Microsoft recently released its new generation quantum chip, Majorana 2. The company claims the average survival time of qubits on this chip has reached an astonishing 20 seconds, a 1000-fold improvement in reliability over the previous generation. Based on this, Microsoft made a bold statement: "By 2029, we will have a commercially valuable, scalable quantum computer." Just last year, the mainstream industry expectation was "ten years from now." Now, Microsoft has directly halved that timeframe.
In the R&D process of this chip, Microsoft heavily leveraged its own AI platform "Microsoft Discovery's" agentic AI, allowing AI teams to collaborate like human research teams, autonomously analyzing massive experimental data, proposing hypotheses, and optimizing manufacturing processes. One is the hardware breakthrough in quantum computing, the other is the software assist from artificial intelligence. Two of the most cutting-edge fields are now reinforcing each other.
What is a Quantum Chip?
A quantum chip manipulates qubits (quantum bits). A qubit can exist in a superposition of 0 and 1. Before measurement, it's like a spinning coin, simultaneously carrying the possibility of both 0 and 1. The superposition state of two qubits can encompass four possibilities: 00, 01, 10, 11; three qubits correspond to eight possibilities... and so on. The quantum state of n qubits is 2^n-dimensional. Through carefully designed operations, quantum gates (the logical gates in quantum computing) manipulate the phase of the quantum state, causing probability amplitudes to interfere within the superposition, thereby amplifying the correct answer. This is why quantum computers are said to possess "exponential computational power."
Furthermore, two entangled qubits share a peculiar correlation: measuring one instantly determines the state of the other, regardless of the distance between them. By leveraging these quantum properties, quantum computers are expected to tackle tasks that are difficult for classical computers to perform efficiently.
A quantum chip is a specialized processor used to generate, manipulate, and measure these qubits. It doesn't use traditional transistors but employs superconducting circuits, trapped ions, photons, or even topological materials to capture quantum states and make them perform calculations according to a prescribed logic, i.e., quantum gates.
Weaknesses of Quantum Chips
Despite their power, qubits possess a fatal weakness: extreme sensitivity and fragility.
The superposition state of a qubit can collapse instantly into a definite 0 or 1 when subjected to even minor external disturbances, such as temperature fluctuations, electromagnetic radiation, or cosmic rays, thereby losing its parallel computing capability. This phenomenon is called decoherence.
Before Microsoft's Majorana 1 chip, the typical lifetime of mainstream superconducting qubits was only tens of microseconds. This meant they would often 'die' shortly after being prepared, before much computation could be done. Therefore, a key metric for evaluating a quantum chip is the qubit lifetime, also known as coherence time.
Microsoft's claim of a 20-second qubit lifespan has caused an earthquake-like reaction within the industry. For quantum operations, 20 seconds is an astronomical figure. Performing a quantum gate operation only requires a microsecond (one-millionth of a second). Twenty seconds theoretically allows for twenty million operations, enough to run fairly complex quantum algorithms. Microsoft even drew an analogy: "This improvement is roughly equivalent to inventing a phone battery that lasts nearly three years on a single charge, compared to the original one-day battery."
Twenty seconds is just the average; some qubits can even last up to a minute. The previous generation Majorana 1 only achieved a lifetime on the millisecond scale, which is why Microsoft states "reliability improved 1000-fold."
So, how did Microsoft achieve this? The answer lies in its chosen technical route: topological quantum computing.
Microsoft's Secret Weapon: Topological Qubits
Most mainstream quantum chips, like those from Google and IBM, use superconducting qubits. This technology is relatively mature, but to avoid environmental interference, it requires extremely low temperatures, close to absolute zero (-273°C). It also suffers from short lifetimes and is prone to errors.
For 20 years, Microsoft has pursued a different, theoretically more advantageous but more challenging path: topological qubits.
If you punch one hole or two holes in a piece of paper and crumple it up, the paper deforms, but the holes remain. One hole doesn't become two, and two don't become one. The number of holes in the paper is a topological invariant. Similarly, the sequence in which two ropes are braided together is also a topological invariant. Topological qubits leverage this topological invariance to protect quantum information. The information isn't stored in specific particles but in the braiding texture formed by the exchange of positions of quasiparticles (collective excitations in particle systems). This storage method is non-local, meaning small disturbances like noise and heat struggle to disrupt the overall topological structure. Consequently, topological qubits are inherently much less sensitive to environmental noise and far more stable than other types of qubits.
The quasiparticle Microsoft uses has a legendary name: Majorana fermion. In 1937, Italian physicist Ettore Majorana predicted the existence of a peculiar fermion whose antiparticle is itself. This particle hasn't been conclusively discovered yet. In the early 21st century, scientists began searching for its analog in condensed matter physics: a quasiparticle called a Majorana zero mode (MZM). When two Majorana zero modes exchange positions in two-dimensional space, the overall quantum state changes; the order of exchange affects the final outcome, similar to how different braiding methods result in different plaits.
In 1997, physicist Alexei Kitaev, then at the Landau Institute in Russia, first theorized the use of Majorana fermions for topological quantum computing. In 2005, Microsoft established Station Q, with Kitaev as a core member at the time. Microsoft has since been committed to this technical path, investing nearly 20 years. In 2025, Microsoft released the first-generation Majorana chip, proving the principle feasibility of topological qubits. They revolutionarily used a topological superconductor, enabling a novel state of matter for more stable quantum computing. Today's Majorana 2 represents a tangible leap in performance, turning principle into practice.
A key improvement lies in the material change: the first-generation Majorana chip used aluminum for its topological superconductor, while the second generation switched to lead. Lead is commonly used as radiation shielding. Using it as a superconductor can significantly thicken the qubit's shield, protecting the fragile quantum state from cosmic ray interference. This seemingly non-disruptive change, coupled with AI optimization of hundreds of process parameters, ultimately led to the 1000-fold reliability improvement.
However, currently, Majorana 2 only integrates 12 qubits. To achieve a commercially valuable universal quantum computer, the industry consensus is that at least several million qubits are needed. Bridging the gap from 12 to 1 million involves countless engineering and physics challenges. Microsoft's confidence in stating 2029 indicates strong belief in their topological route. Theoretically, the error-correction overhead for topological qubits is much lower than other mainstream schemes. If successfully implemented, they have the potential to mature faster than other approaches.
AI's Contribution: How Agentic AI Accelerates Quantum Chip R&D
Another crucial factor in Microsoft's 1000-fold reliability leap is the "assist" from agentic AI. Microsoft's Microsoft Discovery platform deploys agentic AI, where multiple AI agents can take on different roles, such as data analyst, experimental designer, and literature researcher, autonomously completing scientific research workflows under human scientist guidance.
It starts with the core material of the Majorana chip. The first generation used aluminum as the superconductor, while the second generation switched to lead. Changing materials is a complex chain reaction. The team spent years understanding all the trade-offs. Finding the precise doping recipe required hundreds of experiments. Now, AI can first identify high-probability targets through simulation, ideally reducing the required experiments to just one.
This is just the beginning. Quantum chip manufacturing involves countless aspects: software, architecture, material stack, processes, measurement, etc. Changing one parameter can trigger chain reactions. Human engineers struggle to monitor all variables simultaneously, but AI agents can. Crucially, Microsoft's quantum team has accumulated nearly two decades of vast experimental data, in various formats, held by scientists in different countries with different backgrounds. AI agents can synthesize this data and uncover correlations invisible to humans, as no single person possesses such a broad perspective.
Another AI killer app is accelerating experiments. Creating topological quantum states requires simultaneously adjusting hundreds of voltage parameters, followed by measurement—the most time-consuming and delicate part of quantum computing. Previously, a scientist might take weeks to manually complete one round of measurement. The team tried automation with early machine learning methods but failed. It wasn't until they trained a dedicated AI agent using the Microsoft Discovery platform that they reduced the entire cycle by several orders of magnitude. AI can scan the entire parameter space in parallel, automatically determining the lowest point where everything can function normally and pinpointing it.
Finally, AI also helped solve the "ghost noise" problem. At one point, experimental data was consistently off, and scientists spent a long time troubleshooting without success. Later, an AI agent integrated physical models, device logs, and process knowledge to identify an uncalibrated temperature sensor from the raw data that was subtly corrupting the measurement results.
It's fair to say that without AI's involvement, the 1000-fold performance leap of Majorana 2 might have taken several more years to achieve. This reinforces a growing consensus: quantum computing and artificial intelligence can mutually reinforce each other. AI accelerates the R&D of quantum computing hardware, and future quantum computers will, in turn, boost AI by providing exponential computational power for machine learning.
Will the Boots Land?
Microsoft is not the only player in this arena. The path to the "quantum shore" isn't limited to topological quantum chips; there are also superconducting quantum chips, trapped-ion chips, photonic quantum chips, and silicon spin qubits. Governments worldwide are also ramping up investments. China has large-scale deployments in quantum communication and quantum computing; the US has allocated significant funding to quantum computing companies; the EU has launched its "Quantum Flagship" initiative.
Can Microsoft truly deliver a commercial-grade quantum computer by 2029? Paul Stevenson, a physics professor at the University of Surrey in the UK, commented that Microsoft seems to have made a breakthrough in manufacturing reliable qubits. If the results hold up to scrutiny, the timeline sounds reasonable. However, many scientists are hoping to see more detailed data that has undergone peer review, as Microsoft's related papers on this achievement have not yet completed the peer-review process.
Of course, amidst Microsoft's high-profile announcements and excitement, several questions warrant calm consideration. First, are 20 seconds enough? A 20-second qubit lifetime is indeed a staggering leap compared to tens of microseconds. However, practical quantum algorithms require hundreds of millions or billions of quantum gate operations. Even at one microsecond per operation, 20 seconds only allows for twenty million steps, still several orders of magnitude short of the numbers needed for tasks like cracking RSA encryption or precisely simulating drug molecules. Remember, decoherence is a limit set by physical laws, something engineering can never completely overcome. Second, there's the compilation cost issue. Every time a quantum computer solves a problem, it first requires compiling on a classical computer to translate the problem into a specific quantum circuit, and then solving equations based on the quantum chip's parameters to obtain the electromagnetic pulse sequences corresponding to the quantum gates. This compilation process isn't universal; it's one-off per problem. Moreover, the classical computing power consumed by compilation itself might be comparable to, or even exceed, the cost of solving the problem directly using classical methods. Third, what if the quantum computer outputs a wrong answer? Humans cannot verify it with a classical computer; if they could, they wouldn't need the quantum computer in the first place. If the final answer is wrong, there's no way to know where the error occurred.
The dream of building a commercial-grade quantum computer is like a boot hanging in the air,迟迟不肯落地. One day, even if it lands, it might just be a dull thud. Looking back at the history of science and technology, progress sometimes resembles "willows planted unintentionally grow into shade, while flowers cultivated with great care fail to bloom." What is highly anticipated may not be realized, and breakthroughs might emerge from the most unexpected places.
References
https://news.microsoft.com/source/features/innovation/majorana-2-microsoft-discovery-agentic-ai/
https://www.bluequbit.io/blog/quantum-chips
https://www.bbc.com/news/articles/cj4p7gyvp52o
https://zhuanlan.zhihu.com/p/2035004303467917427?share_code=14f9XN3e5wlBq&utm_psn=2035105136662553502&utm_source=wechat_session&utm_medium=social&s_r=0&wechatShare=1
This article is from WeChat public account: 心智观察所 , Author: 心智观察所
