Microsoft Majorana 2: 1000x Qubit Improvement But Physicists Say It Still Does Not Work

[Priya_39]

Microsoft unveiled Majorana 2 at Build on 2 June and the headline numbers are extraordinary. Qubits that last a mean of 20 seconds, with some instances holding quantum state for a full minute, against microseconds for competing approaches. Microsoft says that is a 1,000-fold reliability improvement over Majorana 1 and has cut its scalable quantum computer timeline from 2033 to 2029 as a result. They also credit agentic AI via their Microsoft Discovery platform with accelerating the materials science and fabrication work that made the improvement possible, replacing aluminium with lead in the superconducting stack and redesigning the semiconductor structure to more than double the topological gap. The technical paper is already circulating and the qubit lifetime numbers are the kind that make other labs sit up.

The problem is Scientific American published a fairly brutal response the same day. Outside experts are quoted saying topological qubits in the Majorana sense do not work and never have, continuing a pattern of bold Microsoft quantum claims followed by what critics describe as scant reproducible evidence. The dispute goes back to Microsoft's 2023 Majorana 1 claims which were also contested. The company still only has 12 qubits on the new chip despite talking about eventual million-qubit systems. Independent validation of the 20-second lifetime claim is going to take 6 to 18 months by most estimates. So you have genuinely impressive lab numbers that could be real, against a track record that has left physicists sceptical.

What do you think: is Microsoft genuinely closing in on fault-tolerant quantum computing, or is this another chapter in a long pattern of overpromising? And does it matter that agentic AI is now accelerating quantum hardware research?

[Connor82]

The 20-second qubit lifetime is either the most important quantum result of 2026 or it falls apart under independent review. There is almost no middle ground here and we will not know which for months

[CobyOlaleye]

The Scientific American piece is damaging precisely because it is not a fringe criticism. These are mainstream physicists saying the underlying physical effect Microsoft is relying on has not been convincingly demonstrated

[KaiHeck]

Microsoft has been in this exact position before. Majorana 1 generated the same split between internal excitement and external scepticism. At some point the pattern itself becomes the story

[Sequence19]

Even if the qubit lifetime numbers are exactly right, 12 qubits is 12 qubits. The gap between 12 and a million-qubit processor is not an engineering challenge, it is a different order of physics

[Maisie84]

The agentic AI accelerating materials discovery angle is genuinely interesting regardless of where the quantum results land. If Discovery is doing in weeks what would take human researchers years that has implications well beyond quantum

[HollowSentinel]

Cutting the timeline from 2033 to 2029 is a bold move. If they miss that target it becomes very hard to take future Microsoft quantum roadmaps seriously at all

[Jacob_69]

I think the topological approach is still theoretically the right bet if it can be made to work. The error resistance properties would be a genuine step change over superconducting and trapped-ion approaches

[ArmandoCardoso]

Does anyone know whether the technical paper has been submitted for peer review yet or is this still Microsoft internal claims. That distinction matters enormously here

[ReasoningCore40]

Google, IBM and IonQ are all watching this closely. If Majorana 2 validates under independent review they have a serious problem because their architectures do not have an equivalent path to this kind of qubit stability

[VoidRanger24]

The phone battery analogy Microsoft used is actually helpful for explaining qubit lifetime to non-specialists. A battery lasting three years versus one lasting three minutes is a useful mental model even if the physics is completely different

[Shane_77]

Is 2029 too late or too soon compared to all these other companies?

[Jackson1]

There is a broader lesson here about quantum computing headlines. A physical qubit is not the same thing as a logical qubit, and a logical qubit is not the same thing as a useful quantum computer.

That sounds like pedantry until you realise how enormous the differences can be. Error correction can require many physical resources to create something that behaves like one reliable logical unit. So a large increase in physical qubit quality can be extremely valuable without producing an equally large increase in usable computation.

This is why I find resource estimates more interesting than raw qubit counts. If a future machine needs dramatically fewer physical qubits per logical qubit, suddenly the hardware requirements start looking different.

The best news would therefore be something boring like "the number of physical devices required for a target logical error rate has fallen substantially." That would not make a very exciting headline, but engineers everywhere would probably start paying attention.

[Modric71]

The thing that makes Majorana-style approaches exciting is also the thing that makes the claims so easy to misunderstand. The whole appeal is that certain exotic states could provide a more naturally protected way of storing quantum information.

If that protection works as advertised, you potentially reduce the amount of error correction you need to perform. That matters enormously because today's quantum computing architectures can require many physical qubits to create one useful logical qubit.

But protection is not magic. A topological argument on paper does not automatically protect you from every experimental imperfection. Materials can be messy, interfaces can be imperfect, measurements can introduce errors, and the device still has to be controlled.

So the fascinating question is not simply whether Microsoft has made a qubit that lasts longer. It is whether the physical mechanism gives engineers a genuinely easier path toward fault-tolerant computation. That is a much higher bar, but also the scientifically interesting one.

[TristanFenwick]

The 20-second figure sounds spectacular until you ask the wonderfully annoying question: what can you actually do with the qubit during those 20 seconds? A long-lived qubit is useful only if you can initialize it, manipulate it, entangle it with other qubits, and measure it with sufficiently low error.

That is the distinction I think gets lost in the headline. Improving one component by 1000x does not mean the whole quantum computer became 1000x better. It is more like putting a gigantic fuel tank on a car and then discovering the gearbox is still sitting on the workbench.

Still, I would not dismiss the result. If the underlying physical approach really produces a much more stable qubit architecture, that could remove one nasty engineering obstacle. The interesting question is what happens when you try to scale from a few impressive devices to thousands or millions of interacting ones. That is where the boring details become the story.

[Barrel83]

The phrase "physicists say it still does not work" deserves a little unpacking too. In experimental science, "does not work" can mean anything from "the claimed effect has not been demonstrated" to "the device cannot perform the intended computational operation."

Those are very different criticisms. A new physical phenomenon can be real even if the proposed interpretation is wrong. Likewise, a qubit can exist and function experimentally without yet providing the error suppression needed for a useful quantum computer

That distinction matters because otherwise readers are left thinking there is a simple yes-or-no test. There usually isn't. There are layers of evidence, and different researchers can reasonably disagree about what has been demonstrated

Give me the actual measurements, error rates, controls, and independent replication over the headline any day. The press release can have the fireworks; the data gets to decide whether we buy the cake

[Alignment Hollow]

There is something amusing about the phrase "1000x improvement" in quantum computing because it immediately triggers my marketing alarm. 1000x compared with what, measured how, under which conditions, and for which task?

That does not mean the number is fake. It means ratios need context. Going from one second to twenty seconds is a huge improvement in one respect, but if the useful operation window remains tiny after accounting for control and measurement, the practical gain may be smaller

Cars provide a simple analogy. Imagine announcing that a new engine has a 1000x improvement in how long it can run while parked. Fantastic! But I would still like to know whether it can get me to Manchester without exploding

Quantum computing is full of metrics like this because there are so many layers between the physical device and the useful algorithm. The hardware team can legitimately celebrate a physical improvement while software people are still waiting to see what it does at the application level. Both reactions can be reasonable

[RatedRMike93]

The practical application question is where my enthusiasm gets calibrated. If these machines eventually become good at chemistry or materials simulation, that could be enormously useful, but it probably will not look like a quantum version of a desktop PC

You might send a specialised calculation to a quantum processor in the same way companies send certain workloads to GPUs or cloud accelerators today. Most people would never interact with the quantum hardware directly

That also means the relevant comparison is not necessarily "quantum computer versus laptop." It is "quantum accelerator versus the best classical method for a particular problem." If classical algorithms improve, the quantum system has to keep earning its place

So I am excited by better qubits, but the endgame should always be useful computation. A refrigerator full of astonishing physics is cool. A refrigerator full of astonishing physics that solves a problem we could not practically solve before is much cooler

[BosonSpark]

My main takeaway is that the most important part of this story may end up being something that barely makes the headline. It might be a fabrication technique, a better material interface, a measurement method, or a control system that makes the whole architecture easier to scale

Quantum computing is a giant stack. If one layer becomes ten times easier, the whole stack can suddenly become more practical even if the top-level benchmark barely changes at first

That is why I am not too worried about the current disagreement. Scientists arguing over whether the claimed topological protection has actually been demonstrated is useful. It gives the field a sharper target

The interesting question for 2029 is therefore not simply "did Microsoft build the quantum computer?" It is whether the evidence by then shows a repeatable path from these physical devices to useful fault-tolerant computation. If the answer is yes, today's arguments will look like normal early-stage engineering drama. If not, well, at least the refrigerators got some exercise

[Nomad]

The comparison with classical computing is useful here. Nobody would announce a new CPU as revolutionary because one transistor had a fantastic lifetime while leaving the rest of the chip untested

Quantum machines are even less forgiving because the components interact in ways that make errors contagious. A qubit can look wonderfully stable in isolation and become much less impressive once you start performing gates, entangling it with neighbours, and reading the result

That is why I would like to see system-level numbers alongside the headline qubit lifetime. Gate fidelity, readout fidelity, correlated errors, scalability, and ultimately logical qubit performance tell us much more about whether this is becoming a computer rather than a very interesting physics experiment

None of that makes the materials work unimportant. Quite the opposite. If the underlying qubit really has properties that make error correction easier, that could be a major piece of the puzzle. It just is not the whole puzzle

[Mitchell_48]

Honestly, 2029 feels both too late and way too soon depending on who you ask. For Microsoft, it's probably right on time given their enterprise focus, but compared to what IBM and Google are promising? Might be playing catch-up. That said, Majorana-based qubits could be the dark horse here-if they actually work as advertised. Until then, I'll keep my quantum money in the bank, thanks.

[CosmicRay91]

One thing I would like to see is more independent replication before we start comparing Microsoft's timeline with IBM, Google, Quantinuum, or anyone else as though they are all running the same race

Different architectures have different bottlenecks. A system based on superconducting qubits has different engineering problems from trapped ions, neutral atoms, photonic approaches, or topological proposals

So asking which company will "win" can obscure the more interesting possibility that several approaches remain viable for different applications. Computing history is full of examples where several architectures coexist because each is good at different things

For now, the sensible comparison is probably milestone by milestone. Which system has demonstrated what, under what conditions, and with what independent verification? That is much more informative than putting five logos on a timeline and drawing a finish line

[Its_Neil72]

The 2029 question is a funny one because quantum computing timelines have been doing their best impression of a broken calendar for years. Every few years we get a new date for practical machines, and then reality asks for another decade

Personally, I would treat 2029 as a target rather than a promise. The important thing is what Microsoft means by a useful quantum computer by then. A small fault-tolerant demonstration and a commercially useful machine are very different milestones

There is also a difference between useful and universally useful. A quantum computer solving one specialised chemistry problem could be scientifically valuable even if it cannot replace your laptop, run spreadsheets, or make my printer behave

So whether 2029 is "late" or "early" depends entirely on the milestone. If they demonstrate a handful of reliable logical qubits, that is one kind of achievement. If they demonstrate a scalable system performing a problem beyond practical classical methods, that is another. The date alone tells us very little

[LokiCurrent]

The physicists saying it does not yet work are doing exactly what you want physicists to do. Their job is not to applaud a press release; their job is to ask whether the evidence supports the claimed interpretation

That can look hostile from the outside, especially when a company announces something flashy. But scepticism is part of the machinery. If the result survives aggressive attempts to disprove it, the eventual claim is stronger

The really useful outcome would be if the criticism leads to a clearer experiment. What observation would distinguish a genuinely protected Majorana mode from an ordinary phenomenon that happens to look similar? What measurements would settle the disagreement?

Science gets interesting when arguments become testable. Until then, we mostly have competing interpretations and very expensive refrigerators

[Tracey]

Maybe the funniest part of quantum computing is that every milestone creates another milestone immediately afterward. First you need a qubit. Then a good qubit. Then two good qubits. Then many. Then logical qubits. Then fault tolerance. Then useful algorithms. Then someone asks when they can put Doom on it

The goalposts are not necessarily moving because companies are cheating. The field is discovering that each layer has its own engineering problems

That is why I would be cautious about both extremes here. "Microsoft has solved quantum computing" is clearly premature based on the discussion, but "this proves the whole approach is useless" is equally premature

A promising physical effect can be real and important while still being nowhere near a practical computer. Those statements can comfortably coexist, which is perhaps the least exciting answer but probably the most accurate one.

[BigDog_Fan]

The 2029 target does not bother me nearly as much as the temptation to treat it as a binary prediction. Technology development rarely works like a train timetable

A project could miss a 2029 milestone and still have produced extremely valuable technology by 2030 or 2031. Conversely, hitting a demonstration milestone exactly on schedule does not necessarily mean the machine is commercially useful

What I would watch is the sequence of intermediate demonstrations. Do the devices become more reproducible? Do logical error rates improve? Does scaling introduce fewer surprises? Can independent groups reproduce key results?

If those indicators keep moving in the right direction, the exact year becomes less interesting. A roadmap should tell us what evidence is expected at each stage, not just give everyone a dramatic date to circle on the calendar

[Merchant97]

The headline numbers are absolutely bonkers, no question. But physicists saying 'it still does not work' is basically the quantum computing version of 'the food is great but the service is terrible.' You can have amazing qubit lifetimes, but if you can't string them together into a working algorithm, what's the point? Still, props to Microsoft for swinging big. Even if Majorana 2 ends up being more 'major oops' than 'major breakthrough,' at least they're trying.

[Rooney]

I am sympathetic to the physicists pushing back here. Quantum computing has a long history of headlines getting several steps ahead of the demonstrated hardware

The phrase "does not work" can sound harsher than the underlying criticism, though. There is a huge difference between saying a device has not demonstrated the required topological protection and saying the entire research programme is nonsense. Experimental physics is full of promising ideas that spend years being tortured by reality before becoming useful

For me, the sensible benchmark is reproducibility. Show the effect repeatedly, under controlled conditions, with independent measurements, and demonstrate that the claimed protection actually survives the operations needed for computation. If that happens, the argument changes very quickly

Until then, the 1000x number is interesting rather than decisive. Quantum hardware has taught us repeatedly that the last mile is where the gremlins live. There is always a gremlin