The unbreakable secret.
This golden chandelier is a dilution refrigerator. At its base, around 50 thousandths of a degree above absolute zero, colder than deep space, sits Majorana 2, the quantum chip Microsoft announced in June 2026.
To see why it matters, start with a cup of coffee.
One molecule, 10⁴⁸ bits.
Caffeine has just 24 atoms. But describing its electrons exactly is so hard that IBM has estimated it would take around 10⁴⁸ bits, a number with 49 digits. No supercomputer and no AI can hold that.
Nature does this calculation effortlessly, all the time. So why not build a computer from nature's own rules?
A coin that's still spinning.
A bit is a coin lying flat: 0 or 1. A qubit is closer to a coin in the air: a blend of both outcomes, each with a weight (an amplitude) and a timing (a phase). Measuring it makes it land, at random, according to those weights.
Wrong answers cancel out.
A quantum computer does not simply try every answer at once. Its possibilities add like waves. A good quantum algorithm arranges them so the waves for wrong answers cancel and the waves for the right answer build up. That's interference, and it's why speed-ups exist only for problems with the right structure.
Every qubit doubles the space.
Entangled qubits share one joint state. Describing n of them takes 2ⁿ numbers: 10 qubits, about a thousand; 50, a quadrillion; 300 qubits, about 10⁹⁰, more than the roughly 10⁸⁰ atoms in the observable universe.
A whisper of heat, and it's gone.
Qubits are absurdly fragile. Heat, vibration, stray light or a wandering particle leaks information into the environment, a process called decoherence, and the spinning coin collapses into noise. Many qubits hold their state for only a fraction of a second.
A choir for a single note.
The standard fix spreads one reliable logical qubit across many fragile physical qubits that constantly check each other. Common surface-code estimates run to hundreds, even around a thousand, physical qubits per logical one, depending on hardware quality.
What if a qubit could
protect itself?
Half a secret at each end.
A topological qubit stores information non-locally: split between two exotic states, Majorana zero modes, at opposite ends of a nanowire. Noise that touches one end learns nothing and changes nothing. Only a joint measurement of both ends reads the bit. The particles are named after Ettore Majorana, who predicted them in 1937.
A wire, a coat, a magnet, the cold.
Microsoft grows indium arsenide semiconductor wires (3.5 µm long, 35 nm wide) coated with a thin superconductor, then cools them to about 50 mK in a strong magnetic field (3.8 T). If it works, Majorana modes appear at the ends. Microsoft calls the material a topoconductor.
Two wires, one H, four ends.
Two wires joined in an H make one qubit, a tetron, with a Majorana at each of its four ends. Instead of finely tuned analog pulses, it's operated by measurements: digital voltage pulses connect and disconnect tiny quantum dots, like drawbridges, and read what comes back. Microsoft argues that digital control is what could let it scale to a million qubits on a chip.
The bit is a question of parity.
The stored information is parity: is the number of electrons shared by a pair of Majoranas even or odd? The enemy is quasiparticle poisoning: one stray electron slips in and flips the answer.
A flip every few milliseconds.
With aluminium-coated devices, parity flipped roughly once every millisecond or so (lifetimes of 1–12 ms). The 2025 announcement was contested: Nature published the paper with an editorial note that it was not evidence that Majorana modes were present.
From milliseconds to 22 seconds.
Swapping aluminium for lead raised the superconducting gap about fourfold (≈1.3 vs ≈0.3 meV) and the topological gap from ~30 to ~70 µeV. Parity now lasts 22 ± 1 seconds, the "1,000x" in the headline. Microsoft's analogy: a phone battery that lasted a day now lasting nearly three years.
One number, one device.
The 22 seconds is the Z-parity lifetime, one of two measurements a qubit needs; the other (X) wasn't reported. The data come from one wire of one tetron in a four-qubit prototype, in a preprint that isn't peer reviewed. Critics such as Henry Legg and Sergey Frolov remain sceptical; others call it real progress. No gates, entanglement or algorithms have been shown yet.
Everyone else is moving too.
Google's 105-qubit Willow showed errors falling as codes grow (Dec 2024). Quantinuum's Helios runs 48 error-corrected logical qubits (Nov 2025). IBM targets 200 logical qubits with Starling in 2029. Microsoft also hedges with neutral atoms from Atom Computing. Microsoft and PsiQuantum are in the final stage of DARPA's benchmarking programme, which tests whether designs can be built, not whether they're proven.
Nature's problems, not chatbots.
Quantum computers won't speed up everything. Their promise is simulating nature: catalysts for low-energy fertiliser, battery materials, drugs. And breaking today's RSA encryption, which a 2025 estimate put at under a million noisy qubits, which is why the world is moving to post-quantum cryptography. The likely partnership: AI proposes, quantum simulates.
Nature's own language.
Microsoft says a scalable, practical quantum computer could arrive by 2029. Whether topological qubits prove a shortcut or a detour, the next frontier isn't just a faster computer. It's a computer that works the way nature does.