Superconducting vs. trapped-ion qubits: how do they differ?

Ask a copilot which quantum hardware is "winning" and you will get whichever architecture's press release it read last. The honest answer is structural: the two dominant classes of quantum machine — superconducting circuits and trapped ions — are built from opposite physics, and each one wins exactly the axes its physics gives it for free while paying on the axes its physics charges for. As of August 2026, no axis has been resolved, no conversion rate between the two scorecards exists, and neither class has demonstrated end-to-end advantage on a useful problem.

Status as of: August 2026.

What are the two machine classes?

A superconducting qubit is a manufactured circuit: a loop of superconducting metal lithographed onto a chip, oscillating at microwave frequencies, held near 15–25 millikelvin. Because it is printed, you can print many, wire them densely, and drive them with fast electronics. Because it is manufactured, no two are perfectly identical, and each one couples to every stray field in its environment — which is why coherence is short.

A trapped-ion qubit is a single atom (typically ytterbium or barium) suspended in an electromagnetic trap and driven by lasers or, more recently, by precision electronics. Atoms of the same isotope are identical by nature, superbly isolated, and hold their state for seconds or longer. But moving quantum information between atoms means physically shuttling ions or sharing motional modes — processes measured in microseconds to milliseconds, not nanoseconds.

That is the whole story in miniature: fabrication buys speed and density; nature buys precision and memory. Everything measurable below follows from it.

What does each class win, measured?

Read the scoreboard column by column and the pattern is exact: each class wins the axis its physics gives it and pays on the axis its physics charges.

Three honest annotations. First, speed: the ~1,000× gap in gate time compounds — a trapped-ion error-correction cycle with shuttling and recooling runs near 1 ms, so a computation of fixed depth that takes minutes on a superconducting chip can take hours-to-days on ions. Second, fidelity: the numbers above are each vendor's own, measured under each vendor's own protocol — there is no neutral body that runs both classes under one ruler, a problem we mapped in our vendor-metrics explainer. Third, the record that isn't a machine: IonQ reported 99.99% two-qubit fidelity in October 2025 — on R&D-lab prototypes using electronic qubit control, explicitly not a production system. It is a real physics result and an honest press release; it is not a spec you can rent today.

What does a shot cost on each?

The per-shot spread between the cheapest superconducting device and the trapped-ion device on the same public cloud is roughly 188× ($0.000425 vs $0.08, plus a flat $0.30 per task on every QPU). That spread is mostly the speed axis wearing a price tag: an ion machine executes orders of magnitude fewer circuits per hour, so each one costs more. The hourly reservation prices tell the same story compressed — IonQ Forte reserves at $7,000/hour against Rigetti's $4,100/hour, only 1.7× apart, because an hour is an hour; it is the shots inside the hour that differ. Per-shot price is not a quality ranking — a shot on a higher-fidelity machine can be worth more per answer, a calculus we worked through in what it costs to run a problem on a real quantum computer.

Where does neither class win?

On the axis that decides everything: useful work. Neither architecture has demonstrated end-to-end advantage over a strong classical baseline on a problem anyone needs solved — the measured count of such wins is zero for both, as we tallied in which problems have proven quantum advantage. The fidelity-vs-speed duel matters enormously for when one of them might get there, because error correction multiplies both axes: higher physical fidelity means fewer physical qubits per logical qubit (the ion argument), while faster gates mean logical clock speeds that don't starve the algorithm (the superconducting argument — and the best published logical-qubit demonstration to date, Google's Willow, is superconducting, as covered in our qubit explainer). Which argument wins at scale is precisely the open question that the 2029–2033 vendor roadmaps are racing to answer, and no roadmap date has come due yet.

Rosetta's own data: none in this class. Our sealed runs are simulated or from single QPUs; we have no comparative benchmark of architectures and claim none. This post maps public, vendor-published measurements only.

What we know / what we don't know

What we know (measured, sourced above): superconducting two-qubit gates run ~1,000× faster than trapped-ion gates (~60 ns vs ~70 µs); the best production-machine median two-qubit fidelity belongs to a trapped-ion system (~99.92%, Helios); trapped ions offer all-to-all connectivity while superconducting lattices are fixed at ~4 neighbors; ion coherence exceeds superconducting coherence by 4–6 orders of magnitude; a superconducting shot on a public cloud costs ~188× less than a trapped-ion shot.

What we don't know: whether either class scales to cheap logical qubits — the ion fidelity edge and the superconducting speed edge pull the error-correction overhead calculation in opposite directions, and no one has run the experiment at scale. Whether all-to-all connectivity survives scaling, since ion shuttling time grows with trap complexity. What the fidelity numbers would look like under a neutral protocol — every figure above is the vendor's own measurement under the vendor's own methodology, and no independent body measures both classes with the same ruler. Whether IonQ's 99.99% lab-prototype fidelity transfers to a production machine at scale. And the largest unknown of all: which class, if either, first delivers a measured end-to-end win on a useful problem. Nobody has. When someone does, the scoreboard above becomes history's footnote.

Rosetta Q publishes verdicts with reproducible raw data. This is educational content, not a product claim.