What Is a Quantum Computer? What It Can and Can't Do
A quantum computer is a machine that computes using the rules of quantum mechanics. It does not try every answer at once. Instead, it uses "weights" that can cancel each other out, so the right answer becomes more likely to come out. It is expected to be much faster for some problems, such as simulating molecules and breaking certain kinds of encryption. But it will not replace your PC. As of 2026, it still makes many errors and remains a research project.
What you'll learn
- How an ordinary bit differs from a qubit (superposition and entanglement)
- Why "it tries every answer at once" is a myth
- What it is expected to be good at, and what it is bad at
- Why it is hard to build, and how far error correction has come
- What it means for encryption, and how to read the numbers in the news
Prerequisites: it helps to know that computers store information as 0s and 1s. This is explained in "How Do Computers Use Binary?"
Where you see it: quantum computers in the news
"A calculation that would take a supercomputer 1025 years, done in under five minutes." Headlines like this spread around the world in December 2024, when Google announced a new quantum chip called Willow. In Japan, Fujitsu and the research institute RIKEN announced a 256-qubit machine in April 2025.
Today's quantum computers sit in places such as research labs. Researchers and companies use them over a network.
How it works 1: bits and qubits
An ordinary computer represents information as a string of 0s and 1s. Each digit is called a bit. Inside a chip, a bit is held by things like the state of a switch called a transistor. So a bit is always set to exactly one value: 0 or 1.
In a quantum computer, each digit of information is called a qubit. Before it is measured, a qubit can hold a "weight" for 0 and a weight for 1. These weights can be positive or negative. In physics, this weight is called the amplitude. It is really a kind of number called a complex number. To keep things simple, this article treats it as a positive or negative number.
When you measure a qubit, you always get just one result: 0 or 1. Which one comes out is a matter of chance. The weight multiplied by itself (squared) gives the probability of that result. For example, say the weight of 0 is 0.8 and the weight of 1 is −0.6. Then you get 0 with a 64% chance and 1 with a 36% chance. After the measurement, the qubit turns into the state that came out.
A state like this, with weights on both 0 and 1, is called superposition. It is sometimes described as "both 0 and 1 at once," but that is not enough. What matters is that the weights can be positive or negative. This leads to the "canceling out" explained next.
How it works 2: canceling weights to make the right answer stand out
With 2 qubits, each of the 4 patterns 00, 01, 10 and 11 gets its own weight. With 3 qubits there are 8 patterns, and with 10 there are 1,024. Each extra qubit doubles the count. With 50, there are more than a quadrillion (a million billion). An ordinary computer would quickly run out of memory trying to write down all these weights.
But a measurement gives you just one string of 0s and 1s, one digit per qubit. If you simply measured, it would be no better than picking one possible answer at random.
So before measuring, a quantum computer applies a series of steps that change the weights. Each step is called a quantum gate. As the steps build up, weights that lead to the same result are added together. Weights with the same sign reinforce each other. A positive and a negative weight cancel out. In physics, this is called interference.
The steps of a quantum computation (its algorithm) are designed so the weights of wrong answers cancel and those of the right answer reinforce. Then, at the final measurement, the right answer comes out with high probability. However, such designs have been found only for some problems with a special structure.
How it works 3: entanglement, results that match even far apart
Two qubits can be put in a state where measuring them gives "both 0" or "both 1," each with a 50% chance. Look at either qubit alone and its result is random. Yet the two results always match. A state like this, where the results of qubits are linked, is called quantum entanglement.
The link holds even when the two qubits are moved far apart. Still, entanglement cannot be used to send information faster than light. A person who measures one qubit just sees a random 0 or 1. Only by comparing results over ordinary communication can you tell that they match.
But results that merely match can also be produced by ordinary means. Entanglement is a stronger link: the results stay linked even when you change how you measure (we skip the details here).
What it is expected to be good at
Calculating how molecules and materials behave
Electrons in a molecule behave according to quantum mechanics. To calculate them exactly on an ordinary computer, the amount of numbers to handle grows very quickly with the number of electrons. So in a 1982 paper, the physicist Richard Feynman proposed using a machine that obeys quantum mechanics to imitate nature. It is hoped this will help research on drugs, battery materials and catalysts.
A 2023 paper by researchers at Microsoft and elsewhere also says the most promising areas are chemistry and materials problems. But as of 2026, no clear case has been established where a quantum computer beats the best ordinary methods on a useful problem.
Factoring large numbers
Splitting a large number into prime numbers multiplied together is called prime factorization. It takes ordinary computers a very long time. In 1994, Peter Shor published a method that lets a quantum computer do this quickly. It is called Shor's algorithm, and it has a big impact on encryption (explained later).
Searching: only a square-root speedup
Lov Grover published a method in 1996 that cuts the number of lookups needed to find a target in data with no helpful order. For 1 million items, an ordinary search checks 500,000 on average. This method needs about 800. The number of lookups drops to around the square root of the number of items.
However, each step on a quantum computer is much slower, and fixing errors adds extra work. The 2023 paper mentioned above points out that a speedup of this size makes it hard to win on practical problems. There are also hopes for "optimization," such as planning delivery routes. As of 2026, solid evidence is still scarce.
What it is bad at: it won't replace your PC
Browsing the web, playing videos, writing documents: these tasks are not shaped so that canceling weights makes an answer stand out. A quantum computer, with slow steps and many errors, cannot compete with an ordinary computer at them.
It is also bad at reading in large amounts of data or taking out many results. One measurement gives only one string of 0s and 1s. So quantum computers are thought to suit problems with small inputs and very heavy calculation.
Why it is hard to build
The weights are very fragile
A qubit's weights are slowly disturbed by heat, faint radio waves and vibration around it. Soon they can no longer be used for canceling. This is called decoherence. In Google's Willow, a rough measure of how long a qubit holds its energy is 68 microseconds on average. That is less than a ten-thousandth of a second.
Some metals lose all electrical resistance when cooled, which is called superconductivity. Machines that use it cool the chip to an extremely low temperature. The chip Google announced in 2019 was cooled below 20 millikelvin by a machine called a dilution refrigerator. That is only 0.02 degrees above the lowest possible temperature (absolute zero, −273.15 °C).
Operations make many errors
In Willow, operations that use two qubits had an average error rate of 0.33%, about 1 in 300. At that rate, if you run 1,000 operations, the chance that an error happens somewhere is about 96%.
Error correction and logical qubits
With ordinary data, you can fix errors by making several copies and taking a majority vote. But a qubit's state cannot simply be copied. Instead, one piece of information is spread across many qubits. Only clues about "where an error happened" are measured, and the error is then fixed. This is called quantum error correction.
One unit protected by error correction is called a logical qubit. Each of the qubits it is built from is called a physical qubit. Adding more qubits makes error correction work better, but only when the physical qubits' error rate is below a certain line (the threshold).
In December 2024, Google published a paper in the journal Nature showing that Willow had gotten below this line. The logical qubit was made larger in three steps, and the error fell by about half (a factor of 2.14) at each step. The largest used 101 physical qubits. It held information about 2.4 times longer than the longest-lasting physical qubit.
However, this experiment showed only the "memory" part, which holds information. Doing useful calculations while fixing errors needs many more qubits and even fewer errors. In June 2025, IBM announced a plan to build a machine with 200 logical qubits by 2029. This is the company's own plan.
Different ways to build qubits
There are five main ways to build qubits. As of 2026, it is not yet settled which one has the edge.
- Superconducting: Tiny circuits of superconducting metal are made on a chip, using methods similar to those for semiconductors. Google, IBM, and Fujitsu with RIKEN use this approach. It needs extremely low temperatures.
- Trapped ions: Atoms with an electric charge (ions) are held floating in a vacuum and controlled with lasers. They make few errors, but operations are fairly slow.
- Neutral atoms: Atoms with no electric charge are lined up with "tweezers" made of laser light. This approach makes it easy to increase the number of qubits.
- Light: Uses particles of light (photons). It can build on optical communication technology, but making photons act on each other is hard.
- Spin: Uses a property (spin) of a single electron trapped in silicon or a similar material. It is hoped this can build on the technology of semiconductor fabs.
What it means for encryption
When you shop online, your browser and the site protect their connection with encryption ("What Does HTTPS Mean?"). In some steps, such as setting up a secret key at the start, a type called public-key cryptography is used. Its best-known forms, RSA and elliptic-curve cryptography, can be broken by Shor's algorithm.
But that needs a large machine with error correction. In 2025, a Google researcher estimated what it would take to break 2048-bit RSA. The answer was fewer than 1 million physical qubits and less than a week. A 2019 estimate said about 20 million qubits and 8 hours. So the new one takes longer but needs far fewer qubits. As of 2026, real machines still fall short, both in qubit count and in low error rates.
Even so, people are hurrying to prepare. The reason is a possible attack: collect encrypted messages now and decrypt them in the future. Information that must stay secret for decades needs a new kind of protection starting now.
In August 2024, the US National Institute of Standards and Technology (NIST) published three new encryption standards. They are believed to be hard to break even with a quantum computer, and are called post-quantum cryptography. They are FIPS 203, 204 and 205. In a draft from November of the same year, NIST set out a plan to stop using RSA and similar methods by 2035.
Encryption where both sides share the same key is called symmetric-key encryption. Against it, and against password guessing, a quantum computer only offers Grover's square-root speedup. Longer keys are thought to be enough to handle this. Today, passwords are mostly cracked because they are easy to guess or reused ("How Are Passwords Cracked?").
How to read the news
Claims of "quantum supremacy" and "quantum advantage"
There are two names, "quantum supremacy" and "quantum advantage," for using a quantum computer to solve a problem that ordinary computers cannot solve in a realistic time. In 2019, Google announced that its 53-qubit chip did in about 200 seconds a calculation that would take a supercomputer 10,000 years. IBM researchers quickly pushed back with an estimate that, with smarter methods, it could be done in 2.5 days. In 2022, a paper reported solving it in about 15 hours using 512 GPUs.
As methods for ordinary computers improve, the gap shrinks. Also, the calculation compared was a problem that favors quantum computers, with no practical use of its own. Willow's "1025 years" from the start of this article is also Google's estimate for this kind of problem. The October 2025 claim of "about 13,000 times faster than a supercomputer" is also Google's estimate. The company describes it as a step toward practical use.
Don't compare by qubit count alone
In December 2023, IBM announced Condor, with 1,121 qubits, and Heron, with 133 qubits, at the same time. The one IBM put the most weight on was Heron, which makes fewer errors. Even with many qubits, a machine with many errors cannot run long calculations.
When you see a news story, check three things. What problem was solved? Was the comparison against the best method for ordinary computers? Are the error rate and the number of logical qubits given?
By the numbers
| Item | Number | When / source |
|---|---|---|
| Physical qubits in Willow | 105 | 2024, Google |
| Error rate of Willow's two-qubit operations | 0.33% on average | 2024, Google |
| Error rate of two-qubit operations on Helios (trapped ions) | About 0.08% (98 qubits) | 2025, Quantinuum announcement |
| Estimate for breaking 2048-bit RSA | Fewer than 1 million physical qubits, less than a week | 2025, Google researcher (assuming a 0.1% error rate) |
One million is about 10,000 times Willow's 105. On top of that, the estimate assumes an operation error rate of 0.1%, lower than Willow's 0.33%.
Common misconceptions
Myth: A quantum computer tries every answer at the same time and picks the best one
In fact, a measurement gives only one result. You need steps that make the weights of wrong answers cancel out, and such steps are known for only a limited set of problems. The myth comes from a short explanation that spread widely on its own: "superposition = holding many states at once."
Myth: More qubits means better performance
In fact, what a machine can compute depends heavily on its error rate, how long it keeps its weights, and how its qubits are connected. For useful calculations, what matters is the number of logical qubits protected by error correction. The myth arises because qubit counts make easy headlines.
Myth: Quantum computers will soon break all encryption
In fact, the main type at serious risk is public-key cryptography, and no machine big enough to break it exists yet. But because of the risk of "collect now, decrypt later," the move to new encryption is already under way.
Learn more
- Compare the approaches in a table: "Superconducting vs Trapped-Ion Qubits"
- Compare encryption methods in a table: "RSA vs Elliptic-Curve vs Post-Quantum Cryptography"
- Qubit terms: transmon, surface code (the error-correction method used in Willow), NISQ
- Encryption terms: RSA cryptosystem, ML-KEM (one of the post-quantum standards)
- Quantum mechanics at work in ordinary chips: "Quantum Tunneling in Transistors and Flash Memory"