Majorana 2 halves Microsoft's timeline to a scalable machine and reports a 20-second parity lifetime. The preprint measured one observable on one nanowire, unreviewed; critics say that is not yet a demonstrated topological qubit.
Microsoft's quantum team published two things on June 2. One was a company blog post announcing Majorana 2, a topological quantum processor the company says brings a "scalable quantum computer" within reach by 2029, roughly half the timeline it was citing a year ago. The other was a preprint on arXiv, posted without peer review, titled "20 Second Parity Lifetime in an InAs-Pb Tetron Device." The distance between those two documents is what this story examines.
Microsoft describes the result as improved qubit performance; the preprint reports parity-lifetime measurements. Those are not the same claim, and the physicists who have followed this program through two retracted Delft nanowire papers in Nature and a contested 2025 result are once again pointing at the space between them.
Strip the marketing and the experiment is narrow and specific. Microsoft's team brought up a single tetron inside a multi-tetron array and ran interferometric single-shot parity measurements on one of that tetron's hybrid nanowires. By switching the wire's parity and watching for h/2e-periodic bimodal shifts in the quantum capacitance of a coupled quantum dot, they extracted a characteristic parity switching time of about 20 seconds, with some instances reaching the minute scale. That is a real, measured number, resolved by a new rf technique the paper says reads out low-energy wire-end states and measures their energy splitting "with μeV precision."
The number is impressive and the word attached to it is where the trouble starts. A 20-second parity lifetime is not a 20-second qubit. Parity lifetime is how long the nanowire holds a fixed even-or-odd electron occupation while being passively monitored. Coherence time, the quantity that bounds how long a qubit can compute, is how long a superposition survives while gates are applied to it. The paper is unambiguous that these are different regimes: the abstract notes that "such extremely long parity lifetimes are orders of magnitude longer than typical qubit operation times, which are on the order of μs." Twenty seconds against a microsecond is about seven orders of magnitude, and the larger figure describes the quieter quantity. The paper closes by discussing "potential implications for the fidelity of Pauli measurements": implications, not demonstrated gate fidelity. None of Microsoft's headline framing, the "1,000× more reliable" and "20-second qubit" language, should be read as a coherence result, because the experiment did not operate the device as a qubit.
A qubit is defined by two non-commuting observables. You need to read it in at least two complementary bases, in this architecture an X-parity measurement and a Z-parity measurement, and show the device behaves as a two-level quantum system across both. The single-nanowire parity measurement the preprint reports corresponds to Z in Microsoft's tetron encoding; the complementary X measurement would require reading parity across a second pair of the tetron's wire ends. Majorana 2's preprint reports Z only.
That single omission is, in the critics' reading, the central objection. "Nothing in this preprint resolves the fundamental issues," Henry Legg of the University of St Andrews told Science News. "Nothing in the presented data proves the existence of a topological qubit or Majoranas in these devices." A long-lived Z-parity state in a superconducting wire is, by their account, consistent with trivial Andreev bound states that mimic Majorana signatures without carrying any of the topological protection the entire program is built to exploit. Legg was blunter still on the review posture, telling Scientific American that if the work "was from any other group or Ph.D. student, it would never make it through peer review," and flagging the perennial single-device risk: "You can see something amazing in one device and never see it again because it's just some random artifact." This is one nanowire of one prototype, not the four-qubit array in the marketing renders.
Sergey Frolov of the University of Pittsburgh, who called Microsoft's Majorana 1 data "just noise" when it was presented at last year's APS March Meeting, was similarly unmoved: the new preprint, he said, "is not based on a research track record that can be considered a solid foundation." Reproducibility across devices is the specific thing this program has historically failed to show, which is why a characterization run on a single wire, however clean, does not close the question that matters. The readout machinery itself has advanced: a QuTech/TU Delft and ICMM-CSIC team recently demonstrated single-shot parity readout in a Majorana-style minimal Kitaev-chain device, underscoring progress on readout without settling Microsoft's topology claim. A clean readout settles how a state is measured, not what it actually is.
Microsoft's own paper argues that, in its architecture, parity lifetimes translate directly into qubit-level benefits. The lifetime itself is not in question; what remains unsettled is whether the measured states have been shown to be topological and qubit-complete.
Majorana 2 does not arrive into a vacuum. Two related Delft nanowire papers in Nature were retracted in 2021 and 2022, and a Majorana 1 result published in Nature in February 2025 carried an accompanying note from Nature's editorial team in the peer-review file stating that the data do not represent evidence for the presence of Majorana zero modes. Legg's separate arXiv comment on the topological gap protocol, the very test Microsoft cites to certify its devices as topological, argues the protocol lacks a consistent definition of the "gap" it claims to find and can be fooled by trivial states that mimic Majorana-like signatures while carrying none of the useful properties. Posting Majorana 2 to arXiv and the company's own site with no peer review at launch repeats the pattern that produced the trust problem in the first place.
Microsoft is not retreating, and it deserves a fair hearing. Jason Zander, the executive vice president who runs the quantum group, framed the work to Scientific American as "the start of a new chapter," and reached for a historical analogy: "Bell Labs didn't have to prove there was an electron to invent the transistor." Roman Lutchyn, rebutting Legg's gap-protocol critique at the APS March Meeting, said flatly that "a lot of statements here are just simply incorrect." The company stands behind its papers.
Some of the skepticism comes with credit attached. Kartiek Agarwal of Argonne National Laboratory, who wants more evidence before calling the topology settled, nonetheless called the new nonlocal measurement technique "fantastic progress." The materials change underneath the result is real engineering, not a press-release flourish: swapping aluminum for a higher-gap lead superconductor, on an InAs and InAs-antimonide semiconductor stack, plausibly buys better shielding against the quasiparticle poisoning that scrambles parity. That is the most credible explanation for the longer lifetime, whether or not the states involved are topological. The honest position credits the fabrication advance and the readout method while leaving the topology claim open, which is roughly where the measured-middle of the field sits. Microsoft also credits its agentic-AI research platform with accelerating the materials and measurement work; treat that as a corporate-narrative layer over the physics rather than an independently verified cause.
The calendar is where the Practitioner Core dispute becomes a Category Frame question. Microsoft now targets 2029 for a scalable quantum computer, pulled in from roughly 2033. That is the same year IBM has set for fault tolerance, by an entirely different route of transmon qubits plus quantum error correction, backed by a published roadmap and a disclosed plan to spend more than $10 billion over five years. Two roadmaps, one finish line, and very different things to show for it. IBM's transmon-based path rests on a more widely established qubit modality, though its 2029 fault-tolerance target still carries major engineering risk; Microsoft's path still has to establish that its qubit behaves as one.
What makes the topological wager worth this much scrutiny is also what makes it attractive. If topological protection delivers intrinsically lower physical error rates, it collapses the error-correction overhead that dominates every competing roadmap, the proliferating zoo of QEC schemes that transmon and ion builders need precisely because their physical qubits are noisy. A quantum-classical hybrid machine built on topological qubits could need far fewer physical qubits per logical qubit, which is the whole economic argument for the approach. That upside is exactly why the topology claim cannot carry a 2029 timeline on assurances alone. The data point that would move the dispute is not a longer parity lifetime; it is an X measurement, on a second device, that behaves the way the theory says a Majorana qubit must.
