Majorana 2: The Quantum Bet Microsoft Can Afford to Lose

This is the story of Microsoft’s quantum bet, built from Microsoft’s own filings and investor materials and read against the public scientific record. It is a split story: I am genuinely admiring of the structure of the bet, and genuinely skeptical of the near-term science, which much of the physics community does not yet accept. In future notes I will dig deeper into the physics underneath the headlines and revise where the facts demand it.

I want to start with the strangest fact I found, because it frames everything that follows. Microsoft’s fiscal 2025 Form 10-K runs to hundreds of pages of business description, risk factors, and audited financial statements. The word “quantum” appears in it exactly zero times.¹ Search the whole annual report and you will find quantum mentioned precisely once, in a single sentence of Satya Nadella’s shareholder letter, sandwiched between datacenters and a data-analytics product, where he notes that the company “made meaningful progress on the next frontier in cloud systems: quantum,” announced Majorana 1, and deployed a Level 2 machine with a partner.² One sentence. Then he moves on to Microsoft Fabric.

In the document where Microsoft tells its owners what it does, the word “quantum” appears zero times.

So here is the puzzle. Three days ago, on June 2, 2026, Microsoft unveiled Majorana 2, a chip its scientists describe as 1,000 times more reliable than its predecessor and a step toward a machine that could “tackle intractable problems in global health, food supply, sustainability, energy production and more.”⁴ The company that made this announcement is, by revenue and profit, one of the most powerful enterprises that has ever existed. And in the document where that company tells its owners what it does and what could go wrong, the thing it is most excited about scientifically does not register at all. That gap, between the grandeur of the claim and the silence on the balance sheet, is the whole story. Let me tell it.

The machine that pays for the dream

Start with the money, because the money is the reason this bet is even possible.

In fiscal 2025 Microsoft did $281.7 billion in revenue, up 15 percent, and turned that into $128.5 billion of operating income, up 17 percent.²¹ Three segments carry the load: Productivity and Business Processes at $120.8 billion, Intelligent Cloud at $106.3 billion, and More Personal Computing at $54.6 billion.¹ The engine inside the engine is Azure, which crossed $75 billion in revenue for the first time, growing 34 percent.² This is a company that throws off more operating profit in a single year than the entire annual revenue of all but a handful of corporations on earth.

Now look at what it spends to stay ahead. Research and development ran $32.5 billion, about 12 percent of revenue.¹ Capital expenditure, the money poured into datacenters and AI silicon, hit $64.6 billion in fiscal 2025, up from $44.5 billion the year before and $28.1 billion the year before that.¹ That is not a budget; that is a flood, and it has more than doubled in two years.

Figure 1: Microsoft’s reinvestment, fiscal 2023 to 2025. Capex more than doubled to $64.6 billion in two years while R&D climbed steadily to $32.5 billion. The quantum program is funded from inside these bars; it is not large enough to show as its own stripe.

It is not a business. It is an option, paid for in pocket change, written on a future that may never arrive.

Here is why that matters for our story. The quantum effort is somewhere inside that $32.5 billion of R&D, and it is small enough that the company does not bother to break it out, name it, or warn shareholders about it as a discrete risk. Against numbers like these, Microsoft’s quantum research effort is almost certainly a rounding error, although the company does not disclose its total cost or headcount. That is the first thing to understand about Majorana 2: it is not a business. It is an option, paid for in pocket change, written on a future that may never arrive. The genius and the danger of the bet both flow from that single fact.

The twenty-year gamble on a state nobody could build

Now the science, told as the story it actually is.

Almost twenty years ago, when experimental quantum computing was still in its infancy, Microsoft made a choice that looked, for most of the intervening years, like a mistake. Everyone else building quantum computers went after qubits they could actually make: superconducting loops, trapped ions, photons. Microsoft went after a qubit nobody could yet make, because it depended on creating a highly unusual quantum state known as a Majorana zero mode. It had to be carefully engineered inside a device. It has to be coaxed into existence with superconductors and magnetic fields, in materials sprayed down atom by atom.³ The bet was that this harder path led somewhere better: a topological qubit designed to be naturally more resistant to errors. If it works as intended, it could require far less error correction than competing approaches.³

The protagonist of this chapter is Chetan Nayak, a Microsoft technical fellow who has become the public face of the effort, alongside fellow scientists Matthias Troyer and Krysta Svore. Nayak frames the whole quest in a single ambition: “Let’s invent the transistor for the quantum age. What properties does it need to have?”³ It is a Campbell-style hero’s journey rendered in condensed-matter physics: a departure into uncharted material science, a long ordeal in the wilderness, and, the company would have you believe, a return bearing the prize.

The ordeal was real, and it is the honest reversal this story needs, so I will not soften it. Along the way, the broader Majorana research effort suffered a serious credibility blow. A 2018 Nature paper was retracted in 2021 after problems were found in how the data had been selected and presented. A related 2017 Nature paper was also retracted in 2022.⁵⁶ That is not a footnote you skip past. In a field where the central particle is invisible and the measurements are exquisitely subtle, a retraction is a wound, and it taught a generation of physicists to demand the data before they believe the press release.

Figure 2: The arc of a twenty-year bet. The red node marks the two retractions, the genuine setback the rest of the story has to overcome. The two hollow nodes on the right are promises, not results: Microsoft’s own 2029 target and the 2033 horizon of the DARPA program in which it is a finalist.

Then came the return, or what Microsoft presents as the return. In February 2025 the company announced Majorana 1, which it called the world’s first quantum chip with a topological core, built on a new class of material it named a topoconductor, made of indium arsenide and aluminum.³ Microsoft described Majorana 1 as a chip with eight topological qubits, architected to scale eventually to a million, small enough to hold in your palm and slot into an Azure datacenter.³ A million qubits is the threshold the team treats as gospel: below it, Nayak argues, “you’re going to hit a wall before you get to the scale at which you can solve the really important problems.”³

What is new in Majorana 2, and the consolation prize

Which brings us to last week. Majorana 2’s headline is reliability. The team swapped aluminum for lead in the materials stack, a change that, in Nayak’s words, “led to big, big improvements in device quality.”⁴ Lead shields the fragile qubits from cosmic disturbances the way it shields radiologists in a hospital. The result, as Microsoft reports it: a key stability measurement lasted about 20 seconds on average, with some instances reaching a full minute. That is more than 1,000 times better than in its earlier devices. The company offers a lovely analogy: it is “roughly comparable to inventing a phone battery that instead of dying in a day could last for nearly three years on a single charge.”⁴ Operations run in about a microsecond, and each qubit is roughly one hundredth of a millimeter across.⁴ On the strength of this, Microsoft now says it expects a scalable, commercially valuable quantum computer by 2029, cutting its previous timeline in half.⁴

The quantum lab became the showroom. That is how a Bully hedges a moonshot: it monetizes the search itself.

There is a second, quieter move here that I think is the cleverest part of the whole story. Microsoft says Majorana 2 was developed “with the help of Microsoft Discovery’s agentic AI,” and on the same day it announced that Discovery, the platform its own quantum scientists used to automate measurements, filter noise, and hunt for materials, is now generally available for any company doing frontier research.⁴ Read that again. The quantum lab became the showroom. Even if the topological qubit never ships, the tools built to chase it are now a product Microsoft can sell into life sciences, chemicals, energy, and materials. That is how a Bully hedges a moonshot: it monetizes the search itself. The dream may or may not pay; the toolmaking already does.

The reversal, told straight

Now I have to put the brakes on, because this is where an honest story diverges sharply from a press release.

Much of the physics community does not believe Microsoft has done what the headlines imply. The skepticism is not vague grumbling; it is specific, and it is coming from serious people. When Majorana 1 was announced, it came largely by press release, and the peer-reviewed Nature paper that accompanied it did not settle the question, and Nature’s own news coverage emphasized that the published results were not evidence for the Majorana zero modes the qubit depends on.⁵⁶ A central method Microsoft uses to declare a device “topological” in the first place, the topological gap protocol, was criticized by physicists including Henry Legg for giving different answers depending on which slice of the data you feed it.⁵

“Nothing in the presented data proves the existence of a topological qubit or Majoranas in these devices.”

Majorana 2 has not quieted this. Nature’s own coverage was titled, bluntly, that researchers remain skeptical.⁸⁷ The technical objection is precise and, to my reading, devastating in its narrowness: the new result demonstrates a long-lived Z-parity state in a single superconducting wire, but it does not present the complementary X-parity measurements, shows no quantum gates, no entanglement, and no algorithm, and at announcement it had not been peer reviewed.⁸ As physicist Henry Legg told Science News, “Nothing in the presented data proves the existence of a topological qubit or Majoranas in these devices.”⁸ The 20-second stability measurement is a real and impressive engineering achievement. Whether it is a qubit lifetime, in the sense the word “qubit” is meant to carry, is exactly the thing in dispute.

So let me run the bet through the three hurdles I hold every story to. Is the topological quantum computer possible? Yes; the theory is sound and the materials progress is real. Is it plausible, given competition and physics, that Microsoft rides a long-lived parity state in one nanowire to a million-qubit fault-tolerant machine by 2029? That is a stretch, and most quantum stories die right here. Is this outcome probable, the single most likely future among all the plausible ones? Not yet, and I want to be honest that it is not close. For the topological story to clear that bar, I would need to see the missing complementary measurement, demonstrated quantum gates, peer-reviewed entanglement, and evidence that the architecture can scale from small devices to thousands of qubits. None of that exists today. What would have to be true for the story to fail at the plausibility step is depressingly easy to state: the topological test turns out to be reading artifacts, the promising result in one wire does not survive integration at scale, and another retraction lands. That is not a tail risk I am inventing for balance. It is the base case much of the field already holds.

Why a Bully can rationally fund a Disruptor’s bet

And yet I do not think the bet is foolish, and the reason is the puzzle I opened with.

For Microsoft as a whole, the archetype is unmistakable: this is a Bully. Dominant share across productivity, cloud, and developer tools; a brand that lets it set prices; the financial mass to steamroll. On the corporate life cycle it sits in mature growth, large and hugely profitable yet still compounding at 15 percent on the back of the AI platform shift. But the quantum effort inside it is a completely different creature. It is a Disruptor’s bet, the kind of swing a venture-stage startup makes: pre-revenue, pre-product, narrative almost all the way down, trying to change how the thing is fundamentally done. The fascinating structure of this story is a start-up-stage science experiment nested inside a mature-stable balance sheet, and the manager who makes that nesting coherent is Nadella, with his repeated mantra to “think in decades and execute in quarters.”² Quantum is the purest expression of the “think in decades” half he has ever funded.

When the downside is a rounding error and the upside is a new substrate for civilization, you do not need the bet to be probable. You need it to be possible, cheap, and to be in the room.

Here is the asymmetry that makes it rational. The cost of being wrong about Majorana is trivial: a sliver of a $32.5 billion R&D line, plus the reputational bruising of skeptical Nature articles, which a company this size can absorb. The cost of being absent if it is right is enormous, because a genuine topological qubit would be a foundational, winner-take-most technology, the literal transistor of a new computing age. When the downside is a rounding error and the upside is a new substrate for civilization, you do not need the bet to be probable. You need it to be possible and cheap, and you need to be in the room. Microsoft is in the room: it is one of only two companies DARPA advanced to the final co-design phase of its program for utility-scale quantum computing, the other being the photonics company PsiQuantum, with the program’s own utility horizon set around 2033.³¹² And while it chases its own qubit, Microsoft hedges by selling access to other people’s working machines through Azure Quantum, having partnered with Quantinuum and Atom Computing and deployed what it calls the first operational Level 2 quantum computer.³² The moonshot sits on top of a hedged platform business, not in place of one.

That hedging matters, because the competition is not standing still, and on the only metric that is testable today, working qubits, Microsoft is behind.

Company	Approach	Where it stands today	Posture
Microsoft	Topological (Majorana)	Claimed 8-qubit topological design; ~20-sec stability measurement in one wire; no demonstrated topological gate yet³⁸	Highest risk, highest reward; betting the qubit, not just the scale
Google	Superconducting (Willow)	105-qubit chip; claimed verifiable quantum advantage, Oct 2025⁹	Working hardware now, scaling error correction
IBM	Superconducting	Roadmap to a fault-tolerant “Starling,” ~200 logical qubits, by 2029¹⁰	Industrial roadmap, conventional qubits
IonQ	Trapped ion	Reported >99.99% two-qubit gate fidelity in an R&D demonstration¹¹	High fidelity, commercial traction
PsiQuantum	Photonics	DARPA US2QC co-finalist alongside Microsoft¹²	The other underexplored bet

The table holds the tension of the whole story. Google, IBM, and IonQ have qubits that demonstrably compute today; their challenge is scaling and error-correcting their way up. Microsoft has a claimed eight-qubit design and a contested measurement, and its entire thesis is that everyone else is climbing the wrong mountain, that conventional qubits will hit a wall that a topological qubit sails past. If Microsoft is right about the physics and its architecture scales as intended, the qubit count it is behind on today becomes irrelevant tomorrow, because its design aims for a million qubits while conventional approaches continue to wrestle with heavy error-correction demands. If Microsoft is wrong about the physics, it is years behind in a race it framed wrong. There is very little middle ground, and that is precisely what makes it a story worth telling rather than a line item worth ignoring.

From story to value drivers

Every claim I have made points to a lever someone valuing this company would have to set. I do not set them here; I only name them and anchor each to a number from the record, and hand them on.

Story claim	Value driver it implies	Supporting figure
Quantum is invisible optionality, not a business	Negligible drag on near-term margins; cost lives inside total R&D	R&D $32.5B, ~12% of revenue; “quantum” appears 0 times in the 10-K¹
The bet is funded by the AI cash machine	Reinvestment capacity; durability of the funding source	Azure +34%, >$75B; Intelligent Cloud revenue $106.3B²¹
The lab became a product (Microsoft Discovery)	New addressable market that pays even if the qubit fails	Discovery now generally available for frontier R&D⁴
A topological qubit, if it works as intended, could require far less error correction	Reinvestment efficiency; a smaller, cheaper machine	Qubit ~1/100 mm; palm-sized chip that fits in Azure datacenters³⁴
External validation lowers the perceived risk of the path	Signal on cost of capital / probability weighting	One of two DARPA US2QC finalists³¹²
The science is genuinely contested	A heavy probability discount on any quantum upside	One key measurement shown; complementary measurement still missing; no demonstrated topological gate; two relevant Nature-paper retractions⁸⁵⁶
Whole-company scale absorbs the failure case	Low downside; asymmetric payoff structure	Operating income $128.5B; cash and short-term investments $94.6B¹²

The shape of it is clear. Almost nothing in the quantum story moves a near-term number, because the program is too small to register and the science is too unproven to credit. What it changes is the width of the distribution of outcomes far out on the horizon, a tail that is cheap to hold and very fat if it pays.

What would change this story

No story is permanent, so let me say plainly what I am watching, sorted by how much each would matter.

A break, the kind of event that decimates this thread of the story, would be another retraction, or a credible independent demonstration that the topological gap protocol is reading artifacts rather than Majoranas, or the 2029 roadmap slipping by years, or DARPA dropping Microsoft from the final phase. Any one of those turns Majorana from a moonshot into a cautionary tale, though, tellingly, none of them would meaningfully dent Microsoft’s earnings, which is the whole point.

A change, a fundamental shift in the bet’s logic, would be a competitor, Google or IBM or IonQ, reaching fault-tolerant, utility-scale computing first with conventional qubits. That would not just put Microsoft behind; it would weaken the premise, the claim that you need a fundamentally better qubit to scale, and demote the topological program from “Disruptor with a hidden winning hand” to “scientifically interesting also-ran.”

A shift, the ordinary adjustment, is everything in next year’s update: the stability and fidelity numbers, the first credible complementary measurement, the first demonstrated topological gate, and early revenue traction for Microsoft Discovery, which is the consolation prize that quietly tells you whether the whole apparatus can pay for itself even if the headline dream does not.

I will leave you where I started, with the silence in the 10-K. A lesser company would have splashed Majorana across its risk factors and its investor decks and asked to be valued on the dream. Microsoft buried it in a single sentence and let the experimental evidence speak for itself, or fail to. There is a kind of confidence in that, and a kind of honesty, whatever the physicists conclude. The most valuable thing about being the size of Microsoft is not that you can make the bet. It is that you can afford to be wrong about it, and still come to work the next morning to a $281.7 billion business that does not need the qubit to work.

Sources & notes

Microsoft Corporation, Form 10-K for the fiscal year ended June 30, 2025, filed July 30, 2025, and the Microsoft 2025 Annual Report. Microsoft reported total revenue of $281,724 million; operating income of $128,528 million; research and development expense of $32,488 million, approximately 12% of revenue; additions to property and equipment of $64,551 million in fiscal 2025, $44,477 million in fiscal 2024, and $28,107 million in fiscal 2023; and cash, cash equivalents, and short-term investments of approximately $94.6 billion. Segment revenue was $120,810 million for Productivity and Business Processes, $106,265 million for Intelligent Cloud, and $54,649 million for More Personal Computing. The word “quantum” does not appear in the standalone Form 10-K.
Satya Nadella, Microsoft 2025 Shareholder Letter, October 15, 2025. Nadella reported fiscal-2025 revenue of $281.7 billion, up 15%; operating income of $128.5 billion, up 17%; and Azure revenue exceeding $75 billion, up 34%. The letter includes a brief reference to progress on “the next frontier in cloud systems: quantum,” the Majorana 1 announcement, and a Level 2 machine deployed with Atom Computing. It also uses the phrase “think in decades and execute in quarters.”
Microsoft, “Microsoft’s Majorana 1 chip carves new path for quantum computing,” February 19, 2025, and the Microsoft Azure Quantum announcement. Microsoft described Majorana 1 as the first quantum chip with a Topological Core, using an indium-arsenide and aluminum materials stack, with eight proposed topological qubits and an architecture intended eventually to scale to one million qubits; digital control through voltage pulses; a palm-sized chip intended for Azure datacenters; partnerships with Quantinuum and Atom Computing; and participation in DARPA’s US2QC program. For the peer-reviewed paper, see “Interferometric single-shot parity measurement in InAs-Al hybrid devices,” Nature 638 (2025), which states that the reported measurement alone does not unequivocally distinguish Majorana zero modes from alternative explanations.
Microsoft, “Introducing Majorana 2,” June 2, 2026, and “20 Second Parity Lifetime in an InAs-Pb Tetron Device,” arXiv preprint, June 2026. Microsoft described Majorana 2 as a next-generation topological chip using an indium-arsenide and lead materials stack, with a greater than 1,000-fold reliability improvement over its earlier aluminum-based devices. Microsoft’s public article uses the phrase “mean qubit lifetime of 20 seconds,” with some instances reaching approximately one minute, and describes intended microsecond-scale operations, a proposed qubit footprint of approximately one hundredth of a millimeter, a revised 2029 target, and the role of Microsoft Discovery. The accompanying preprint uses narrower language: an approximately 20-second characteristic parity switching time measured in one hybrid nanowire of an InAs-Pb tetron device.
Emily Conover, “Physicists are mostly unconvinced by Microsoft’s new topological quantum chip,” Science News, February 2025; and Henry Legg, “Comment on Microsoft’s topological gap protocol,” arXiv preprint, February 2025. The sources discuss scientific skepticism surrounding the Majorana 1 announcement and criticism of the topological gap protocol used to classify devices; Legg argues the protocol can produce different conclusions depending on choices made during the analysis.
Nature news coverage, 2025, on the Majorana 1 claim: “Microsoft claims quantum-computing breakthrough: but some physicists are sceptical” and “Microsoft quantum-computing claim still lacks evidence: physicists are dubious.” See also the Nature retraction notices for two earlier papers from the broader Majorana-device research effort: “Retraction Note: Quantized Majorana conductance,” 2021, and “Retraction Note: Epitaxy of advanced nanowire quantum devices,” 2022. The first describes problems involving data processing, presentation, and calibration; the second describes unjustified data removal and cropping in several figures.
Nature, “Microsoft upgrades controversial quantum chip: researchers are still sceptical,” June 2026; and Emily Conover, “Microsoft upgraded its controversial quantum chip. Physicists still have questions,” Science News, June 2026. The coverage discusses the improved parity-lifetime measurement and the remaining scientific questions surrounding its interpretation.
Microsoft Quantum team, “20 Second Parity Lifetime in an InAs-Pb Tetron Device,” arXiv preprint, June 2026; Emily Conover, “Microsoft upgraded its controversial quantum chip. Physicists still have questions,” Science News, June 2026; and Nature coverage of Majorana 2. The preprint reports an approximately 20-second characteristic parity switching time from measurements on one hybrid nanowire in an InAs-Pb tetron device. At announcement, the work had been released as a preprint rather than a peer-reviewed paper, and the coverage notes that the new result does not yet demonstrate complementary parity measurements, topological quantum gates, entanglement, or an algorithm running on Microsoft’s proposed topological architecture.
Google, “Meet Willow, our state-of-the-art quantum chip,” December 2024; Google, “Quantum Echoes: The first verifiable quantum advantage,” October 2025; and the associated peer-reviewed paper, Nature, 2025. Google described Willow as a 105-qubit superconducting processor and later reported a verifiable quantum-advantage result using the Quantum Echoes algorithm.
IBM, “The path to large-scale fault-tolerant quantum computing,” and the IBM Quantum Development and Innovation Roadmap. IBM’s roadmap places its Starling system in 2029, with approximately 200 logical qubits and the ability to execute approximately 100 million quantum gates.
IonQ, “Accelerating towards fault tolerance: Unlocking greater than 99.99% two-qubit gate fidelity,” and the associated technical preprint, “Trapped-ion two-qubit gates with greater than 99.99% fidelity without ground-state cooling,” arXiv, 2025. IonQ reported a greater than 99.99% two-qubit-gate-fidelity result in an experimental trapped-ion demonstration.
DARPA, “DARPA advances quantum computing approaches to final phase of US2QC program,” 2025; the Underexplored Systems for Utility-Scale Quantum Computing program page; and the Quantum Benchmarking Initiative program page. DARPA advanced Microsoft and PsiQuantum to a final validation and co-design stage intended to evaluate whether their approaches could support utility-scale quantum computers. The broader benchmarking initiative is designed to test whether an industrially useful quantum computer can be built by 2033. Selection for evaluation is a meaningful external signal, but not confirmation that Microsoft has demonstrated a working topological qubit.