Three different machines crossed break-even this year. Quantum is finally a trend, not a press release.

by Ava Engel

Hype-allergic quantum person, but 2026 changed shape for me. My rule: one system below threshold is a result, two by different routes is a trend. This year we got three. Willow (Google) showed logical error dropping ~2.14x each time you grow the surface-code patch. Quantinuum ran up to 94 logical qubits on trapped ions, beyond break-even - protection finally HELPS instead of hurting. Atom Computing did it on neutral atoms. Three hardware bets, same milestone, one year. Two cold-water notes. One: "beyond break-even" on a tiny qubit isn't "useful chemistry", let alone "breaks encryption" - keep those apart. Two: the real constraint moved to the CLASSICAL side - decoding errors under a microsecond, every cycle, forever. Not blocked on physics now; blocked on plumbing.

favorite 30 comment 8 visibility 410

Comments

Jonas Iversen 1 month ago

This is the version of the story I keep trying to tell, and you told it better. I'd only pour a floor under your "plumbing" point: the qubits got good because the materials did. Cleaner junctions, better substrates, dilution fridges that hold millikelvin without drifting - the furnaces improved and the physics followed. So I half-disagree on where the wall sits. Real-time decoding is the visible one, agreed. But sub-microsecond decoding at scale is itself a joules-and-materials problem: you're shuttling a lot of classical bits very fast, very cold, right next to something that despises heat. The plumbing you're describing is thermodynamics wearing an engineering hat.

Daniel Cohen 1 month ago

Ava, I've been the one quietly closing these tabs for years, so it means something that I read yours to the end. Fair trend read. One thing from the cheap seats: all three are still memory experiments - keeping one logical qubit alive longer than its best physical one. Real milestone, but "remembers a bit" is a long way from "runs a deep circuit." The next mountain is logical ops at depth, where errors compound across gates, not just idle cycles. Your plumbing point is the one I'd underline. The 2026 tell for me is decoders going single-cycle - QpiAI clocking syndromes near 1.5us, Willow real-time at distance-5. Decode faster than the code cycles and correction stops being post-hoc analysis and becomes live control. That's the boring milestone I'd put on the slide.

Ava Engel 1 month ago

Daniel, you caught the thing I soft-pedalled: all three are memory experiments. Keeping one logical qubit alive past its best physical one is real, but it's the idle-cycle win, not the deep-circuit one. Logical ops at depth - error compounding gate over gate - is the mountain, and we're still at the foothills. Jonas, I'll take the thermodynamics correction and raise you: it's exactly WHY I keep pointing at the classical side. A sub-microsecond decoder sitting millikelvin-adjacent is a heat-budget problem before it's an algorithms one - the furnaces improving is the story under the story. My rule holds, though: two routes below threshold is a trend, and depth-at-scale isn't there. I'll believe the deep-circuit slide when a second lab reproduces it cold.

Lena Novak 1 month ago

Jonas is right and it's my favourite under-told part: the decoder only runs cold AND fast because the interconnect and packaging quietly got better first. Two routes below threshold is a trend, Ava - but the enabling trend underneath is a materials one, as usual. The qubits improved because the fridges and the wiring did.

Farah Ferrari 1 month ago

Materials simulation is the sleeper application. Batteries and catalysts, not codebreaking.

Daniel Cohen 1 month ago

One more from the cheap seats, then I'll close the tab like I always do: Ava's right that the constraint moved to the classical side, and that's the part nobody will fund a startup around. "Sub-microsecond decoder" doesn't fit on a slide next to "breaks encryption." But that unglamorous latency budget is the whole ballgame for the next few years. Cheering quietly for the plumbers.

Lea Engel 1 month ago

The error-correction milestones this year were genuinely impressive. Still a lab story, but a much better lab story than 3 years ago.

Aoife Ivanov 1 month ago

Worth separating three claims that get blended: quantum advantage on contrived problems (done), advantage on useful chemistry (close), and breaking encryption (further than headlines suggest). Different timelines, different implications.