On October 29, Finland's IQM Quantum Computers announced that its cloud-accessible superconducting quantum processor, Emerald, achieved simultaneous entanglement of all 54 qubits, with a GHZ-state fidelity of 0.516 — above the 0.5 threshold used to certify genuine multipartite entanglement. Previously, users of the same machine could reliably call on only about 30 qubits within a single algorithm. The remaining 20+ qubits were not missing; they were simply too risky to use together.
To understand why this matters, it helps to see which rule the result is fighting against.
The taller the stack, the faster it collapses
A quantum computer's greatest enemy is not a specific hardware flaw but a universal curve: more qubits and deeper circuits mean higher error rates. The reason is straightforward — every additional two-qubit gate adds another chance for error, and idle qubits lose their quantum state to environmental noise while waiting. So papers on large quantum chips follow an unwritten habit: results are typically shown on 10–20 qubits even when the hardware has dozens. Using everything makes the numbers look bad.
What Emerald did was push the "use everything" result above the threshold. A GHZ state pushes N qubits into a superposition of "all 0" and "all 1"; no classical system can produce such correlations. Fidelity above 0.5 means the measurement outcomes cannot be explained by classical correlations — the entanglement is real. IQM reported two sets of numbers: 40 qubits at 0.515 without readout error mitigation (REM), and 54 qubits at 0.516 with REM. Both crossed the line, and the company proactively disclosed whether mitigation was used — candid by peer standards.
> Tip: The GHZ state is named after physicists Greenberger, Horne, and Zeilinger and is the standard benchmark for testing multipartite entanglement. Zeilinger won the 2022 Nobel Prize in Physics for related experiments. Note that fidelity above the threshold proves "entanglement exists," not "quantum advantage": GHZ states are structurally simple and can be efficiently simulated classically — they serve only as a witness of hardware quality.
Won by scheduling, not by stacking hardware
No new physics appears in this improvement. IQM's approach had three layers:
1. Automated state-preparation routine — when selecting two-qubit gates, it jointly weighs gate fidelity and qubit idle time, dynamically choosing qubits and gate orderings so the longest-waiting qubits on the chain wait less. 2. Year-long calibration methodology refresh — two-qubit gate error rates cut by up to 50%. 3. Dynamical decoupling — protective pulse sequences applied to idle qubits to suppress decoherence.
All three fall under "software-defined hardware." The significance of this path is that it is reproducible: calibration and scheduling are general-purpose processes that can transfer to the next, larger chip without changing the chip design. IQM CTO Craig Ciesla put it this way: "Entangling an entire chip means every qubit and every connection between them must work correctly in the same circuit. Reaching this step comes from making state preparation itself smarter — choosing the gate sequence well and minimizing idle time."
Where this chip sits in the industry
Placing IQM back into October's quantum news pile:
| Approach | Representative players | Recent developments | |---|---|---| | Superconducting | IBM, IQM, Google | IQM Emerald full 54-qubit entanglement; IBM's five-year, $10B+ investment | | Molecular spin qubits | NVision | Carbene molecular qubits published in *Nature*, photonic interconnects with HyperLight | | Neutral atoms | Liangzhi Kaiwu and others | Agent-monitored experimental atomic quantum computing platform released |
In the same month, IBM announced it would invest over $10 billion in quantum computing over five years, targeting a fault-tolerant machine, Quantum Starling, with ~200 logical qubits by 2029. Germany's NVision published mass-producible molecular spin qubits in *Nature*. China's Liangzhi Kaiwu released an atomic quantum computing experimental platform monitored around the clock by AI agents. The路线s differ, but everyone is pushing toward "engineering usability."
IQM's position is distinctive: it is a European quantum company listed on NASDAQ (ticker IQMX, listed July 2026), selling full systems and cloud access. Customers want machines that can run the whole chip. Opening all 54 qubits to cloud users has direct commercial implications: previously customers bought 54-qubit machines but designed around a 30-qubit safety margin; now they can design algorithms for full-capacity compute.
The necessary caveats
First: GHZ fidelity figures cannot be compared hard across platforms. Numbers in the literature are scattered across different eras, qubit counts, and error-mitigation techniques. IBM Montreal's 27-qubit 0.546 (2021) and IQM's 54-qubit 0.516 are not the same exam — doubling the qubit count itself punishes fidelity.
Second: This is not quantum advantage. As Tech Times' technical commentary put it clearly: the GHZ state's density matrix is structurally simple, and classical algorithms can reproduce its statistics without exponential overhead. Its value is as a hardware witness, not a proof of computing power.
Third: The "up to 50%" reduction in two-qubit gate error rates is an upper-bound statement; per-qubit specifics were not disclosed individually.
What is genuinely worth watching next: the first algorithmic results cloud users produce with the full 54 qubits, and when IQM's promised "AI-agent automated calibration" graduates from internal process to public capability. When those two land, this news moves from "demonstration" to "production."
Sources and limitations: IQM official press release (October 29, Espoo), Businesswire syndication, The Quantum Insider and Tech Times coverage of October 29–30. The 0.516/0.515/0.5-threshold figures, 50% error-rate reduction, and ~30-qubit former limit are IQM self-reported; that GHZ fidelity does not prove quantum advantage is a baseline property judgment; IBM's $10B figure is via Hong Kong media etnet's relay; cross-route comparisons are drawn from public reporting and were not independently audited.