Microsoft Majorana 1 and Majorana 2: A Full Guide to Topological Qubits, Tetrons, the 2029 Target and Limits

Microsoft says its newest quantum chip holds a qubit's state for about 20 seconds, more than a thousand times longer than the chip it replaced. Physicists who reanalyzed the first chip's data say Microsoft still hasn't shown it has a working topological qubit at all.

Both statements are on the record, in Microsoft's own blog and in Nature. That gap between a 2029 product roadmap and an unsettled physics question is the real story of Microsoft Majorana.

This full guide covers what Majorana 1 and Majorana 2 are, how a tetron qubit is supposed to work, what each paper actually measured, the critiques and Microsoft's replies, where DARPA fits, how the approach compares with trapped ions, neutral atoms and photonics, and the honest limits. Last updated: November 9, 2026.

Key Takeaways

• Majorana 1 (February 2025) placed eight topological qubits on a chip Microsoft says is designed to scale to one million.
• Majorana 2 (June 2026) swaps aluminum for lead and reports parity lifetimes averaging 20 seconds, versus 1 to 12 milliseconds before: a claimed 1,000x gain.
• The Majorana 2 preprint reports Z-parity measurements on one nanowire of one tetron. It does not report the X measurements a working qubit also needs, as Science News notes.
• A June 2026 Nature Matters Arising by Henry Legg argues the 2025 parity readout happened in disordered, apparently gapless regions. Microsoft's published reply disputes it.
• Microsoft is in the final phase of DARPA's US2QC program and now targets a scalable quantum computer by 2029, cut from a 2033 timescale.

What Microsoft Majorana actually is

Majorana is the name Microsoft gives to its topological quantum processors. Majorana 1 was announced on February 19, 2025, alongside a Nature paper. Majorana 2 followed at Microsoft Build on June 2, 2026.

Neither chip is a product you can rent. Microsoft hasn't put a Majorana device on Azure Quantum, published a gate fidelity table, or run a customer workload on one. They're research processors, and the public evidence is a set of physics papers.

That framing matters because the marketing language is product language. Microsoft calls Majorana 1 the world's first quantum processing unit powered by a Topological Core, and says the design can reach a million qubits on a single chip. Treat that as an architectural ambition, not a demonstrated capacity.

The topoconductor, in plain terms

Microsoft's chips combine a semiconductor (indium arsenide) with a superconductor. On Majorana 1 the superconductor was aluminum. Cooled near absolute zero and tuned with a magnetic field, a gate-defined nanowire in that stack is supposed to enter a topological superconducting phase with Majorana zero modes (MZMs) at each end. Microsoft brands that material stack a topoconductor.

An MZM is a quasiparticle state, not a particle you can isolate. Two of them share one fermion's worth of information. The wire's parity, whether it holds an even or odd number of electrons, becomes the stored bit, and in theory a local disturbance can't read or flip it without touching both ends at once.

That's the whole pitch. If the protection is real, topological qubits need far less error correction than superconducting transmons or trapped ions. If it isn't, the device is an unusually complicated way to build a noisy qubit.

How a tetron qubit is supposed to work

The tetron is Microsoft's single-qubit building block. Two parallel topological nanowires are joined by a trivial superconducting backbone, giving four MZMs, one at each wire end. Microsoft's roadmap figure draws it as an H shape.

Operations are measurements, not analog rotations. Digital pulses connect quantum dots to the wire ends, and microwave reflectometry reads the dots' quantum capacitance, which shifts with the wire's parity. Measuring one wire's parity is a Pauli-Z measurement. Measuring across both wires at one end is a Pauli-X measurement.

You need both. A Z measurement alone tells you a two-state system exists and how long it holds. Only Z and X together show you can prepare and read superpositions, which is what makes it a qubit. Remember that distinction, because it's where most of the 2026 argument lives.

The roadmap Microsoft published

Microsoft's February 2025 blog laid out four device steps. First, a single tetron. Next, a two-qubit device for measurement-based braiding. Then a 4x2 tetron array to run quantum error detection on two logical qubits. Finally, a 27x13 tetron array for full quantum error correction.

As of November 2026, the public papers cover the first step. Majorana 2 is fabricated as a multi-tetron device, and the June 2026 preprint brings up one tetron inside a prototype array. No entanglement, braiding or error-detection result on a Majorana chip has been published.

Majorana 1: what the February 2025 paper showed

The Nature paper behind Majorana 1 is titled Interferometric single-shot parity measurement in InAs-Al hybrid devices. It reports a single-shot parity readout with a 1% assignment error, a quantum-capacitance shift of up to 1 fF, and parity dwell times longer than 1 ms at in-plane fields near 2 T.

The paper is more careful than the press launch. In its own words, the measurements do not, by themselves, determine whether the detected low-energy states are topological. The authors argue the data tightly constrain trivial explanations such as fine-tuned Andreev bound states, but they stop short of claiming proof.

Microsoft's blog went further. It called the result a hardware-protected topological qubit and reported a parity flip roughly once per millisecond on average, a sign the shielding was working. It also said Station Q data showed both Z and X measurements on a tetron.

The X measurement arrived later, and drew its own critics

The X and Z tetron data were posted as an arXiv preprint in July 2025. It reports parity switching times of 14.5 microseconds in the X loop and 12.4 milliseconds in the Z loop, with assignment errors of 16% for X and 0.5% for Z.

A 16% X readout error is a long way from fault-tolerant territory. The authors attribute the short X timescale to intra-wire parity switches. Critics, including Sergey Frolov and Henry Legg, weren't persuaded the data show topological behavior, as Science News reports.

Majorana 2: what changed in June 2026

Majorana 2 is a materials change first. Microsoft replaced the aluminum superconductor with lead and moved the semiconductor active region to a composite of indium arsenide and indium arsenide antimonide. Microsoft says the resulting topological gap is more than double the first chip's.

The headline number is lifetime. Microsoft reports qubit lifetimes between 1 and 12 milliseconds on aluminum-based Majorana 1 devices, and lifetimes averaging 20 seconds on Majorana 2, occasionally longer than one minute. That's where the 1,000x claim comes from.

Microsoft also credits its Microsoft Discovery agentic AI platform with helping design the new material stack. The Majorana 2 blog frames AI as a reason the timeline moved, which is a marketing link between two Build announcements more than a physics argument.

What the Majorana 2 preprint actually measured

The technical paper is 20 Second Parity Lifetime in an InAs-Pb Tetron Device. It brings up one device in a multi-tetron array and performs parity measurements on one of that tetron's two nanowires, aggregating 324 dwell intervals.

The preprint compares those Z-parity lifetimes with roughly 112 ms in earlier aluminum tetrons and argues non-equilibrium quasiparticles no longer limit qubit operations. It reports no X measurement, a gap Science News flagged at launch. Microsoft's Chetan Nayak has told reporters the team holds unpublished data showing full qubit control.

Source: Microsoft Majorana 1 and Majorana 2 blogs; Nature 2025; arXiv 2507.08795; arXiv 2606.03884.

The dispute: Legg's Nature critique and Microsoft's reply

On June 24, 2026, Nature published a Matters Arising by Henry Legg of the University of St Andrews titled On the robustness of topological gap detection via transport. He reanalyzed the transport data behind the 2025 paper, data that the original article didn't present.

His argument has two parts. First, the claimed parity readout happened in regions with considerable disorder that appear gapless, which would undercut the topological interpretation. Second, the topological gap protocol (TGP) Microsoft used to find those regions gives different answers depending on field range and junction settings, and Legg reports coding artefacts in the original TGP outcomes.

Microsoft's reply in Nature says its quantum-capacitance analysis doesn't assume a gap, and that a gapless system wouldn't produce a stable h/2e-periodic bimodal signal. Nayak told reporters that Legg hasn't proposed an alternative model that fits all of the data.

Frolov went further, telling Scientific American the paper likely needs to be retracted. Legg said the 2029 roadmap can't be sold as credible when the underlying physics is not there.

Why the field is wary: the 2021 retraction

This isn't the first round. In March 2021, Nature retracted the 2018 paper Quantized Majorana conductance from a Delft-led team whose corresponding author, Leo Kouwenhoven, was listed with Microsoft Station Q Delft. That retraction is why many physicists now ask for raw data and full parameter sweeps before accepting any Majorana claim.

It also explains the tone. Nayak acknowledged to Science that Microsoft had been cagey with peers in the past, and Nature's news team described Majorana 2 as an upgrade to a highly controversial chip.

DARPA, US2QC and the 2029 target

Microsoft's strongest outside validation is DARPA. In February 2025 DARPA selected Microsoft and PsiQuantum for the Validation and Co-Design stage of US2QC, the pilot program that grew into the Quantum Benchmarking Initiative (QBI). DARPA says that final phase has the same goals as QBI Stage C: verifying whether a design can reach utility scale by 2033.

Context matters. DARPA picked 11 companies for QBI Stage B in November 2025, including IBM, IonQ, Quantinuum and QuEra. Microsoft and PsiQuantum skipped ahead because they were already in US2QC, not because DARPA ranked them above everyone else.

DARPA's role is to verify and validate a design, components and prototypes. It's a serious review by a government team of more than 50 experts, but it isn't a statement that topological qubits work. Inside Deep Tech's PsiQuantum Omega guide covers the other company in that same final phase.

PsiQuantum Omega: A Full Guide to the Photonic Chipset, Million-Qubit Ambition, Brisbane/Chicago Buildouts, and Honest Limits

The other company in DARPA's final US2QC phase, and how photonic qubits approach utility scale.

Inside Deep Tech

What 2029 actually means

Nayak told reporters Microsoft had talked about a scalable machine on a 2033 timescale and now believes it can reach it by 2029. The Majorana 2 blog calls this cutting the timeline in half. Microsoft hasn't published a logical qubit count, error rate or application target for that 2029 machine.

Compare that with the public roadmap it already has. The 4x2 error-detection array and the 27x13 error-correction array are still ahead of any published result. Going from one measured nanowire to a scalable fault-tolerant computer in roughly three years would be faster than any qubit modality has moved.

How Microsoft Majorana compares with other qubit approaches

Microsoft isn't betting only on Majorana. It co-engineered Magne with Atom Computing and QuNorth in Copenhagen, a neutral-atom machine targeting about 50 logical qubits from more than 1,200 physical qubits, with availability listed for early 2027. That hedge says a lot about how Microsoft itself weighs timing risk.

Source: Microsoft Majorana 2 blog and arXiv 2606.03884; QuNorth Magne specs; Inside Deep Tech guides linked in the table.

The difference is the type of risk. Ion and atom platforms already run logical-qubit experiments, so their risk is engineering and cost. Microsoft's risk is still partly physics. If the topological phase holds up, its upside is larger. If it doesn't, there's no partial credit. For the wider error-correction picture, see Inside Deep Tech's quantum error correction in 2026 explainer.

What this means for you

• Enterprise quantum buyers: there's nothing to buy. Plan pilots on trapped-ion, neutral-atom or superconducting systems that you can access today, and treat Majorana as a 2029-plus option.
• Investors: Majorana is a line item inside Microsoft, not a pure-play stock. Its news moves sentiment across quantum names more than it moves Microsoft. The Top 100 Quantum Companies directory maps the listed and private alternatives.
• Researchers and engineers: the useful signals are papers, not keynotes. Watch for X-measurement data on lead-based tetrons and any independent replication.
• Policy and security teams: a 2029 claim from one vendor shouldn't change post-quantum migration plans. Those timelines rest on any credible fault-tolerant machine, not on Microsoft alone.

Inside Deep Tech's take

Honest limits

• No published qubit operation on Majorana 2. The preprint measures Z parity on one wire of one tetron. Lifetime of a parity state isn't the same as coherence of a superposition.
• The Majorana 1 evidence is formally contested. Nature has published Legg's critique and Microsoft's reply. Neither side has retracted.
• No independent replication. As of November 2026, no outside lab has reproduced Microsoft's topological parity readout.
• The roadmap's middle steps are unpublished. The 4x2 error-detection array and two-qubit braiding results haven't appeared.
• 2029 has no published specification. Microsoft hasn't defined logical qubits, error rates or workloads for the scalable machine.
• Unpublished data can't be checked. Nayak's statements to reporters about unpublished control data can't be assessed until the data are released.

FAQ

What to watch next

• Published X-measurement data on lead-based tetrons, ideally with lower error than the 16% reported on aluminum devices.
• Peer review of the Majorana 2 preprint, and whether reviewers ask for the full tuning data critics want.
• Any two-qubit braiding or 4x2 error-detection result, the next steps on Microsoft's own roadmap.
• Independent replication of topological parity readout outside Microsoft.
• Whether Magne delivers its logical qubits on time, which would show how Microsoft's software stack performs while Majorana matures.

Microsoft has made the topological bet more concrete than ever, with a material change, a real lifetime number and a date. The next proof point is simple to state and hard to fake: a published qubit, measured in both bases, that someone outside Redmond can reproduce.