Analog quantum simulator vs universal quantum computer: what is the difference?

TL;DR (September 27, 2026): An analog quantum simulator dials a Hamiltonian into hardware and lets physics evolve continuously — no gates, no error correction, one family of problems per machine. A universal gate-based quantum computer applies discrete gates from a universal set and can, in principle, run any quantum algorithm — but only behind fault tolerance it does not yet have at scale. The largest public quantum arrays today are analog — 256 atoms on a public cloud since November 2022 (arXiv:2306.11727), 3,000 atoms in continuous operation for over 2 hours in the lab (Nature, 16-Sep-2025) — precisely because skipping gates is cheaper. The price: an analog machine will never run Shor, and analog advantage claims keep drawing classical re-runs — the latest landed September 3, 2026 (arXiv:2609.01719). Rosetta Quantum files both paradigms under one rule: a result counts when an independent party can re-run it.

What is an analog quantum simulator?

A machine that is built to be the problem, not to compute it. The idea is Feynman's (1982): to study a quantum system, build another quantum system whose native physics follows the same equations, tune it, and measure. There is no instruction stream. The program is a set of dials: where the atoms sit, how strong the driving field is, how the detuning ramps over microseconds. On QuEra's Aquila — 256 neutral atoms, the largest analog QPU on a public cloud, on AWS Braket since November 2022 (AWS) — the user chooses atom positions and waveforms within a published envelope: Rabi drive 0–2.5 × 2π MHz, detuning ±20 × 2π MHz (QuEra spec sheet, accessed 27-Sep-2026). That is the whole programming model — a Hamiltonian, not a circuit (arXiv:2306.11727).

What can that buy? Real, peer-reviewed physics: in 2021 a 256-atom array of the same family mapped out quantum phases of matter (Ebadi et al., Nature 595, 227–232, 2021). What it cannot buy is generality: an analog machine answers questions about Hamiltonians it can natively express, and nothing else.

What does «universal» mean in a gate-based computer?

Universal means: any quantum algorithm, in principle. A gate-based machine applies discrete operations from a universal gate set, and a universal set can approximate any quantum evolution to any precision — that is a theorem, not a slogan. Everything Rosetta has covered on the algorithm side lives in this model and only in this model: Shor, Grover, quantum phase estimation (our QPE explainer). Quantum error correction is also defined here: you correct discrete gate errors, which is why every fault-tolerance roadmap is a gate-based roadmap.

The symmetric price: generality costs overhead. Discrete gates accumulate discrete errors, so the useful-scale versions of those algorithms will require fault tolerance that no vendor has at scale — the measured state of that race is one logical qubit running a full QPE instance, against millions of physical qubits required for cryptography-scale runs (both sourced in our QPE post).

Where is the line in practice?

The two columns are two contracts. Analog: you buy the physics of one Hamiltonian family — big, cheap per atom, running today — and you accept no gates, no error correction, no Shor, and a verification story that is still research. Digital: you buy generality and a defined path to error correction, and you accept smaller machines and a longer wait. Quantum annealing — D-Wave's paradigm — is an analog special case aimed at optimization; Rosetta Quantum drew that map in our annealing vs gates explainer and it stands.

Every row is one of the two contracts. When a vendor says «qubits», the first question Rosetta asks is which column the machine sits in — dials or gates — because the number means a different thing in each.

What has each side actually measured?

The analog side's résumé is real measured physics. The 256-atom phase diagrams of 2021 were peer-reviewed and reproduced across groups (Nature 595, 227–232). Google's analogue–digital hybrid ran 69 superconducting qubits through a magnetism simulation (Nature, Feb-2025) — with a margin («more than a million years» on the Frontier supercomputer) that is the team's own estimate, stated on Google's blog, not an externally verified number. The 3,000-atom continuous array (Nature, 16-Sep-2025) is a hardware milestone — over 2 hours of sustained operation with atoms reloaded on the fly — not a computation.

The field's own flagship review goes further than we will: Daley, Bloch, Troyer, Zoller and co-authors wrote in 2022 that «a first practical quantum advantage already exists in the case of specialized applications of analogue devices» (Nature 607, 667–676, 2022). Rosetta Quantum applies a stricter bar: an advantaged solve requires an independent re-run against a matched classical baseline on the same instance. Under that bar, the public record holds zero verified analog advantaged solves — and zero digital ones. The two columns are tied at 0; they are just tied for different reasons.

Why do analog advantage claims keep drawing classical re-runs?

Because analog outputs are statistical estimates of physics, and classical physics methods keep getting better at estimating the same numbers. The live case is D-Wave's: the claim — «beyond-classical computation in quantum simulation», with the largest problems put at nearly a million years on Frontier — was published in Science 388, 199 (Mar-2025). Tensor-network re-runs answered within months (our tensor-networks explainer has that clock); D-Wave answered back on 26-May-2026 that the re-runs «do not reproduce the full scope» of the demonstration (D-Wave press release). Then, on September 3, 2026, a single-author paper from the Flatiron Institute reported classical time-dependent variational Monte Carlo reproducing the experiment's two-spin correlations at the QPU's own error level, below 5%, on all four lattice topologies — including the biclique instances the earlier re-runs had not attempted — at a cost of ≈2,100 GPU-hours against seconds on the QPU (Wiersema, arXiv:2609.01719). The author's own verdict is the honest one: the results «sharpen, rather than settle, the question». Both sides are linked above; neither disputes the measured physics — the dispute is entirely about the classical baseline, which is exactly where our annealing explainer said it would live.

There is theory pulling the other way, and it deserves the same visibility: Trivedi and co-authors argue that for a class of observables, analog simulators are provably stable to realistic noise — advantage could survive imperfection (Nature Communications, 2024). The question is open in both directions. That is what «contested» means on the clock above.

Which paradigm should a decision-maker bet on?

The one whose contract matches the problem — and the dates should be read in future tense. If the problem is quantum matter itself — magnets, phases, dynamics — analog machines are the near-term bet the field itself makes, and access is cheap to try (our pilot-evaluation gates apply unchanged). If the problem is optimization, cryptography or general algorithms, only the universal column applies, and its scoreboard today is the one Rosetta tracks: zero verified advantaged solves (the ledger's unit).

The vendors themselves are hedging toward convergence, in the future tense: Pasqal sells analog machines today and declares a digital ramp — 10,000 physical qubits targeted for 2028, «100 high-fidelity logical qubits by 2029», 200 by 2030 (Pasqal roadmap); those numbers will be measured when the machines exist, not before. QuEra operates Aquila in analog mode on Braket while its fault-tolerant gate-based demonstrations live in the lab (our neutral-atoms map carries the measured numbers). A buyer's question for any «dual-mode» pitch: which contract is shipping today, and which one is a date on a slide.

What we don't know

We don't know whether any analog advantage claim will survive the re-run cycle — the September 2026 re-run itself says it sharpens rather than settles the question, and its author flags his ≈2,100 GPU-hours as a property of today's ansatz, not a lower bound. We don't know whether the noise-stability theory for analog simulators extends to the observables buyers would pay for. We don't know whether the dual-mode vendors' declared logical-qubit dates (2027–2030) will hold — they are roadmaps, and they will be measured on arrival. And a scope note: Rosetta Quantum has not yet run an analog job on our own bench — our in-house experiments so far are gate-model and classical-baseline work. When we run one, the run report will be published here, win or lose.

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