In December 2024, Google announced that its new quantum chip, Willow, had completed a calculation in under five minutes that would take the world’s fastest supercomputer 10 septillion years. That number is so large it makes the age of the universe look like a rounding error. A few months later, D-Wave published peer-reviewed results showing its quantum annealer could simulate magnetic materials in minutes that would take a classical supercomputer nearly a million years. And in October 2025, Google followed up with an algorithm that ran 13,000 times faster than anything a supercomputer could manage.
These are real results, published in Nature and Science, from machines that exist today. So does that mean quantum computers have won? Not exactly. The reality is more nuanced, more interesting, and more important to understand than the headlines suggest.
What quantum computers actually do differently
Classical computers process information as bits: ones and zeros. Every calculation, from sending an email to training an AI model, comes down to flipping and reading those bits very quickly. Modern processors handle billions of operations per second, and they are extraordinarily good at it.
Quantum computers use qubitsA quantum bit, the basic unit of information in a quantum computer. Unlike a classical bit, it can exist as both 0 and 1 simultaneously until measured. instead. A qubit can exist in a “superposition” of both 0 and 1 at the same time, and multiple qubits can become “entangled,” meaning the state of one instantly influences the others. This is not just faster classical computing. It is a fundamentally different way of processing information that lets quantum machines explore many possible solutions simultaneously.
Think of it this way: a classical computer trying to find its way through a maze tries one path at a time. A quantum computer can, in a sense, explore all paths at once. For certain types of problems, this gives quantum machines an exponential advantage.
But here is the critical caveat: “certain types of problems” is doing a lot of work in that sentence.
What “quantum advantage” actually means
When researchers say a quantum computer has achieved “quantum advantage” or “quantum supremacy,” they mean it has solved a specific problem faster than any known classical method. That is a narrower claim than it sounds.
Google’s 2024 Willow result, for instance, used a benchmark called Random Circuit Sampling (RCS). This tests whether a quantum computer can produce output that a classical computer cannot efficiently replicate. Willow passed that test decisively. But RCS has no known practical application. It is a proof of capability, not a product.
D-Wave’s 2025 result was different and arguably more significant: it solved an actual materials science problem, simulating the behavior of magnetic materials called spin glasses. This has real applications in understanding superconductors, sensors, and electronic components. It was published in Science and is considered the first demonstration of quantum supremacy on a useful problem.
Then came the most mathematically rigorous proof yet. In September 2025, researchers from UT Austin and Quantinuum demonstrated what they called “unconditional” quantum supremacy: a memory task that 12 qubits could solve but that would require at least 62 classical bits. Unlike previous claims, no future classical algorithm can close this gap. The math proves it impossible.
The error problem
The biggest obstacle standing between today’s quantum computers and practical usefulness is errors. Qubits are extraordinarily fragile. They need to be cooled to temperatures colder than outer space, and even then, the slightest disturbance can corrupt a calculation.
Google’s Willow chip made a genuine breakthrough here. Its Nature paper showed that adding more qubits actually reduced the error rate, cutting it in half with each step up in code size. This is called going “below threshold,” and researchers had been chasing it for nearly 30 years since Peter Shor first proposed quantum error correctionA method of encoding a logical qubit across many physical qubits to protect quantum information from errors caused by noise and decoherence. in 1995.
But context matters. Willow’s logical error rate of about 0.143% per cycle is still far above the roughly 0.0001% error rates that large-scale, fault-tolerant quantum algorithms would need. The gap between “below threshold” and “actually useful at scale” remains significant.
What quantum computers are good at (and what they are not)
Quantum computers will not replace your laptop. They will not run your spreadsheets faster or make your web browser snappier. They are not better at general-purpose computing. For the vast majority of tasks, classical computers will remain superior, cheaper, and more practical.
Where quantum machines show genuine promise is in problems that are inherently quantum or combinatorially explosive:
- Molecular simulation: Understanding how molecules interact is fundamental to drug discovery and materials science. Classical computers struggle to model quantum behavior at the atomic level because they are, by definition, not quantum. Google’s Quantum Echoes algorithm demonstrated molecular structure analysis that matched Nuclear Magnetic Resonance data, pointing toward a future “quantum-scope” for chemistry.
- Optimization: Logistics routing, portfolio optimization, and supply chain management involve finding the best solution among astronomically many possibilities. Quantum annealingA quantum computing approach that finds optimal solutions to complex problems by using quantum tunneling to escape local minima, suited for optimization tasks., the approach D-Wave uses, shows promise here.
- Cryptography: A sufficiently powerful quantum computer could break most current encryption. Recent estimates suggest breaking RSA-2048 might require around one million qubits, down from earlier estimates of 20 million. We are nowhere near that today, but the trajectory matters.
- Materials discovery: Simulating new materials for batteries, solar cells, or superconductors is a natural fit for quantum computing.
Bain & Company estimates the total market potential at up to $250 billion across pharmaceuticals, finance, logistics, and materials science. But today’s market for quantum computing hardware and services is less than $1 billion annually. The gap between potential and reality remains vast.
The race to build a real quantum computer
The major technology companies are pursuing different approaches, and the diversity itself is telling: nobody has figured out the winning design yet.
Google uses superconducting qubits and is focused on scaling error correction. Its Willow chip has 105 qubits, and the company’s next goal is a long-lived logical qubit as Milestone 3 on its roadmap.
IBM is building what it calls “quantum-centric supercomputers,” combining quantum processors with classical ones. Its Heron processor can run circuits with up to 5,000 two-qubit gate operations, and IBM aims to deliver fault-tolerant computing by 2029 with its Starling processor: 200 logical qubits running circuits of 100 million gates.
Microsoft took a radically different bet with Majorana 1, the first processor based on topological qubits. These store information in exotic quantum states that are inherently more resistant to errors. Microsoft has placed eight topological qubits on a chip designed to scale to one million. DARPA selected Microsoft for the final phase of its program to build a utility-scale quantum computer.
D-Wave offers quantum annealing systems that are already commercially available, with its Advantage2 processor handling thousands of qubits. Its approach is narrower but more immediately practical for optimization problems.
The honest timeline
The quantum computing field has a history of overpromising. For years, practical quantum computing has been “five to ten years away.” The honest assessment in 2026 is that we are in what the industry calls the “fault-tolerant foundation era.” Significant engineering problems have been solved in the lab. But translating lab results to machines that can run useful algorithms reliably, at scale, for hours at a time, remains a formidable challenge.
The Riverlane QEC Report highlights a critical workforce problem: there are only an estimated 600 to 700 quantum error correction specialists worldwide, but 5,000 to 16,000 are needed by 2030. You cannot build a revolution without the people who understand the machines.
Most experts now agree that quantum computers will not replace classical ones but will work alongside them in hybrid architectures. Bain’s analysis projects that the quantum computing market will reach $5 billion to $15 billion by 2035, with early practical wins in molecular simulation and optimization. The full potential depends on breakthroughs that have not happened yet.
What this means for you
If you are not a quantum physicist or a CISO, quantum computing probably will not affect your daily life for several more years. But it is no longer a theoretical curiosity. Real machines are solving real problems faster than classical supercomputers, even if those problems are still narrow.
The most immediate practical concern is cybersecurity. The “harvest now, decrypt later” threat, where attackers store encrypted data today to crack it with future quantum computers, is already motivating governments and enterprises to adopt post-quantum cryptographyEncryption methods designed to remain secure against quantum computers, which can break widely used algorithms like RSA by exploiting quantum properties.. If you handle sensitive data with a long shelf life, this is worth paying attention to now.
Quantum advantage is real. It is also specific, limited, and early. The machine that changes everything has not been built yet. But for the first time, the components that will make it possible have been demonstrated to work.
In December 2024, Google’s Quantum AI team published results in Nature demonstrating that their 105-qubitA quantum bit, the basic unit of information in a quantum computer. Unlike a classical bit, it can exist as both 0 and 1 simultaneously until measured. Willow processor had achieved below-threshold quantum error correctionA method of encoding a logical qubit across many physical qubits to protect quantum information from errors caused by noise and decoherence. using the surface code. The headline number, completing Random Circuit Sampling in under five minutes versus an estimated 1025 years on Frontier, is dramatic. But the real breakthrough was the error suppression factor: Λ = 2.14 ± 0.02 when increasing code distance by two, culminating in a distance-7 code with 0.143% ± 0.003% error per cycle. For the first time, adding qubits to a superconducting quantum processor made it more reliable, not less.
This result arrived alongside several other milestone demonstrations in 2025, collectively redefining what “quantum advantage” means in practice and bringing the field closer to useful, fault-tolerant quantum computation.
Below threshold: why it matters
Quantum error correction (QEC) encodes a logical qubit across multiple physical qubits to protect against decoherence and gate errors. The surface code, the most studied QEC code, relates logical error rate to physical error rate through the approximate relation εd ∝ (p/pthr)(d+1)/2, where d is the code distance, p is the physical error rate, and pthr is the threshold. When p < pthr, the logical error rate is suppressed exponentially with increasing code distance.
Google’s Willow demonstrated this behavior across distance-3, distance-5, and distance-7 surface codes on two processors (72-qubit and 105-qubit). The distance-7 logical qubit achieved a lifetime of 291 ± 6 μs, exceeding the best constituent physical qubit lifetime (119 ± 13 μs) by a factor of 2.4 ± 0.3. This “beyond breakeven” result is an unfakable sign that error correction is genuinely improving the system.
However, the 0.143% logical error per cycle is still orders of magnitude above the ~10-6 rates needed for algorithms like Shor’s factoring or practical quantum chemistry simulations. The gap between “below threshold” and “fault-tolerant at scale” remains the central engineering challenge.
The 2025 quantum advantage landscape
Three distinct demonstrations in 2024-2025 collectively established quantum advantage as an empirical reality rather than a theoretical promise.
Random Circuit Sampling (Google, December 2024)
Willow’s RCS result is the strongest separation yet between quantum and classical computation on this benchmark. Google’s estimate of 1025 classical years assumed generous conditions for the Frontier supercomputer, including unlimited secondary storage with no bandwidth overhead. Classical algorithms have improved since Google’s 2019 Sycamore demonstration, but the gap is growing at a double exponential rate. RCS remains classically hardest but has no known practical application.
Materials simulation (D-Wave, March 2025)
D-Wave’s Advantage2 prototype performed quantum dynamics simulations of programmable spin glasses published in Science. The collaboration simulated a suite of lattice structures across multiple evolution times, modeling magnetic materials behavior in minutes that would take Frontier nearly one million years. This is notable as the first peer-reviewed quantum supremacy claim on a problem with direct scientific and industrial applications, using quantum annealingA quantum computing approach that finds optimal solutions to complex problems by using quantum tunneling to escape local minima, suited for optimization tasks. rather than gate-based computation.
Unconditional quantum information supremacy (UT Austin/Quantinuum, September 2025)
Researchers constructed a communication complexity task where 12 qubits sufficed but any classical protocol required at least 62 bits of memory. The key distinction: this separation is unconditional. Unlike RCS-based claims, which rest on computational complexity conjectures (the assumption that polynomial hierarchy does not collapse), the UT Austin result carries a mathematical proof that no classical algorithm, however clever, can close the gap. It directly demonstrates access to the exponential Hilbert space resource.
Verifiable quantum advantage (Google, October 2025)
Google’s Quantum Echoes algorithm computed out-of-order time correlators (OTOCs) on Willow, running 13,000 times faster than the best classical algorithm on a leading supercomputer. Unlike RCS, this algorithm models physical experiments and tests for both computational complexity and precision. The results were verified against Nuclear Magnetic Resonance data for molecules with up to 28 atoms, demonstrating a path toward quantum-enhanced molecular structure analysis.
The QEC code explosion and the path to fault tolerance
The 2025 landscape saw a dramatic expansion in QEC research. Riverlane’s QEC Report documents 120 new peer-reviewed QEC code papers between January and October 2025, up from 36 in 2024. All seven major QEC code families have now been implemented on hardware.
The shift from surface codes to quantum low-density parity-check (qLDPC) codes, initiated by IBM’s transition in 2024, is expected to proliferate across the industry in 2026. qLDPC codes offer better encoding rates (more logical qubits per physical qubit) and are critical for reducing the overhead required for fault-tolerant computation.
The practical implications are significant. One analysis combining improved QEC codes, higher qubit quality, and smarter classical co-processing estimated that breaking RSA-2048 might require approximately one million physical qubits, a 20x reduction from earlier estimates of 20 million. This compresses the timeline for cryptographic relevance considerably.
Hardware architectures: a diverging field
The absence of a dominant qubit technology is itself informative. Four fundamentally different approaches are being pursued at industrial scale:
Superconducting transmons (Google, IBM): The most mature platform. Google’s Willow features 105 qubits with T1 coherence times approaching 100 μs. IBM’s Heron achieves record two-qubit gate fidelities and can accurately run certain classes of circuits with up to 5,000 two-qubit gate operations. IBM’s roadmap targets its fault-tolerant Starling processor by 2029: 200 logical qubits running circuits of 100 million gates.
Topological qubits (Microsoft): Majorana 1 uses a novel topoconductor material (indium arsenide/aluminum heterostructures) cooled to near absolute zero to form topological superconducting nanowires with Majorana Zero Modes at the endpoints. Information is encoded in the parity of electrons shared between MZM pairs, providing inherent topological protection against local noise. Currently at eight qubits on a chip designed for one million. Measurement-based control via digital pulses rather than analog rotations. DARPA selected this approach for its final US2QC phase.
Trapped ions (Quantinuum, IonQ): UT Austin researchers demonstrated unconditional quantum information supremacy using Quantinuum’s H-series trapped ion processor. Oxford Ionics demonstrated 99.99% two-qubit gate fidelity in 2025. Trapped ion systems offer longer coherence times and all-to-all connectivity but face challenges in gate speed and scaling.
Quantum annealing (D-Wave): A fundamentally different computation model optimized for optimization and simulation problems. D-Wave’s Advantage2 prototype already operates at thousands of qubits with its supremacy result published in Science. The company offers commercial cloud access today via its Leap platform.
The hybrid future and the honest timeline
The consensus among industry and analysts is that quantum computing will augment, not replace, classical computation. Bain & Company’s 2025 Technology Report projects a market potential of up to $250 billion, but current revenue is under $1 billion annually. The projected near-term market of $5 billion to $15 billion by 2035 depends on early wins in simulation and optimization.
IBM’s vision of quantum-centric supercomputing, integrating QPUs with CPUs and GPUs in heterogeneous workflows, reflects where the field is heading. RIKEN and Cleveland Clinic are already running quantum-classical hybrid algorithms for electronic structure problems on IBM Quantum System One.
Several critical bottlenecks remain:
- Logical error rates: Current best (~10-3 per cycle) need to reach ~10-6 or below for most practical algorithms. This requires both better physical qubits and larger code distances.
- Decoder latency: Real-time decoding must keep pace with cycle times of ~1 μs for superconducting systems. Google demonstrated 63 μs average decoder latency at distance 5, but scaling to larger codes will be challenging.
- Correlated errors: Google’s repetition code experiments revealed rare correlated error events occurring approximately once per hour (every 3 × 109 cycles), setting an error floor of 10-10. The origins of these events are not yet understood.
- Talent: Riverlane estimates only 600-700 QEC specialists exist worldwide, against a need of 5,000-16,000 by 2030.
- Algorithm maturity: Over half of Bain’s projected $250 billion market value sits in quantum machine learning, which remains largely theoretical. The highest-value use cases (QML for LLMs, generative AI) are the most speculative.
Cryptographic implications
The “harvest now, decrypt later” threat is the most time-sensitive concern. Adversaries may be storing encrypted traffic today for future quantum decryption. NIST finalized its first post-quantum cryptographyEncryption methods designed to remain secure against quantum computers, which can break widely used algorithms like RSA by exploiting quantum properties. (PQC) standards in 2024, and migration is underway but slow. Bain’s survey found that 73% of IT security professionals expect quantum cryptographic threats to materialize within five years, but only 9% of tech leaders have a PQC migration roadmap.
The revised estimate that RSA-2048 might fall to approximately one million qubits (down from 20 million) compresses the threat timeline. While no one is close to a million fault-tolerant qubits today, the convergence of improving hardware, better QEC codes, and novel architectures like Microsoft’s topological approach means this is a moving target that warrants proactive planning.
Where we actually are
Quantum advantage has been demonstrated on benchmarks, on useful scientific problems, and with unconditional mathematical proof. Error correction works below threshold. Multiple viable hardware platforms are scaling. The first fault-tolerant prototypes are expected within two to three years.
But the machine that solves industrially relevant problems faster, cheaper, and more reliably than classical alternatives has not been built. The engineering path from 105 physical qubits to millions of fault-tolerant logical qubits is long and uncertain. Quantum computing in 2026 is in the position that classical computing occupied in the transistor era of the late 1950s: the fundamental principles are proven, the components work, and the engineering scaling challenge is the defining problem of the field.
