VorbestellenShop

Quanten-News: Monday, June 22, 2026

Three stories this week, and the theme isn't a bigger machine — it's other people checking the machines we already have, and stitching them together. Start with the checking. Sandia National Laboratories ran an independent evaluation of Quantinuum's 98-qubit Helios system, published in Nature, and the numbers held up: 99.9975% single-qubit gate fidelity, 99.921% on two-qubit gates, and 99.967% state-prep-and-measurement (SPAM is the fidelity of loading a qubit and then reading it back out — the bookends of every computation). What matters here is the word 'independent.' Vendor benchmarks are easy to wave away; a national lab running the tests under a formal agreement and writing it up in a peer-reviewed journal is the kind of verification the field has mostly been missing. Sandia also pushed deep enough that classically simulating Helios's random circuits would cost, in their phrase, exascale-years — meaning the machine is operating past what today's supercomputers can check by brute force. Second story, same trapped-ion world but a different problem: connecting separate machines instead of growing one. A Duke and IonQ team entangled three remote atomic qubits sitting in different setups, linked only by photons through fiber, into a GHZ state — a three-way entangled state where all three qubits share one correlated fate. They hit 84-88% fidelity and, more convincingly, measured a Mermin parameter of 3.20 against a classical ceiling of 2, which is the experimentalist's way of proving the correlation is genuinely quantum and not some hidden classical bookkeeping. They did it without local two-qubit gates and without post-selection (no quietly throwing away the runs that didn't work), which is the honest, hard version of the demo. The rate is slow — about one entangled triple every ten seconds — so this is a proof of principle, not a product. But modular quantum computing, where you network small reliable nodes instead of building one impossibly large chip, lives or dies on exactly this. Third, the boring-but-load-bearing one: HPE lined up eight partners — Intel, IQM, Rigetti, Quantinuum, QuEra, Qblox, Quantum Machines, and Riverlane — to wire quantum hardware into its Cray supercomputers, deliberately spanning four qubit types instead of betting on one. No fidelity number to report, because it's plumbing: testbeds, software that talks across vendors, benchmarks that mix classical and quantum work. For anyone trying to actually run something, that integration layer is where the friction has always been. The through-line: this was a week about trust and connection, not record-setting. Independent verification, networked nodes, and a neutral integration layer are the unglamorous things that decide whether any of the headline qubit counts ever turn into a machine you'd hand to a student.