So What Are Quantum Computers Even For?
Quantum computers are being built primarily to simulate quantum systems, which is why the US Department of Energy names materials science, quantum chemistry and high-energy physics as the fields the first reliable machines are expected to change. That list is short on purpose, because chemistry is quantum mechanics and a machine that obeys the same rules as the thing it models is the natural instrument for modeling it.
Separately, and much less discussed, a different branch of the same physics is already commercial. Quantum sensing ships today in atomic clocks, gravimeters and magnetometers, and one of those has been in orbit doing a job for you every time your phone finds itself on a map.
As of 2026 no quantum computer has outperformed a classical computer on a problem somebody genuinely needed solved, which is the sentence most vendor material is written to avoid.
Source: Testimony of Dr Tanner Crowder, Office of Science, US Department of Energy, before the House Committee on Science, Space, and Technology, 22 January 2026, energy.gov; R. P. Feynman, “Simulating Physics with Computers,” International Journal of Theoretical Physics 21, 467–488 (1982), doi.org/10.1007/BF02650179.
The short version:
- Quantum technology is not one technology on one timeline. Sensing is commercial now, computing is at an early and expensive stage, and treating them as the same field is what makes people believe the whole thing is either finished or fake.
- The DOE names three target areas for the first fault-tolerant machines: materials science, quantum chemistry, and high-energy physics. All three are problems about simulating quantum systems.
- Quantum sensing already works and already ships. Atomic clocks and atom interferometers are, in the DOE’s own words, “being commercialized and providing technological solutions today.”
- No quantum computer has yet beaten a classical one at a job anyone needed done. The 2026 IBM–UChicago verified-advantage result is a real milestone on a sampling benchmark, and a sampling benchmark is not a useful problem.
- The heavily marketed applications sit outside the DOE’s list. Portfolio optimization, logistics, quantum machine learning and “quantum AI” have no demonstrated advantage on a real problem, and classical methods in those fields are strong and improving.
An everyday way to picture it
A wind tunnel is the closest ancestor here. You could try to predict how air moves over a wing by writing equations and grinding through them, and for simple shapes that works. For a real wing it stops working, so instead of computing the airflow you build a box that has airflow in it, put the wing inside, and read the answer off the physical system. A quantum computer is that box for quantum mechanics. Molecules, electrons and magnetic materials behave according to rules a classical computer has to approximate at enormous cost, and a quantum computer is a machine that obeys those same rules natively. That is why the target list is chemistry and materials rather than spreadsheets and delivery routes.
What can quantum technology actually do today?
The part of quantum technology that is commercially deployed right now is sensing, and it is close to absent from mainstream coverage of the field.
Quantum sensing works because quantum systems are extraordinarily sensitive to their surroundings. In a computer that sensitivity is the central engineering problem, which is why those machines sit inside refrigerators colder than deep space. In a sensor, that same sensitivity is the product. The DOE’s description is precise: quantum sensing “harnesses the exquisite sensitivity of quantum systems to make precise measurements of physical quantities, such as magnetic fields, temperature, acceleration, and gravity, that surpass classical limits.”
Three families are worth knowing by name:
| Sensor | What it measures | Where it goes |
|---|---|---|
| Atomic clock | time, by counting atomic state transitions | satellite navigation, telecom and financial timestamping, metrology |
| Atom interferometer | gravity and acceleration | subsurface mapping, fundamental-constant measurement, navigation without satellites |
| Diamond magnetometer (nitrogen-vacancy center) | magnetic fields at nanoscale | biomagnetic measurement, high-radiation and high-temperature environments |
Two capabilities in that table are worth stating plainly because they sound like claims and are not. Diamond-based magnetometry reaches resolution “capable of detecting single neurons firing,” which is the same measurement problem as non-invasive brain and heart imaging. And atom interferometers “can detect gravitational fields with unprecedented accuracy with applications that range from high-precision measurements of fundamental physical constants to terrestrial subsurface sensing and precision, navigation, and timekeeping,” which in practice means reading what is under the ground without digging, and navigating when satellite positioning is unavailable or jammed.
Source: Crowder testimony, 22 January 2026, energy.gov.
Why is quantum sensing commercial when quantum computing is not?
Because a sensor needs one quantum system to behave well for a moment, and a computer needs many of them to behave well together, repeatedly, for the length of a calculation.
A quantum sensor exploits sensitivity. A quantum computer has to protect against it, holding fragile states in coherence long enough to run thousands of operations while the environment tries to destroy them. That is the error-correction problem, and it is why a machine’s logical qubit count is orders of magnitude below its physical one. The physics underneath the two is the same. The engineering burden is not remotely comparable.
The practical consequence for anyone reading coverage of this field: a claim that “quantum technology is decades away” is false about sensing and defensible about computing, and a claim that “quantum is here” is true about sensing and unearned about computing. Both statements circulate as if the field were one thing.
What are quantum computers being built for?
The DOE’s quantum lead names the target explicitly. The first fault-tolerant machines “will serve as novel scientific instruments, unlocking unprecedented opportunities for scientific discovery,” and may eventually “redefine scientific boundaries in applications such as those across materials science, quantum chemistry, and high-energy physics, among many others.”
Those three, from the agency writing the checks. What each one reaches:
Quantum chemistry. Modeling how molecules actually behave, which is a quantum-mechanical problem that classical methods approximate. The worked example most people already depend on is ammonia synthesis. The Haber-Bosch process makes it with enormous heat and pressure, and the International Energy Agency puts ammonia production at roughly 2% of global total final energy consumption, with about 70% of that output going to fertilizer. Close to half the world’s population is fed on crops grown with synthetic nitrogen. Bacteria run the same nitrogen-fixing reaction at ordinary temperature and pressure with an enzyme called nitrogenase, and the mechanism is still not fully modeled. A machine that could simulate it properly changes the energy cost of feeding people.
Materials science. Battery chemistry, superconductors and catalysts are all questions about how electrons arrange themselves in matter. Today those are answered largely by synthesizing candidate materials and testing them, which is slow and expensive. Simulation shortens that loop, and every argument about grid storage, electric vehicles and device battery life sits downstream of it.
High-energy physics. Simulating quantum field theories and lattice gauge theories, which is pure science rather than a product, and is a legitimate reason to build an instrument on its own terms.
Source: Crowder testimony, 22 January 2026, energy.gov; International Energy Agency, “Ammonia Technology Roadmap,” 2021, iea.org; Our World in Data, “How many people does synthetic fertilizer feed?”, ourworldindata.org.
Why chemistry and materials science instead of logistics and finance?
Because the advantage comes from a structural match between the machine and the problem, and only some problems have it.
Richard Feynman made the argument in 1981, and it has not needed revision: a classical computer simulating a quantum system pays an exponential cost in the size of the system, because it has to track a state space that grows exponentially. His proposal was to stop approximating and build a simulator that obeys the same physical rules as the thing being simulated. Chemistry is quantum mechanics, so it inherits that match directly.
Portfolio optimization, vehicle routing and scheduling do not. They are classical combinatorial problems with decades of extremely well-funded classical algorithm development behind them, and the quantum algorithms proposed for them offer modest theoretical speedups under assumptions that rarely survive contact with real instance sizes and error rates. The problem is not that quantum methods are forbidden there. It is that the classical competition is excellent and the theoretical margin is thin.
Source: R. P. Feynman, “Simulating Physics with Computers,” International Journal of Theoretical Physics 21, 467–488 (1982), doi.org/10.1007/BF02650179.
Has a quantum computer beaten a classical computer at a useful problem?
No. As of 2026 there is no demonstration of quantum advantage on a problem that somebody outside the experiment needed solved.
What does exist is a genuine and improving record on benchmarks, and the distinction between those two things is the most abused idea in coverage of this field.
The strongest current result. On 30 July 2026, IBM and the University of Chicago announced a verified quantum advantage: a computation finished in roughly 15 minutes that leading classical simulation methods cannot practically reproduce, run on 70 logical qubits with 2,415 logical two-qubit operations and 468 logical T gates, at logical error rates about 10 times lower than the underlying physical error rates. The method is doped Clifford sampling, and the verification is the genuinely new element, because it allows the answer to be trusted rather than merely received. It is a real milestone in the science. It is also a sampling task, chosen because it is hard for classical machines rather than because anyone wanted the output.
Why that caution is earned. In 2019 Google claimed quantum supremacy on its 53-qubit Sycamore processor with a sampling task completed in about 200 seconds that Google estimated would take a classical supercomputer 10,000 years. IBM responded within days that a classical machine could do it in about 2.5 days, and subsequent classical work narrowed the gap further. The headline number did not survive a motivated classical community, and that pattern has repeated often enough to be the default expectation.
The milestone that matters more than any qubit count. Google’s Willow chip, 105 qubits, went below the surface-code threshold in December 2024: making the error-correcting code larger made the logical error rate go down rather than up. That is the property a scalable machine requires, and it is peer-reviewed rather than announced.
And classical computing is not standing still. A 2026 joint result from the University of Osaka and Fixstars ran one of the largest classical simulations of quantum-chemistry circuits on 1,024 GPUs, pushing past the previous 40-qubit ceiling, and the classical simulation outperformed the quantum version on every metric tested for accuracy and efficiency. Any timeline that assumes one side improves while the other waits is wrong before it starts.
Where the machines actually stand:
| Builder | Best demonstrated |
|---|---|
| IBM | 1,121-qubit Condor (2023); public target of 200 logical qubits in 2029 |
| Willow, 105 qubits, below-threshold error correction (2024) | |
| Quantinuum | Helios (November 2025): 98 physical, 48 logical, 99.921% two-qubit fidelity |
| IonQ | AQ 64 on Tempo (October 2025) |
Source: IBM Newsroom, “IBM and the University of Chicago demonstrate quantum advantage,” 30 July 2026, newsroom.ibm.com, and the paper “Sampling hard circuits with verifiably high fidelity,” arxiv.org/abs/2607.25941; F. Arute et al., “Quantum supremacy using a programmable superconducting processor,” Nature 574, 505–510 (2019), doi.org/10.1038/s41586-019-1666-5, and IBM Research, “On quantum supremacy,” research.ibm.com; R. Acharya et al., “Quantum error correction below the surface code threshold,” Nature 638, 920–926 (2025), arxiv.org/abs/2408.13687; University of Osaka and Fixstars, 2026, eurekalert.org; IBM, “Large-scale fault-tolerant quantum computing,” ibm.com; Quantinuum, “Introducing Helios,” quantinuum.com; IonQ, “IonQ hits AQ 64,” ionq.com.
Which quantum computing applications are being oversold?
The marketed list and the DOE’s list overlap very little, and the gap is where most commercial quantum material lives.
| Sold as imminent | Where it actually stands |
|---|---|
| Portfolio optimization and quantitative finance | The most heavily marketed category and among the weakest technically. Classical solvers are mature, well funded and improving |
| Logistics and route optimization | Same shape. Demonstrations are small, and classical heuristics are strong on real instance sizes |
| Quantum machine learning | A legitimate research area with no demonstrated advantage on a real dataset |
| Drug discovery | Sits downstream of quantum chemistry, so it inherits chemistry’s timeline rather than improving on it |
| ”Quantum AI” as a product category | Largely a naming exercise. The useful question is which of the two words is doing the work |
| Quantum annealing as general-purpose computing | Annealers target one narrow class of optimization problem, and the classical competition on that class is fierce |
None of that makes the companies fraudulent, and several of them do serious research. It makes the application claims premature, which matters because those claims are what procurement decisions and press coverage rest on.
How do you tell a real quantum result from a marketing one?
Two questions decode most coverage, and neither requires physics.
Is that a logical qubit or a physical one? Physical qubits are noisy hardware. Logical qubits are error-corrected assemblies of many physical ones, and they are the unit that determines what a machine can actually run. A headline quoting a four-digit physical count next to a claim about breaking encryption is comparing two different things. See Logical vs Physical Qubits.
Was the problem chosen because it is useful, or because it is hard for classical computers? A benchmark selected for classical hardness demonstrates capability. A problem selected because someone needed the answer demonstrates value. Both are worth reporting, and conflating them is how a sampling result becomes “quantum computers now outperform supercomputers.”
How Do You Tell Real Quantum Progress From Hype carries the longer version of this, and Quantum Advantage and Quantum Supremacy defines the terms precisely.
Is breaking encryption what quantum computers are for?
Breaking public-key cryptography is the one application that is proven on paper before the machine exists, which makes it unusual in this list rather than central to it.
Peter Shor published the method in 1994, and it has been checked continuously since. Everything else in this note is waiting for somebody to find a job worth doing with the machine; this one already has its job and is waiting only on hardware. That asymmetry is why the security world plans against it while the application world is still surveying.
It is also narrower than most coverage implies. Quantum computing breaks public-key cryptography, meaning RSA, Diffie-Hellman and elliptic curve, the mathematics that lets two parties who have never met agree on a shared secret. Symmetric encryption such as AES-256 survives, and NIST’s own assessment is that Grover’s speedup “does not render cryptographic technologies obsolete” and that “doubling the key size will be sufficient to preserve security.” The NSA advisory that retires RSA and elliptic curve for national-security systems keeps AES-256 as the required symmetric cipher. See What Can a Quantum Computer Actually Break for the full split, and Harvest Now, Decrypt Later (HNDL) for why the timeline question is not the same as the deadline question.
Source: P. W. Shor, “Polynomial-Time Algorithms for Prime Factorization and Discrete Logarithms on a Quantum Computer,” 1995, arxiv.org/abs/quant-ph/9508027; NIST, “Report on Post-Quantum Cryptography,” NISTIR 8105, April 2016, csrc.nist.gov; NSA, “Announcing the Commercial National Security Algorithm Suite 2.0,” nsa.gov.
What is being built with public money?
The United States funds this as an instrument program rather than as a product program, and the structure says what the government thinks it is buying.
The National Quantum Initiative Act of 2018 created 5 DOE National Quantum Information Science Research Centers, each led by a national laboratory. Together they span more than 50 academic institutions across 22 states and more than 18 industry partners, have hired over 300 scientists, supported more than 600 PhD students and postdoctoral researchers, and trained over 2,500 external personnel. The centers were renewed in November 2025 for a further 5 years.
The Genesis Mission is the DOE platform connecting supercomputers, AI systems and quantum computers, including using AI to assist with error correction and with the discovery of new quantum algorithms.
Read the shape of that rather than the dollar figures: the money is buying position in front of an instrument and the people who can operate it, which is what a government does when it believes the applications will be found after the machine exists rather than before.
Source: Crowder testimony, 22 January 2026, energy.gov.
Has this happened before?
Yes, and the parallel is close enough that the DOE’s own quantum lead reached for it. Speaking at Quantum USA 2026, Crowder placed quantum computing at its “ENIAC moment,” still needing applications and orders of magnitude more performance.
In 1943 the US Army was designing artillery faster than it could produce firing tables, the books of numbers telling a gunner how to correct for range, wind, temperature and air density. The tables were computed by roughly 200 women whose job title on the paperwork was literally “computer,” because that is what the word meant. One trajectory took a person 20 to 40 hours on a mechanical calculator, and a single firing table required about 1,800 of them. The Army funded a machine to do that one narrow job, and John Mauchly and J. Presper Eckert built ENIAC at the University of Pennsylvania.
When it was finished in 1945, the first substantial calculation run on it was a problem for Los Alamos, months before the machine was shown publicly. It was not a firing table. What the machine became after that, the general-purpose computer and everything downstream of it, was not written down by anyone funding it in 1943, because nobody could have written it down.
The lesson people take from ENIAC is usually that every expensive instrument turns into a consumer product, which is the wrong lesson and the one vendors prefer. The defensible lesson is narrower: the uses arrived after the machine. That cuts against confident optimism and confident dismissal in equal measure, because it means anyone naming what quantum computing will be worth in 2050 is guessing, whether they are selling it or dismissing it.
Common misconceptions
“Quantum computers are just much faster computers.” They are not faster at general computation, and for most tasks they are dramatically worse. They offer an advantage on a narrow class of problems with the right mathematical structure, and running ordinary software on one would be slower than a laptop.
“Nothing quantum works yet.” Quantum sensing is deployed and commercial, and atomic clocks have been operating in satellite navigation for decades. The statement is defensible about computing and false about the field.
“A quantum computer tries every answer at once.” Superposition is not parallel search. Extracting a useful answer requires interference that cancels wrong answers and reinforces right ones, which is why useful quantum algorithms are rare and hard to design rather than automatic.
“More qubits means a better machine.” Physical qubit counts without error rates, connectivity and coherence times describe very little. See Logical vs Physical Qubits and NISQ vs CRQC.
“Quantum advantage means quantum computers are now more useful than classical ones.” Demonstrated advantage so far is on benchmarks selected for classical hardness. No result has yet beaten a classical machine at a task somebody needed performed.
“Quantum computing and quantum encryption are the same subject.” Quantum key distribution and quantum random number generation are separate technologies with separate maturity levels and separate arguments. See Quantum-Native Security MOC.
Questions people ask
Is quantum sensing really in my phone? Not in the phone itself. The satellites your phone reads for positioning carry atomic clocks, and positioning is computed from the timing they broadcast, so the quantum device is in the system rather than in your hand.
When will a quantum computer do something useful? Nobody credible names a year. The DOE frames the current state as needing both applications and orders of magnitude more performance, and the useful posture is to track logical qubit counts and error rates rather than announcements.
Are quantum computers going to replace regular computers? No. They are special-purpose instruments expected to work alongside classical systems, which is also how the DOE’s Genesis Mission describes the architecture.
If chemistry is the target, why does every quantum ad mention finance? Finance and logistics have buyers with budgets and short procurement cycles, and chemistry research does not buy machines the same way. The marketing follows the money rather than the physics.
Does the 2026 IBM result mean the encryption timeline moved? Not directly. It is a sampling benchmark on 70 logical qubits, and breaking RSA-2048 requires a machine several orders of magnitude larger. It is evidence that error-corrected computation is progressing, which is a different claim. See Cryptographically Relevant Quantum Computer (CRQC).
Is quantum annealing a quantum computer? An annealer is a quantum device built for one narrow class of optimization problem. It cannot run Shor’s algorithm and it is not a general-purpose gate-model machine, so treating the two as interchangeable produces wrong conclusions in both directions.
Should I believe a vendor who says quantum will transform my industry? Ask which of the DOE’s three areas their claim reduces to, and ask whether their evidence is a demonstration or a projection. If the claim is not a simulation problem and the evidence is a roadmap, it is a projection.
Is any of this relevant to me if I am not a physicist? The security half is, because the timeline for protecting long-lived data does not wait for the machine. See Harvest Now, Decrypt Later (HNDL) and Is the Quantum Threat Overhyped.
The map is free. Turning it into a migration your auditors accept is the work I do at LaMarr Labs.
Last verified 2026-09-04 · Maintained by Addie LaMarr, LaMarr Labs.