum computing. It shows that one required ingredient, certified remote entanglement between atomic memories, is physically real.

That cube, $p_i^3$, is the heart of the "how does this scale" question. At 1% collection, multiplication is the enemy. The authors point to two honest exits, neither free.¹ The first is to collect photons far more efficiently, for instance by surrounding each ion with a high-cooperativity optical cavity, a pair of finely tuned mirrors that funnel the atom's photon into the fiber instead of letting it scatter away, pushing each $p$ toward 100% where the cube stops biting. The second is to change the recipe: build the GHZ state from remotely generated Bell pairs, the simple two-particle form of entanglement, stitched together by local operations, a route whose rate scales as $p_i^2$ rather than $p_i^3$. Better, but still fighting the same arithmetic.

Hand-drawn infographic: three boxes labeled A, B and C, each about one percent, multiplied together equal about one in a million, because all three photons are needed in one attempt. Two ways out are labeled: better cavities to raise p, and a Bell-pair route that scales as p squared, not p cubed. — Figure 2: Why one percent hurts. Because a usable photon is needed from all three nodes at once, the per-node collection chances multiply, so a 1% chance at each node becomes about one success in a million attempts. Pushing collection toward 100%, or switching to a two-photon Bell-pair recipe, are the two roads out.

The investor's read

So I put on the investor's hat. Where does this fit in IonQ's story?

The open question is whether it can become fast, reliable, and cheap enough to scale.

It fits best as a scientific proof point for IonQ's networking and modular-scaling story, not as a validation of the commercial roadmap. The paper carries IonQ affiliations and flows from the company's founding scientific lineage,¹ and it shows something real: separated ion memories can share a certified GHZ state through light. That matters, because IonQ's bull case rests partly on becoming a full-stack platform that treats quantum networking, the work of connecting machines into something larger than any one machine, as one of its core pillars.²

But the honest bottom line is the one I keep returning to. This result shows that remote atomic qubits can be entangled and certified. It does not yet show that those remote states can be generated fast, reliably and cheaply enough to serve as practical interconnects for large-scale distributed computation. A GHZ state once every ten seconds is a triumph of physics and an interconnect that does not yet exist.

That gap is where the investment question lives, and it is why, for the next few years, IonQ's investor story will likely be carried by nearer-term levers than remote networking: system sales, the performance milestones it reports such as qubit count and gate fidelity, and its published roadmap. By the company's own account, that roadmap runs from the 100-qubit Tempo system it describes as in the market today to a next-generation 256-qubit machine, with far larger systems sketched beyond, and its 2026 revenue guidance is 260 to 270 million dollars.² I treat these as IonQ's claims about itself, useful for reading the company's intentions rather than as independently verified facts. Remote networking, for now, is closer to scientific infrastructure than a near-term revenue engine, though the company does report an early commercial step in that direction: the first sale of a quantum memory node into the Mid-Atlantic Regional Quantum Internet at the University of Maryland.²

Where I land

I will end the week wearing both hats at once, which is the whole point of this notebook. As a physicist, I find the result beautiful: three distant atoms, each alone in its own trap, woven into one shared state simply by catching the light they emit and measuring it away. As an investor, I want to keep the verdict honest and narrow. This is a scientific proof point for IonQ's networking and modular-scaling story, evidence that the physics under the vision is real, not a validation of the commercial roadmap and not a near-term revenue result. It shows that one necessary ingredient, a certified GHZ state shared between separated atomic memories, genuinely exists. Whether that ingredient can be made fast, reliable, and cheap enough to wire real machines together is the question the next few years will answer, and it is where the investment case will be won or lost. Important, but early.

Sources & notes

I. Goetting, A. Kalakuntla, M. Shalaev, H. B. Shi, A. Ferrari, S. Saha, G. Toh, S. Male, and C. Monroe, "Tripartite entanglement of remote atomic qubits," arXiv:2606.17173v1 (15 June 2026), Duke Quantum Center and IonQ, Inc. Source for all quantitative claims here: the bounded fidelity 0.841 to 0.881; the average ion-photon fidelity 0.983; the populations, parity contrast and error budget; the Mermin parameter 3.203 and the closing of the detection loophole; the per-node collection efficiencies near 1% and the 0.095 per second generation rate; and the $p^3$ versus $p^2$ scaling and cavity outlook.
IonQ, Investor Overview, May 2026 (investor presentation), used here only to convey the company's own claims and roadmap rather than as verified performance data: the 100-qubit Tempo system it describes as in market and the next-generation 256-qubit system (slides 5 and 7), its 2026 revenue guidance of $260 to $270 million (slide 4), its framing of quantum networking as a core platform pillar (slide 5), and its reported first commercial sale of a quantum memory node into the Mid-Atlantic Regional Quantum Internet at the University of Maryland (slide 8). Monroe's co-founding role and the trapped-ion lineage are drawn from the public corporate record; see also the companion company story, IonQ: The Quantum Company That Decided to Become an Industry.
D. M. Greenberger, M. A. Horne, and A. Zeilinger, "Going Beyond Bell's Theorem," in Bell's Theorem, Quantum Theory, and Conceptions of the Universe (ed. M. Kafatos), 69-72 (Kluwer Academic, 1989). DOI: 10.1007/978-94-017-0849-4_10.
D. Bouwmeester, J.-W. Pan, M. Daniell, H. Weinfurter, and A. Zeilinger, "Observation of Three-Photon Greenberger-Horne-Zeilinger Entanglement," Phys. Rev. Lett. 82, 1345 (1999). DOI: 10.1103/PhysRevLett.82.1345.
J. O'Reilly, G. Toh, I. Goetting, S. Saha, M. Shalaev, A. L. Carter, A. Risinger, A. Kalakuntla, T. Li, A. Verma, and C. Monroe, "Fast Photon-Mediated Entanglement of Continuously Cooled Trapped Ions for Quantum Networking," Phys. Rev. Lett. 133, 090802 (2024). DOI: 10.1103/PhysRevLett.133.090802. Preprint: arXiv:2404.16167. The same group's earlier two-node barium result, and the basis for the atom-photon entanglement step.
N. D. Mermin, "Extreme Quantum Entanglement in a Superposition of Macroscopically Distinct States," Phys. Rev. Lett. 65, 1838 (1990). DOI: 10.1103/PhysRevLett.65.1838.