Quantum Computing: what is actually working now?
In our quantum computing market deck, you will find everything you need to understand the market
SUMMARY
Quantum Computing: what is actually working now? Quantum computing is already working as a specialized scientific technology: error correction is improving logical information, researchers can run increasingly deep logical circuits, and selected quantum workloads now sit beyond practical classical reproduction. Broad commercial advantage, however, is still rare.
The biggest change is that several old objections are being attacked at once. Better logical qubits, deeper encoded computation and more credible verification methods are arriving together rather than as isolated chip announcements.
Error correction is the strongest technical proof. Google has shown a surface-code logical memory that improves as the code grows, while trapped-ion and neutral-atom teams are reaching similar goals with very different hardware and lower-overhead codes.
Quantum advantage claims are also becoming harder to dismiss. The better recent experiments focus less on raw speed headlines and more on whether the computation is genuinely hard for classical methods and whether the quantum output can still be trusted.
Chemistry and materials research look like the most natural scientific use cases because the underlying systems are quantum mechanical. Even there, the practical model is increasingly hybrid: classical computers handle most of the workflow and quantum processors are used only for the part that becomes especially difficult classically.
Optimization already has some real production deployments, especially through D-Wave, but that evidence should not be confused with proof that quantum systems broadly beat the best classical solvers. The commercial base remains small and classical optimization keeps improving quickly.
Quantum machine learning is much less convincing today. Theoretical advantages exist, but loading classical data into quantum states and competing with rapidly improving GPUs leave little reason for normal AI teams to switch.
Cryptography is almost the reverse case: quantum computers cannot break RSA-2048 today, yet the risk already affects real security planning because encrypted data collected now could remain valuable long enough to be attacked later.
No hardware architecture is clearly winning. Superconducting systems lead in industrial depth and surface-code progress, trapped ions in fidelity, neutral atoms in flexible large arrays, annealers in production optimization, while topological hardware still carries unusually large scientific uncertainty.
Commercial traction is real, but revenue and contracts mostly show that customers and governments are willing to pay for access, hardware and strategic capability. They do not yet show a broad migration of ordinary computing budgets toward quantum because of consistently better economics.
The real race has shifted from proving individual quantum effects to scaling them. Today’s best systems can manipulate tens of logical qubits and thousands of meaningful logical operations; useful large-scale fault tolerance will require those pieces to work together across millions or billions of reliable operations.
This market map, featured in our quantum computing market deck, highlights top companies and startups in the quantum computing market
Why does quantum computing feel more real now?
Quantum computing feels more real today because error correction, logical computation and beyond-classical experiments have all moved forward at the same time.
For years, progress was easy to dismiss as another larger chip or another laboratory benchmark. The recent pattern is harder to wave away.
Google’s Willow work showed below-threshold quantum error correction: as researchers increased the distance of a surface code, the logical error rate fell rather than rising with the extra hardware. The distance-7 experiment used 101 physical qubits and reached a logical error rate of about 0.143% per correction cycle. Its logical memory also lasted about 2.4 times longer than the best physical qubit used to build it. Google reported those results in Nature.
Neutral atoms have reached a different milestone. Harvard, QuEra and collaborators first ran error-corrected algorithms with as many as 48 logical qubits, then pushed experiments to arrays containing hundreds of atoms while demonstrating repeated error correction, logical entanglement and lattice surgery.
IBM and the University of Chicago added an especially important result this year. Their researchers encoded 70 logical qubits, ran 2,415 logical two-qubit operations and 468 logical T gates, and completed the computation in roughly 15 minutes. The team also developed statistical checks on the fidelity of a calculation that leading classical simulation methods could no longer handle practically.
Together, these results attack three weaknesses that used to hang over the field: qubits were too noisy, error correction consumed enormous resources, and the most impressive quantum computations were difficult to verify.
There is still a huge distance between those experiments and a machine that transforms banking, chemistry or logistics. But the argument that quantum computing remains little more than fragile physics experiments has become much harder to make.
What does “quantum computing works” actually mean today?
Quantum computing currently works as a scientific computing technology far more convincingly than it works as an everyday commercial computing technology.
A lot of confusion comes from putting several very different achievements under the same word.
One standard is basic control: can researchers prepare qubits, entangle them and run complex circuits before noise destroys the calculation? Several platforms can clearly do that now.
A tougher standard is error correction. Can many imperfect physical qubits create logical information that behaves better than its components? Google, Quantinuum, QuEra and other teams have now demonstrated important parts of that process.
Then comes computational advantage. Can quantum hardware complete a calculation that serious classical methods struggle to reproduce? The answer is increasingly yes for carefully chosen sampling and quantum-physics problems. IBM’s recent 70-logical-qubit experiment is particularly useful here because the researchers addressed both hardness and verification instead of relying only on the inability to simulate the circuit classically.
The hardest test is economic usefulness. Can an organization take a problem it genuinely needs solved and get a better result, lower cost or faster answer by adding a quantum computer?
That final category remains thin.
| What are we testing? | Where quantum computing stands now | Best evidence |
|---|---|---|
| Can quantum states be controlled reliably? | Clearly yes | High-fidelity systems running increasingly deep circuits |
| Can error correction improve quantum information? | Yes at experimental scale | Below-threshold memories and logical operations |
| Can quantum hardware outrun classical methods? | Yes on selected problems | IBM, Google and D-Wave experiments |
| Can ordinary companies routinely get better economics? | Only in narrow cases | Some annealing and hybrid production workloads |
If you want more recent data on this point, please see our latest quantum computing market report.
As this chart shows, and as featured in our quantum computing market deck, search interest in quantum computing has grown significantly
Is quantum error correction actually working now?
Quantum error correction is working experimentally today, and this is probably the most important technical change in the entire field.
The old problem was brutal. Protecting one useful qubit required many extra physical qubits, yet every added qubit introduced another place where something could go wrong. If error rates stayed above the correction threshold, adding more hardware simply built a larger unreliable machine.
Google’s Willow experiment finally showed the relationship moving in the desired direction. Increasing the surface-code distance progressively reduced the logical error rate, with roughly a factor-of-two improvement each time researchers increased the code distance by two. The largest logical memory ran on 101 physical qubits and survived substantially longer than its strongest individual component qubit.
That experiment also involved real-time classical decoding. The processor repeatedly measured errors while classical hardware interpreted them fast enough to keep the logical state alive. Google ran some tests for as many as one million correction cycles, so the result went well beyond a short demonstration that happened to survive.
Other architectures are getting there differently.
Quantinuum’s Helios has 98 trapped-ion qubits with reported average two-qubit gate fidelity around 99.921% and single-qubit fidelity around 99.9975%. Those unusually clean physical qubits allow much lighter error-correcting codes than surface-code machines require. Quantinuum has demonstrated logical encodings involving dozens of logical qubits and fault-tolerant gate operations.
Neutral atoms offer another route. QuEra and academic teams can rearrange atoms physically, build logical blocks dynamically and perform operations such as lattice surgery between them. Their experiments have involved tens of logical qubits and hundreds of atoms.
Superconducting qubits, trapped ions and neutral atoms now all have credible error-correction results despite using very different physics.
The remaining problem is scale. A useful fault-tolerant algorithm may need millions or billions of logical operations. Current experiments are impressive precisely because they have reached thousands of meaningful logical operations. Those two scales remain far apart.
Can quantum computers actually beat classical computers now?
Quantum computers can beat leading classical methods on some problems today, although the set of problems remains narrow.
The strongest recent experiments have become harder to dismiss because researchers are improving both the quantum calculation and the way they prove that the answer deserves to be trusted.
Google’s early random-circuit experiments showed that a quantum processor could generate samples that were extremely expensive to reproduce classically. Classical researchers later found much better simulation techniques and cut the original gap dramatically. That episode was useful: every serious quantum advantage claim now has to assume that classical algorithms will improve after publication.
Google later moved toward calculations with clearer physical meaning. Its Quantum Echoes work measured quantum dynamics through out-of-time-order correlations and reported a roughly 13,000-fold advantage over the classical approaches used in its comparison. Researchers subsequently tested stronger tensor-network simulations and still found the largest instances extremely difficult.
IBM and the University of Chicago pushed the argument further this year. Their 70-logical-qubit experiment ran 2,415 logical two-qubit gates and 468 T gates in about 15 minutes. The researchers showed that several leading classical simulation methods would require impractical resources, while the encoded quantum circuit had logical error rates about ten times lower than the underlying physical error rates.
Verification is the interesting part. If a classical computer cannot calculate the full answer, we lose the easiest way to check the quantum machine. The IBM–Chicago team instead built statistical bounds around the fidelity of the computation. That makes the result considerably more useful than saying, “our classical simulator could not keep up.”
D-Wave has also published a beyond-classical result using quantum annealing to simulate magnetic materials. The original comparison produced an enormous claimed speed gap, then classical researchers attacked the benchmark with better algorithms. That back-and-forth is exactly what we should expect.
We can now say with reasonable confidence that beyond-classical quantum computation exists. These experiments still do not show a general advantage for the workloads sitting inside normal corporate data centers.
If you want more recent data on this point, please see our latest quantum computing market report.
This chart, included in our quantum computing market deck, illustrates yearly VC funding for quantum computing startups
Are quantum computers solving useful real-world problems yet, especially in chemistry?
Quantum computers are already doing useful scientific work, especially in chemistry and materials, while clear end-to-end business advantage remains rare.
Quantum systems are naturally promising for molecules and materials because the underlying physics is quantum mechanical. Classical computers can approximate these systems extremely well in many cases, but exact simulation becomes painfully expensive as interactions grow.
This is where some current experiments are becoming genuinely interesting.
IBM and Qedma have used quantum hardware to study quantum-material dynamics at sizes where several strong classical techniques began disagreeing with each other. The point was less about producing a flashy runtime comparison and more about entering a regime where classical approximations stop giving one obvious answer.
IBM and RIKEN have also connected an IBM quantum computer directly to Japan’s Fugaku supercomputer. Their chemistry workflow moved back and forth automatically between classical and quantum calculations. Researchers from Oak Ridge National Laboratory, Cleveland Clinic and IBM have used similar approaches for molecular calculations linked to materials that could matter in fusion reactors.
Google’s Quantum Echoes experiment goes in the same direction. The technique can probe Hamiltonians and molecular structure rather than simply generate random samples.
Chemistry has a particularly strong case because molecules contain interacting electrons whose full quantum states become enormously expensive to represent classically. A sufficiently capable quantum computer could model those states more directly.
Researchers are already getting better at combining quantum processors with classical chemistry methods, so the quantum machine handles only the part that classical computation finds hardest. That approach is probably more realistic than waiting for one giant quantum computer to calculate an entire industrial molecule from scratch.
Classical chemistry remains a ferocious competitor. Density-functional theory, coupled-cluster techniques, tensor networks, GPUs and machine-learned interatomic potentials continue to improve. A quantum computer reaching a particular molecular size several years from now will compete against the classical software available then.
The real benchmark is an industrial decision. A pharmaceutical company would want a quantum calculation that changes which molecule it develops. A battery company would want a material that classical methods failed to identify. A chemical company would want fewer physical experiments or a cheaper development cycle.
Public evidence of those outcomes is still scarce.
Is quantum optimization actually useful for companies today?
Quantum optimization has some of the clearest production use cases today, mainly through D-Wave’s annealing systems, but we still lack evidence of a broad advantage over the best classical optimization software.
D-Wave deserves to be treated separately from most gate-model companies. Its Advantage2 system has more than 4,400 annealing qubits and specializes in optimization and energy-minimization problems. Those qubits cannot run arbitrary quantum circuits, so comparing the machine directly with IBM, Google or Quantinuum by qubit count makes little sense.
The commercial evidence is more interesting.
In its latest half-year results, D-Wave said 37.3% of its quantum-computing-as-a-service revenue came from production applications, up from 9.8% during the comparable period a year earlier. Production QCaaS revenue reached $1.3 million, commercial customers generated roughly two-thirds of total revenue, and first-half bookings rose from $2.9 million to $35.5 million.
AT&T offers one concrete example. According to D-Wave’s latest investor presentation, an early network-optimization application cut processing time from roughly one hour to under 15 seconds. The companies are now looking at outage response, technician routing, network planning and traffic management.
Those numbers deserve context. D-Wave reported only $5.9 million of total revenue in the first half of the year because the comparable period included a large hardware sale. We are therefore looking at a small commercial base, even if bookings and production usage are moving sharply upward.
There is another catch. Optimization is unusually difficult to benchmark. Classical solvers combine mathematical programming, heuristics, GPUs, decomposition and years of domain tuning. A quantum system can beat one implementation and lose to a stronger classical method six months later.
D-Wave has already done something many gate-model companies still cannot show: real customers are running quantum workloads in production. Broad superiority over the best classical optimizers is still unproven.
If you want more recent data on this point, please see our latest quantum computing market report.
This chart, included in our quantum computing market deck, looks at IonQ’s strategy in quantum computing
Is quantum machine learning useful yet?
Quantum machine learning currently has interesting theory and experiments, but normal AI workloads still belong on classical accelerators.
The gap between the academic promise and the practical case is especially wide here.
Researchers can construct learning problems where quantum algorithms have provable advantages. Recent theoretical work has expanded the types of supervised-learning problems where those advantages could appear. That tells us something important about what quantum computers might eventually do.
It tells companies very little about whether they should move their fraud model, recommendation system or language model away from GPUs today.
Classical data creates one obvious bottleneck. Images, financial transactions, text and sensor records begin as ordinary bits. A quantum algorithm may gain speed inside the calculation while losing much of that advantage when enormous classical datasets have to be encoded into quantum states.
The classical baseline is also moving absurdly fast. GPUs get faster, AI algorithms change, model compression improves and specialized chips keep appearing. Quantum machine learning therefore has to chase one of the fastest-improving targets in computing.
More promising applications may emerge when the data itself comes from a quantum system. A processor studying a molecule or material could potentially feed quantum information into another quantum routine without reconstructing the entire state classically.
For conventional machine learning, however, we found no current case strong enough to justify replacing a good GPU-based workflow with a quantum one.
Can quantum computers break RSA encryption today?
Quantum computers cannot break modern RSA or elliptic-curve encryption today, and current machines remain far below the scale needed for that job.
Shor’s algorithm has been known for decades. The uncertainty lies in the hardware required to run a cryptographically meaningful version of it.
Small quantum machines can demonstrate parts of Shor’s algorithm and factor tiny numbers. Breaking an RSA-2048 key is an entirely different engineering problem. It would require fault-tolerant logical qubits running enormous numbers of reliable operations, alongside the infrastructure needed for continuous error correction.
Compare that with the machines researchers are building now. Even IBM’s planned Starling system targets roughly 200 logical qubits and 100 million logical gates. That would be a dramatic jump from today’s hardware and still would not amount to an RSA-breaking machine.
NIST consequently treats the arrival date of a cryptographically relevant quantum computer as uncertain rather than imminent. Estimates span many years, and nobody can point to a reliable countdown.
Cybersecurity teams are moving anyway because encrypted information can remain valuable for a long time. An attacker can capture sensitive traffic now, store it and attempt to decrypt it later if sufficiently capable quantum hardware arrives. That “harvest now, decrypt later” problem is one reason NIST finalized its first post-quantum cryptography standards in 2024 and has urged organizations to start migrating.
Quantum computers therefore create a real security problem today through future risk, even though the machine capable of breaking mainstream public-key encryption does not exist.
This chart, included in our quantum computing market deck, illustrates yearly funding for quantum computing startups
Which quantum computer technology is actually winning right now?
No quantum architecture is clearly winning right now because superconducting qubits, trapped ions, neutral atoms and annealers each lead on completely different things.
Superconducting systems have the deepest industrial ecosystem. Google has produced the strongest below-threshold surface-code result, while IBM is building around modular processors, error-corrected circuits and integration with classical supercomputers. Gates are fast and semiconductor manufacturing techniques help with scaling. Noise and cryogenic complexity remain painful.
Trapped ions currently look exceptional on fidelity. Quantinuum’s Helios has 98 fully connected physical qubits, reported single-qubit fidelity of 99.9975% and two-qubit fidelity of 99.921%. High physical accuracy lets Quantinuum build useful logical encodings with much less overhead than many superconducting approaches.
Neutral atoms are scaling raw atom counts very quickly. QuEra and academic collaborators have already used large reconfigurable arrays for logical qubits, error correction, entanglement and lattice surgery. The ability to move atoms around gives this architecture unusual flexibility for creating logical structures.
D-Wave leads raw qubit count with more than 4,400 qubits in Advantage2, although annealing solves a narrower class of problems. Its more meaningful advantage today is commercial maturity in optimization rather than the headline qubit figure.
Microsoft sits in a stranger position. Its topological approach could dramatically lower error-correction overhead if the underlying physics works as intended. Majorana 2 has kept that program alive, but skepticism remains unusually high. Nature reported this year that researchers still dispute whether Microsoft has conclusively demonstrated the topological states it needs, and a peer-reviewed critique argued that conventional quantum-dot behavior can mimic important experimental signatures.
This also explains why raw qubit counts tell us so little. A 4,400-qubit annealer, a 98-qubit trapped-ion processor and a surface-code machine using roughly 100 physical qubits to protect one strong logical memory cannot sensibly be ranked by one number.
| Architecture | What looks strongest today | What could still stop it |
|---|---|---|
| Superconducting | Fast gates, mature engineering, strong surface-code results | Error-correction overhead and cryogenics |
| Trapped ions | Extremely high fidelity and full connectivity | Slower gates and scaling the hardware |
| Neutral atoms | Large reconfigurable arrays and rapid logical progress | Gate fidelity and system engineering |
| Quantum annealing | Thousands of qubits and real optimization deployments | Narrower programmability and contested general speedups |
| Topological | Potentially much lower correction overhead | Fundamental experimental evidence still disputed |
If you want more recent data on this point, please see our latest quantum computing market report.
Are companies actually paying for quantum computing now?
Companies are paying real money for quantum technology today, although current revenue says more about early adoption and hardware deployment than about widespread quantum economic advantage.
IonQ has become the clearest example of how quickly quantum-company revenue can grow. The company reported $80.1 million of second-quarter revenue this year, up 287% from the same quarter a year earlier. First-half revenue reached roughly $145 million, and IonQ raised its full-year revenue guidance to between $280 million and $290 million.
That is substantial for the quantum sector. We should still be careful with what sits inside the number. IonQ has expanded well beyond access to trapped-ion computers through acquisitions and hardware deployment. Around a quarter of its latest quarterly revenue came from its broader multi-product business. Revenue growth therefore cannot be read as a direct measure of customers earning returns from quantum algorithms.
D-Wave gives us a cleaner look at actual quantum usage. As discussed earlier, 37.3% of its QCaaS revenue in the latest half-year came from production applications. At the same time, total first-half revenue was only $5.9 million. That combination is revealing: genuine production usage exists, but the market remains small.
Quantinuum provides another useful comparison. Its 2025 revenue was around $31 million while research-and-development spending was well above $150 million. This is still an industry where building the machines costs far more than customers currently pay to use them.
Government funding adds another layer. The United States has continued directing large amounts of strategic capital toward quantum hardware, including recent CHIPS-related support for D-Wave, Rigetti and Quantinuum. Governments are buying future technological capacity alongside immediate computing output.
Commercial customers, governments and large companies now sign meaningful contracts. What we cannot yet see is conventional computing budgets shifting toward quantum at scale because quantum systems consistently offer better economics.
| Commercial evidence | What it shows | What it does not prove |
|---|---|---|
| IonQ: $80.1M latest quarterly revenue | Quantum companies can now build sizeable commercial businesses | That the revenue comes mainly from quantum-compute advantage |
| D-Wave: 37.3% of QCaaS revenue from production use | Some quantum workloads have entered normal operations | Broad superiority over classical optimizers |
| Quantinuum: about $31M 2025 revenue | Customers will pay for leading quantum capability | Near-term profitability of quantum hardware |
| Large government programs | Quantum is strategically important | Near-term private-sector ROI |
This chart, included in our quantum computing market deck, compares the main business model options for quantum computing hardware startups
Are hybrid quantum-classical systems already the practical model?
Hybrid quantum-classical systems are the most realistic form of useful quantum computing today, and current hardware is increasingly being built around that assumption.
Almost every serious quantum workflow needs classical computing somewhere in the loop.
Google’s error-correction experiment depends on classical decoders interpreting streams of syndrome measurements while the quantum processor continues operating. Without that classical layer, the protected logical state eventually collapses under accumulated errors.
IBM and RIKEN have taken the same idea to supercomputer scale by directly connecting an IBM quantum system with Fugaku. A chemistry calculation can move between the two machines inside one workflow, allowing each processor to handle the part of the problem where it makes sense.
Quantinuum’s Helios integrates NVIDIA GPUs into its control stack as well. The classical hardware can deal with compilation, decoding and other processing without treating the quantum machine as an isolated remote experiment.
Quantum processors increasingly resemble specialist accelerators. CPUs and GPUs still do most of the work, while quantum hardware gets called for the unusually hard subproblem where its physics provides leverage.
The unresolved part is economic. Hybrid computing already works mechanically; we still need more cases where adding the quantum accelerator improves the total result enough to justify the extra complexity and cost.
How far are today’s quantum computers from truly fault-tolerant machines?
Today’s quantum computers have demonstrated many of the ingredients of fault tolerance, but useful large-scale fault-tolerant computing still requires a jump of several orders of magnitude.
IBM’s roadmap makes the gap unusually easy to see.
Current machines can run important circuits involving thousands of gates. IBM’s planned Starling system targets 200 logical qubits capable of 100 million logical operations. Its later Blue Jay architecture is aimed at 2,000 logical qubits and one billion operations.
Moving from thousands of useful operations toward 100 million means adding roughly four to five orders of magnitude of circuit depth.
The hardware count is only part of that challenge. Logical gates have to remain reliable across long computations. Error decoders must run continuously. Machines need ways to create expensive logical resources such as magic states. Separate quantum modules need to communicate. Cryogenic systems, lasers, vacuum equipment, control electronics and software all have to work together for hours rather than for one carefully prepared experiment.
Google has demonstrated below-threshold surface codes and fast real-time decoding. QuEra and its collaborators have demonstrated logical entanglement, repeated correction and lattice surgery. Quantinuum has shown how high physical fidelity can sharply lower encoding overhead. IBM and the University of Chicago have now pushed encoded computation to 70 logical qubits and thousands of logical operations.
Researchers now have credible pieces of a fault-tolerant machine. The difficult part is scaling those pieces together by several orders of magnitude.
This chart, featured in our quantum computing market deck, illustrates how revenue is divided among customer segments in the quantum computing market
So what is actually working in quantum computing now?
Quantum computing is genuinely working now as a specialized scientific technology, while broad commercial quantum computing remains premature.
The evidence is much stronger than a few impressive chip announcements.
Quantum error correction can now improve logical information as codes become larger. Researchers can manipulate tens of logical qubits, execute thousands of logical operations and, in IBM’s latest experiment, run a trusted calculation that leading classical simulation methods cannot practically reproduce. Quantum processors are also contributing to serious materials and chemistry research.
Commercial activity has become real as well. D-Wave has production optimization workloads. IonQ is generating tens of millions of dollars each quarter from its expanding quantum platform. Quantinuum sells access to one of the highest-fidelity gate-model machines available. Quantum computers are being integrated directly with supercomputers and GPUs instead of sitting alone in physics laboratories.
The limits are equally clear.
No gate-model quantum computer currently offers a routine economic advantage across mainstream corporate workloads. Quantum chemistry has promising hardware experiments without a clear industrial breakthrough yet. Quantum optimization has real deployments, although the best classical solvers remain difficult competitors. Quantum machine learning has little reason to displace GPUs today. RSA-breaking hardware remains far away. Microsoft’s topological approach still has fundamental experimental questions hanging over it.
The biggest change is where the uncertainty sits.
A few years ago, researchers were still fighting to prove that error correction could improve as a machine grew. That barrier has now been crossed experimentally on several fronts. The hard question these days is whether quantum systems can jump from tens of useful logical qubits and thousands of meaningful operations to machines capable of millions or billions of reliable operations.
Our judgment is fairly sharp: quantum computing has escaped the category of speculative physics, but it has yet to become a broadly useful computing platform.
What works today is the hard scientific core—qubit control, increasingly credible error correction, logical computation, selected beyond-classical workloads, quantum simulation and a small group of commercial optimization applications. The general-purpose economic revolution still depends on scaling those successes by orders of magnitude.
That is where the quantum computing race actually stands now.
If you want more recent data on this point, please see our latest quantum computing market report.
OUR METHODOLOGY
This analysis tests what is actually working in quantum computing today by separating technical reliability, quantum error correction, beyond-classical computation, scientific usefulness, commercial deployment, cryptographic implications, competing hardware architectures and the distance still left to large-scale fault tolerance.
We kept different kinds of evidence separate. Physical-qubit count, logical-qubit performance, gate fidelity, computational advantage, production use and company revenue answer different questions, and architectures such as superconducting qubits, trapped ions, neutral atoms and quantum annealers cannot be ranked meaningfully by one headline number.
For technical claims, we prioritized peer-reviewed experiments and first-hand research accounts. The core evidence includes Google’s Willow surface-code work in Nature, the Harvard/QuEra/MIT/NIST logical-processor experiments, later neutral-atom fault-tolerance results, IBM and the University of Chicago’s 70-logical-qubit computation, Google’s Quantum Echoes work, and Quantinuum’s published Helios specifications and logical-computing results.
Claims of quantum advantage were treated more cautiously than raw benchmark claims. We looked at the classical methods used for comparison, how those methods improved after publication, and whether researchers could still establish confidence in the quantum output once direct classical reproduction became impractical.
For chemistry and hybrid computing, we used IBM and Qedma’s quantum-material work and IBM and RIKEN’s integration of quantum hardware with Fugaku. These examples are useful because they show quantum processors operating inside larger classical workflows rather than as isolated laboratory machines.
For commercial evidence, we separated market traction from computational advantage. D-Wave’s SEC-filed results were used for revenue, bookings and production QCaaS activity; IonQ’s investor results were used for its latest revenue growth and business mix. Revenue and contracts show that customers are paying for quantum technology, but they do not by themselves prove that quantum algorithms already beat strong classical alternatives economically.
For cryptography, we relied on NIST’s post-quantum cryptography material and migration guidance. IBM’s published fault-tolerance roadmap was used to compare current logical-computing experiments with the much larger scale targeted by Starling and Blue Jay.
Key sources include: Nature on Google Willow and below-threshold error correction, Nature on the 48-logical-qubit neutral-atom processor, Nature on later neutral-atom fault-tolerance work, IBM on the 70-logical-qubit trusted-computation experiment, University of Chicago on the same experiment and its verification work, Nature on Quantum Echoes, Quantinuum on Helios, D-Wave’s SEC-filed first-half commercial results, IonQ’s second-quarter results, NIST’s post-quantum cryptography guidance, IBM’s fault-tolerance roadmap, and Nature’s coverage of Microsoft’s disputed topological evidence.
This chart, included in our quantum computing market deck, shows how cloud quantum computing access technology has evolved over time
Related blog posts
- Quantum Computing: what are the biggest challenges now?
- Quantum Computing: what are the biggest unsolved problems?
- Quantum Computing: what is getting real adoption now?
- What are the latest funding developments in quantum computing?
