Guide

Quantum computing use cases: where quantum computers will help first, ranked by the evidence

Simulating quantum systems has the strongest case: molecules, materials and many-body physics, where qubits hold a quantum state that exact classical methods can only store at exponential cost, and where some physics experiments already run at the edge of what classical simulation can follow. Second is breaking RSA and elliptic-curve cryptography with Shor's algorithm, which no machine can do yet but which is why the move to post-quantum cryptography has already started. Optimisation, finance and machine learning come last: their speed-ups are mostly quadratic or unproven, and the overhead of error correction is large enough to cancel a quadratic one.

Updated

The short answer

First: simulating molecules and materials
The strongest case. Resource estimates for catalysts, enzymes and battery materials have fallen by orders of magnitude, and they still need thousands of error-corrected qubits[3]. An exponential advantage over the best classical chemistry methods is not established in general[8].
Closest today: many-body physics
Experiments on spin models already run where classical simulation struggles. Some were later matched classically[11], and at least one still stands[15].
Second: breaking public-key cryptography
Shor's algorithm factors in polynomial time, where the best known classical methods take super-polynomial time[19]. No machine can run it at a useful size yet, but standards bodies have set dates to retire RSA and elliptic-curve keys[23].
Last: optimisation, finance, machine learning
Speed-ups are mostly quadratic or unproven, and a quadratic one is unlikely to beat classical computers on early error-corrected hardware[24].
What to do now
Learn and test on simulators, check what today's hardware can run, and start a post-quantum cryptography inventory. All three are cheap and none depends on a forecast.

How to rank quantum computing use cases

A use case is worth ranking high when three things hold. The quantum algorithm is much faster than the best classical one, not just faster than a naive one. The input and the output are small, so that loading data and reading results do not eat the speed-up. And someone has worked out the full cost on error-corrected hardware, in qubits and run time, so the claim can be checked.

Hoefler, Häner and Troyer turned these into a rule of thumb: useful quantum advantage needs small-data problems and speed-ups better than quadratic, and by that rule chemistry and materials science stand out[25]. The ranking below follows the same logic, and for each area it names the best evidence and whether classical methods have since caught up.

1. Simulating molecules and materials

A molecule is a quantum system, and storing its exact electronic state on a classical computer takes memory that grows exponentially with its size. A quantum computer holds the same state in a number of qubits that grows with the number of orbitals. The algorithm that uses this is quantum phase estimation, which reads off an energy to a chosen precision. It needs error-corrected qubits, so the evidence so far is resource estimates: worked-out counts of qubits, gates and run time for a specific molecule. The variational method that runs on today's hardware is a different route, tested on our VQE page.

Nitrogen fixation (FeMoco). The iron-molybdenum cofactor of nitrogenase, the enzyme bacteria use to turn nitrogen from the air into ammonia, is the standard benchmark. The first full estimate, by Reiher and colleagues, needed 111 logical qubits[1] and about 1.0×10141.0 \times 10^{14} T gates[1] for one structure at the looser of its two accuracy targets, with run times that assumed fast logical gates later called optimistic[2]. Lee and colleagues cut the gate count by orders of magnitude with tensor hypercontraction: 2,142 logical qubits[3], or about four million physical qubits[3] for under four days[3], assuming a physical error rate of 0.1%[3]. Spectrum amplification then gave a further 4 to 195 times[6] speed-up on iron-sulfur complexes and a catalyst, and a denser surface-code layout brings FeMoco to 89k physical qubits in under a month[7] in a preprint from the same group.

Drug metabolism (cytochrome P450). Goings and colleagues ran the classical side and the quantum estimate on the same models of an enzyme family that metabolises many drugs. The largest model needs about 4.6 million physical qubits[4] and 73 hours[4] at a 0.1%[4] physical error rate. Their conclusion is careful: models large enough to capture both kinds of electron correlation have the potential to be a quantum advantage problem[4].

Catalysts and batteries. For a ruthenium catalyst that turns carbon dioxide into methanol, von Burg and colleagues estimated 101010^{10} to 101110^{11} Toffoli gates[2] and about 4,000 logical qubits[2] per energy, with 101010^{10} Toffoli gates taking about a day[2] at the faster of the gate times they considered. For a lithium-ion cathode material, Delgado and colleagues found 2,375 to 6,652 logical qubits[5] and run times from hours to months[5], and called the requirements daunting.

What this does not show. A survey of the evidence by Lee and colleagues found no proof of an exponential advantage across chemistry: for many molecules, either a good starting state for phase estimation is hard to prepare, or classical methods already handle the molecule well[8]. The case for chemistry rests on specific hard molecules, such as metal clusters with strongly correlated electrons, where a polynomial speed-up over the best classical method can still be large. That is why the most useful papers in this list run the best classical methods on the same molecule. The guide to quantum computing for chemistry goes through what has run on hardware so far and how each run compares.

2. Many-body physics: the closest to an advantage today

Simulating the dynamics of interacting spins and electrons is the one area where today's machines, without error correction, run experiments that classical computers find hard. Analog quantum simulators, built to mimic one kind of physics, were the first to reach this regime[9]. Digital machines have followed, and the record so far is mixed in an instructive way.

Matched classically. IBM ran a kicked Ising model on 127 qubits[10] with error mitigation and reported results beyond brute-force classical simulation. Within weeks, Tindall and colleagues reproduced it with a tensor network that follows the processor's lattice, more accurately than the hardware[11]. D-Wave's annealer simulated the quench dynamics of spin glasses and was claimed to be beyond classical reach[12]; a tensor network method then reached state-of-the-art accuracy on the same models with modest resources[13], and a variational Monte Carlo method matched the annealer's accuracy on up to 128 spins[14]. Tensor networks are the classical method to beat in this area.

Still standing. Google measured out-of-time-order correlators, which track how a disturbance spreads through a quantum system, on a 103-qubit[15] processor. Each circuit took 2.1 hours[15] of data collection, against an estimated 3.2 years[15] on the Frontier supercomputer with the best tensor network contraction found, a factor of about 13,000[15]. The authors tried a range of classical algorithms and none reproduced the measurement. The quantity is a physics observable, not yet an industrial answer, and the paper presents it as a path towards one.

Partly matched, partly open. Quantinuum simulated Ising dynamics on 56 qubits[16] and observed thermalization on timescales that, in the authors' words, severely challenge classical methods[16]. A Google team later reproduced the square-lattice runs with a matrix product state heuristic on a single A100 GPU[17], in close agreement with the hardware; the triangular-lattice runs have not been matched so far. Quantinuum's later preprint measured superconducting pairing correlations in Fermi-Hubbard models, at a scale the authors call likely difficult to reach classically[18], without a quantified classical comparison. It is the kind of claim classical groups test next, and the quantum simulation guide follows each of these replies.

3. Breaking public-key cryptography

Shor's algorithm factors integers and computes discrete logarithms in polynomial time[19], which breaks RSA, Diffie-Hellman and elliptic-curve cryptography. The best known classical methods take super-polynomial time, so this is the clearest large speed-up for a problem people care about, and also the use case nobody wants to see succeed. It needs a large error-corrected machine, and the estimates of how large keep falling. Gidney and Ekerå estimated that a 2048-bit RSA key could be factored in 8 hours[20] with 20 million[20] noisy qubits. Gidney later brought that to less than a million[21] noisy qubits in less than a week[21], under the same assumptions: a physical error rate of 0.1%[21] and a 1 microsecond[21] error-correction cycle.

The practical consequence is already here. NIST published its first three post-quantum standards in August 2024[22]: ML-KEM for key establishment (FIPS 203), and ML-DSA and SLH-DSA for signatures (FIPS 204 and 205), and told administrators to start using them immediately[22]. Its transition report, finalised in October 2026[23], deprecates RSA, ECDSA, EdDSA and Diffie-Hellman after 2030[23] and disallows them after 2035[23]. The reason to move before a quantum computer exists is harvest now, decrypt later: traffic recorded today can be decrypted once one does[23].

4. Optimisation, finance and machine learning: mostly quadratic or unproven

Most proposed business applications rest on one of two kinds of speed-up. Grover-type search and amplitude estimation are proven, but quadratic: N\sqrt{N} steps where a classical method needs NN (see Grover's algorithm and amplitude estimation vs Monte Carlo). Heuristics such as QAOA and quantum annealing have no proven speed-up at all, and on the instances a simulator can hold, exact classical solvers won in our runs (QAOA vs MILP).

A quadratic speed-up runs into the overhead of error correction. A logical gate on an error-corrected machine is many orders of magnitude slower than a classical operation[24], so the quantum computer only wins after a long run. Babbush and colleagues worked through an example of quantum-accelerated simulated annealing against a parallel classical machine: with a quadratic speed-up the break-even run time was 880 millennia[24], and with a quartic one 4.9 hours[24]. Their conclusion: quadratic speed-ups will not give an advantage on early error-corrected hardware unless error correction gets much faster[24]. For derivative pricing with amplitude estimation, the full estimate is 8k logical qubits[26] and a T-depth of 54 million[26], executed in about a second[26] to beat classical Monte Carlo.

There is new work on larger speed-ups for optimisation. Decoded quantum interferometry gives a super-polynomial speed-up over known classical algorithms for one structured problem, fitting polynomials over finite fields; on an unstructured test instance it beat general-purpose heuristics, but a tailored classical solver beat it[27]. Machine learning on classical data is in a similar place, and our guide on when quantum machine learning makes sense goes through the evidence.

Quantum computing use cases at a glance

Each row is a reading of published work as of November 2026. Hardware figures are estimates for error-corrected machines unless the row says otherwise, and they depend on the assumed error rate and gate speed.

ApplicationQuantum methodBest evidenceWhat it needsStatus today
Nitrogen-fixation catalyst (FeMoco)Phase estimationSuccessive resource estimates, each cheaper than the last[3][6]2,142 logical qubits[3], about four million physical qubits[3] for under four days[3]; 89k physical qubits in under a month[7] in a preprintNot run. Estimates only.
Drug metabolism (cytochrome P450)Phase estimationQuantum estimate next to classical DMRG and coupled-cluster runs on the same models[4]4.6 million physical qubits[4] and 73 hours[4] for the largest modelNot run. A potential advantage at large model sizes.
Catalysts, battery materialsPhase estimationResource estimates for a ruthenium catalyst[2] and a cathode material[5]4,000 logical qubits[2] per energy for the catalyst; 2,375 to 6,652 logical qubits[5] for the cathodeNot run. Estimates only.
Many-body dynamics (spin and Hubbard models)Analog or digital simulation, no error correctionCorrelator measurement about 13,000[15] times faster than the best classical estimateRuns today on 56[16] to 127[10] physical qubitsRunning. Some claims matched classically, at least one still stands.
Breaking 2048-bit RSAShor's algorithmResource estimates falling over time[20][21]less than a million[21] noisy qubits for less than a week[21]Not possible on today's hardware. Post-quantum standards in force.
Derivative pricing, riskAmplitude estimation (quadratic)Full resource estimate for real derivatives[26]8k logical qubits[26], run in about a second[26]Not run. Quadratic speed-up only.
Combinatorial optimisationQAOA, annealing, Grover-type search, decoded quantum interferometryQuadratic speed-ups unlikely to pay on early error-corrected hardware[24]; one structured problem with a larger speed-up[27]No credible estimate for an advantage on an industrial problemNo advantage shown. Classical solvers win in published tests.
Machine learning on classical dataVariational circuits, quantum kernelsProven speed-ups only on constructed problems; see the guide linked aboveUnknownNo advantage shown.

Most physical-qubit figures for chemistry and cryptography assume a physical error rate of 0.1%[21] and a surface-code cycle of 1 microsecond[21], as in the RSA estimate.

How far today's hardware is from these numbers

Every application in the first and third places needs error-corrected qubits, and error correction now works in the lab. Google ran a surface-code memory below the threshold where adding qubits lowers the logical error rate, on a 101-qubit[28] code with 0.143%[28] error per correction cycle[28]. A neutral-atom processor ran circuits on up to 48 logical qubits[29] protected by an error-detecting code[29]. The quantum error correction page explains what these milestones mean.

The estimates above ask for thousands of logical qubits running billions of gates. A cross-platform benchmark from Sandia and Quantinuum, still a preprint, translated those requirements into one measure of capability and found that today's machines have to grow by five orders of magnitude[30] to reach them[30]. The gap is large, and the estimates are shrinking from the other side: for FeMoco, from about four million physical qubits[3] to 89k[7] through better algorithms and codes alone. Watch both curves, hardware capability rising and resource estimates falling, rather than a date. The quantum advantage page covers how past claims were checked.

How to judge the next claim of quantum advantage

Five questions that sorted the claims above. A claim that answers all five well deserves attention.

  1. Faster than what?

    The comparison has to be with the best classical method for that problem, tuned, not with brute-force simulation of the circuit. Tensor networks and Monte Carlo methods are the usual challengers in physics; exact solvers and tuned heuristics in optimisation.

  2. Is the output useful, or only hard?

    Sampling from a random circuit is hard to simulate and useful to nobody. A correlator or an energy that a scientist wants is a step up, and an answer a company would pay for is the goal.

  3. Can it be checked?

    A result that another quantum computer, an experiment or a small classical case can verify is worth more than one that has to be taken on trust.

  4. Has a classical group tried to match it?

    Several claims fell within months to tensor networks or variational Monte Carlo[11][14]. Give a new claim time, and look for the reply.

  5. For a future advantage, is there a full resource estimate?

    Logical qubits, gate count, run time and the error rate assumed, for an instance of a size that matters. A speed-up in the exponent without these numbers is not yet a use case.

What you can do now

None of these needs a forecast of when large machines arrive, and each pays off on its own.

  1. Map your problems to quantum formulations

    Most business problems do not map onto a quantum algorithm with a large speed-up, and it is cheaper to find out early. EigenOrbit Adviser takes a plain-language description of a problem and flags which parts have a credible quantum formulation.

  2. Learn the methods on a simulator

    Build Hamiltonians, map them to qubits and solve small ones exactly and variationally, with the exact answer to check against. A simulator is free, exact and enough to learn where the methods break.

  3. Check what today's hardware can run

    Before booking QPU time, look up which devices could run a circuit of your size within their error budget. The Benchmarks link below opens on a Hamiltonian time-evolution circuit, the workload of many-body simulation.

    Run it on a real QPU with EigenOrbit RouterIn eigenkit: pip install "eigenkit[router]", then provider=RouterConfig(device=...) (docs).

  4. Start a post-quantum cryptography inventory

    List where your systems use RSA, elliptic-curve or Diffie-Hellman keys, and which data must stay confidential for years. That list decides the migration order, and the standards to migrate to are published[22].

Questions

What will quantum computers be used for?

The strongest case is simulating quantum systems: molecules for catalysts, enzymes and batteries, and materials and many-body physics. The second is breaking RSA and elliptic-curve cryptography with Shor's algorithm, which is why organisations are moving to post-quantum cryptography now. Optimisation, finance and machine learning are proposed often, but their speed-ups are mostly quadratic or unproven, so they rank last on current evidence.

When will quantum computers be useful?

For physics research, some machines are already useful: they run spin-model experiments that classical simulation struggles to follow. For chemistry and codebreaking, published estimates need thousands of error-corrected logical qubits, far beyond today's machines. The sources give resource estimates, not dates, so the honest way to track progress is hardware capability rising and resource estimates falling.

Has practical quantum advantage been demonstrated?

Not on an industrial problem. Several physics experiments have run beyond what classical simulation could do at the time; some were later matched by tensor network or Monte Carlo methods, and Google's measurement of out-of-time-order correlators still stands. These are scientific observables rather than answers a business would buy.

Will quantum computers break encryption?

A large error-corrected quantum computer running Shor's algorithm would break RSA, Diffie-Hellman and elliptic-curve cryptography. No such machine exists, but the estimated size keeps falling, and data recorded today could be decrypted later. NIST has published post-quantum standards and set dates to retire the vulnerable algorithms. Symmetric ciphers such as AES are not the target of this migration.

Is quantum computing useful for optimisation?

Not yet, and the case is weaker than for chemistry. Proven speed-ups for search are quadratic, which error-correction overhead is likely to cancel on early hardware, and heuristics like QAOA and annealing have no proven speed-up. In published comparisons, tuned classical solvers win on the instances today's machines and simulators can hold.

What should a company do about quantum computing now?

Three things that do not depend on a forecast: check which of your problems have a credible quantum formulation, let a small team learn the methods on simulators, and start a post-quantum cryptography inventory. Re-examine the first two when error-corrected machines reach the sizes the resource estimates ask for.

Sources

  1. M. Reiher, N. Wiebe, K. M. Svore, D. Wecker, M. Troyer. Elucidating reaction mechanisms on quantum computers. Proceedings of the National Academy of Sciences 114, 7555, 2017. arXiv:1605.03590.
  2. V. von Burg, G. H. Low, T. Häner, D. S. Steiger, M. Reiher, M. Roetteler, M. Troyer. Quantum computing enhanced computational catalysis. Physical Review Research 3, 033055, 2021. arXiv:2007.14460.
  3. J. Lee, D. W. Berry, C. Gidney, W. J. Huggins, J. R. McClean, N. Wiebe, R. Babbush. Even more efficient quantum computations of chemistry through tensor hypercontraction. PRX Quantum 2, 030305, 2021. arXiv:2011.03494.
  4. J. J. Goings, A. White, J. Lee, C. S. Tautermann, M. Degroote, C. Gidney, T. Shiozaki, R. Babbush, N. C. Rubin. Reliably assessing the electronic structure of cytochrome P450 on today's classical computers and tomorrow's quantum computers. Proceedings of the National Academy of Sciences 119, e2203533119, 2022. arXiv:2202.01244.
  5. A. Delgado, P. A. M. Casares, R. dos Reis, M. S. Zini, R. Campos, N. Cruz-Hernández, A.-C. Voigt, A. Lowe, S. Jahangiri, M. A. Martin-Delgado, J. E. Mueller, J. M. Arrazola. Simulating key properties of lithium-ion batteries with a fault-tolerant quantum computer. Physical Review A 106, 032428, 2022. arXiv:2204.11890.
  6. G. H. Low, R. King, D. W. Berry, Q. Han, A. E. DePrince, A. White, R. Babbush, R. D. Somma, N. C. Rubin. Fast quantum simulation of electronic structure by spectrum amplification. Physical Review X 15, 041016, 2025. arXiv:2502.15882.
  7. G. H. Low, W. J. Huggins, D. W. Berry, T. Khattar, A. F. White, N. C. Rubin, R. Babbush. A denser planar surface code. arXiv preprint, 2026. arXiv:2605.30455.
  8. S. Lee, J. Lee, H. Zhai, Y. Tong, A. M. Dalzell, A. Kumar, P. Helms, J. Gray, Z.-H. Cui, W. Liu, M. Kastoryano, R. Babbush, J. Preskill, D. R. Reichman, E. T. Campbell, E. F. Valeev, L. Lin, G. K.-L. Chan. Evaluating the evidence for exponential quantum advantage in ground-state quantum chemistry. Nature Communications 14, 1952, 2023. arXiv:2208.02199.
  9. A. J. Daley, I. Bloch, C. Kokail, S. Flannigan, N. Pearson, M. Troyer, P. Zoller. Practical quantum advantage in quantum simulation. Nature 607, 667, 2022.
  10. Y. Kim, A. Eddins, S. Anand, K. X. Wei, E. van den Berg, S. Rosenblatt, H. Nayfeh, Y. Wu, M. Zaletel, K. Temme, A. Kandala. Evidence for the utility of quantum computing before fault tolerance. Nature 618, 500, 2023.
  11. J. Tindall, M. Fishman, E. M. Stoudenmire, D. Sels. Efficient tensor network simulation of IBM's Eagle kicked Ising experiment. PRX Quantum 5, 010308, 2024. arXiv:2306.14887.
  12. A. D. King, A. Nocera, M. M. Rams, J. Dziarmaga, R. Wiersema, et al. Beyond-classical computation in quantum simulation. Science 388, 199, 2025. arXiv:2403.00910.
  13. J. Tindall, A. Mello, M. Fishman, E. M. Stoudenmire, D. Sels. Dynamics of disordered quantum systems with two- and three-dimensional tensor networks. Science 392, 868, 2026. arXiv:2503.05693.
  14. L. Mauron, G. Carleo. Challenging the quantum advantage frontier with large-scale classical simulations of annealing dynamics. arXiv preprint, 2025. arXiv:2503.08247.
  15. Google Quantum AI and collaborators (D. A. Abanin, R. Acharya, et al.). Observation of constructive interference at the edge of quantum ergodicity. Nature 646, 825, 2025. arXiv:2506.10191.
  16. R. Haghshenas, E. Chertkov, M. Mills, W. Kadow, S.-H. Lin, et al. Digital quantum magnetism on a trapped-ion quantum computer. Nature 653, 56, 2026. arXiv:2503.20870.
  17. S. Mandrà, B. Ware, N. Astrakhantsev, S. Isakov, B. Villalonga, T. Westerhout, K. Kechedzhi. A heuristic for matrix product state simulation of out-of-equilibrium dynamics of two-dimensional quantum spin systems. PRX Quantum 7, 033048, 2026. arXiv:2511.23438.
  18. E. Granet, S.-H. Lin, K. Hémery, R. Haghshenas, P. Andres-Martinez, et al. Superconducting pairing correlations on a trapped-ion quantum computer. arXiv preprint, 2025. arXiv:2511.02125.
  19. P. W. Shor. Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer. SIAM Journal on Computing 26, 1484, 1997. arXiv:quant-ph/9508027.
  20. C. Gidney, M. Ekerå. How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits. Quantum 5, 433, 2021. arXiv:1905.09749.
  21. C. Gidney. How to factor 2048 bit RSA integers with less than a million noisy qubits. arXiv preprint, 2025. arXiv:2505.15917.
  22. National Institute of Standards and Technology. NIST releases first 3 finalized post-quantum encryption standards (FIPS 203, FIPS 204, FIPS 205). NIST news release, 2024.
  23. D. Moody, R. Perlner, A. Regenscheid, A. Robinson, D. Cooper. Transition to post-quantum cryptography standards (NIST IR 8547). NIST Internal Report, 2026.
  24. R. Babbush, J. R. McClean, M. Newman, C. Gidney, S. Boixo, H. Neven. Focus beyond quadratic speedups for error-corrected quantum advantage. PRX Quantum 2, 010103, 2021. arXiv:2011.04149.
  25. T. Hoefler, T. Häner, M. Troyer. Disentangling hype from practicality: on realistically achieving quantum advantage. Communications of the ACM 66, 82, 2023. arXiv:2307.00523.
  26. S. Chakrabarti, R. Krishnakumar, G. Mazzola, N. Stamatopoulos, S. Woerner, W. J. Zeng. A threshold for quantum advantage in derivative pricing. Quantum 5, 463, 2021. arXiv:2012.03819.
  27. S. P. Jordan, N. Shutty, M. Wootters, A. Zalcman, A. Schmidhuber, R. King, S. V. Isakov, T. Khattar, R. Babbush. Optimization by decoded quantum interferometry. Nature 646, 831, 2025. arXiv:2408.08292.
  28. Google Quantum AI and collaborators (R. Acharya, L. Aghababaie-Beni, I. Aleiner, et al.). Quantum error correction below the surface code threshold. Nature 638, 920, 2025. arXiv:2408.13687.
  29. D. Bluvstein, S. J. Evered, A. A. Geim, S. H. Li, H. Zhou, et al. Logical quantum processor based on reconfigurable atom arrays. Nature 626, 58, 2024. arXiv:2312.03982.
  30. T. Proctor, O. Hart, O. W. Maupin, M. Girling, D. Hothem, et al. Benchmarking the computational power of quantum computers. arXiv preprint, 2026. arXiv:2609.12146.

Run it yourself with eigenkit

eigenkit is EigenOrbit's open-source Python library: write a circuit once, run it on any simulator or QPU, and test it against a classical baseline on your own data before you invest in it or rule it out.

$ pip install eigenkit