Quantum Computing Explained: What It Is and Why It Matters Now

Quantum computing occupies an unusual position: enormously funded, genuinely profound, and — for almost every practical purpose today — not yet useful. Both the excitement and the scepticism are justified, which makes it one of the harder technologies to think about clearly.

This article explains what a quantum computer actually is, why it might be dramatically faster at certain problems and no faster at most others, what the engineering obstacle really is, and what a sensible organisation should do about it now rather than in ten years. No mathematics beyond arithmetic, no code, and no claims that it will solve everything.

• What a qubit is and how it differs from a bit
• Superposition, entanglement and interference, explained plainly
• Why quantum computers are fast at some problems and not others
• What error correction is and why it dominates everything
• The real timeline, separating demonstrations from utility
• What to do now — particularly about cryptography

1. The one-paragraph version
2. Bits and qubits
3. Superposition: what it is and is not
4. Entanglement
5. Interference: where the speed comes from
6. Measurement, and why it is destructive
7. What a quantum algorithm actually does
8. The algorithms that matter
9. Why most problems get no speedup
10. Noise and decoherence
11. Error correction
12. How qubits are physically built
13. The current state, honestly
14. Quantum advantage and what it proved
15. The cryptography problem
16. What to actually do now
17. Twelve misconceptions
18. A worked example: searching a list
19. Glossary
20. Frequently asked questions

1. The one-paragraph version

A quantum computer stores information in physical systems that obey quantum mechanics rather than in ordinary electrical switches. This allows a register to hold a combination of many values at once, and allows operations to act on all of them simultaneously. That sounds like it should make everything faster, and it does not — because reading the result gives you only one value, chosen at random. The entire art is designing operations so that wrong answers cancel each other out and the right one becomes overwhelmingly likely to be what you read. Only a small number of problems have a known structure that permits this, and building hardware stable enough to run those procedures is the obstacle the field has been working on for thirty years.

2. Bits and qubits

A classical bit is a switch. It is zero or one. Eight of them hold one of two hundred and fifty-six possible values — one, specifically, at any moment.

A qubit is a physical system with two distinguishable states, small enough that quantum mechanics governs it — the spin of an electron, the polarisation of a photon, the state of a tiny superconducting circuit. Like a bit, it can be zero or one. Unlike a bit, it can be in a superposition: a specific combination of both, described by two numbers that determine how much of each is present and how they relate in phase.

The consequential part is what happens with several qubits. Two qubits do not hold two combinations; they hold a combination of all four possibilities. Three hold a combination of eight. Fifty hold a combination of over a quadrillion. The description of an n-qubit system requires two-to-the-n numbers.

This exponential scaling is the source of both the promise and the difficulty. It is why a quantum computer can, in principle, do things no classical machine can — and why simulating one on a classical machine becomes impossible past roughly fifty qubits, since the memory required exceeds anything buildable.

3. Superposition: what it is and is not

Superposition is the most misexplained concept in the field, so it is worth being careful.

What it is not: the qubit is not secretly zero or one with us merely being ignorant. That is classical uncertainty, and experiments have ruled it out. Nor is the qubit "in both states at once" in the sense of running two computations in parallel universes, which is a picturesque description that leads to wrong predictions.

What it is: the qubit's state is genuinely a combination, described by amplitudes — numbers that can be positive or negative, and more generally complex. When you measure, you get zero or one with probability related to those amplitudes, and the state collapses to whichever you got.

The fact that amplitudes can be negative is the entire difference between quantum computing and randomness. Probabilities only add. Amplitudes can add or cancel. A path to a wrong answer with amplitude plus-one and another with minus-one cancel to zero — that answer becomes impossible. Nothing classical does this, and every quantum speedup ultimately comes from it.

4. Entanglement

Two qubits can be correlated in a way with no classical equivalent. Measure one and you instantly know something about the other, regardless of distance, and — critically — the pair has no description as two separate qubits. The state describes the pair as a whole.

The common framing of entanglement as spooky action at a distance is a distraction for our purposes. No information travels; you cannot signal with it. What matters computationally is different: entanglement is what allows an operation on one qubit to affect the whole register's state, and it is what makes the exponential state space genuinely usable rather than merely large.

Without entanglement, a quantum computer is efficiently simulable classically and offers nothing. It is the resource that makes the machine more than the sum of its qubits.

5. Interference: where the speed comes from

Here is the crux, and the part most explanations skip.

Suppose you place a register in a superposition of every possible input — for fifty qubits, a quadrillion values — and apply an operation. The operation acts on all of them. This is where people conclude the machine tried every possibility at once and therefore solved the problem.

It did not, because measurement returns one value, at random. A superposition of a quadrillion answers, measured, gives one random answer, which is no better than guessing.

The actual technique: arrange the operations so that amplitudes for wrong answers cancel and amplitudes for right answers reinforce. Then measurement returns the right answer with high probability. This is interference, and it is the same phenomenon as waves cancelling and reinforcing.

Designing a sequence of operations that produces the right interference pattern for a given problem is what a quantum algorithm is. It requires the problem to have mathematical structure the algorithm can exploit. Most problems do not, which is why quantum computers are not general-purpose accelerators.

6. Measurement, and why it is destructive

Measurement is not passive observation. It forces the system into a definite state and destroys the superposition irreversibly.

Three consequences shape everything about how these machines are used:

• You cannot inspect intermediate states. Looking collapses the computation. Debugging is therefore extraordinarily awkward.
• You get limited information out. Fifty qubits hold a quadrillion amplitudes; measurement yields fifty bits. The algorithm must concentrate the answer into those bits.
• Results are probabilistic. Most algorithms are run many times and the distribution of outcomes analysed. A single run rarely means anything.

There is also a deeper restriction: quantum states cannot be copied. This forbids the backup-and-retry approach that classical error handling relies on, and it is why quantum error correction had to be invented from scratch rather than adapted.

7. What a quantum algorithm actually does

Nearly every useful quantum algorithm follows the same four-step shape:

1. Prepare. Put the register into a superposition, usually of all possible inputs.
2. Compute. Apply operations encoding the problem, acting on the whole superposition and typically writing information into the amplitudes' phases rather than their magnitudes.
3. Interfere. Apply further operations causing wrong answers to cancel and right ones to reinforce. This is the clever part and where the algorithms differ.
4. Measure. Read the result. Repeat if the probability of correctness is high but not certain.

The third step is the whole discipline. Steps one, two and four are broadly mechanical. Finding an interference pattern that isolates the answer to a useful problem is why there are so few quantum algorithms after decades of effort — perhaps a few dozen genuinely distinct ones, of which a handful matter.

8. The algorithms that matter

Factoring large numbers. The famous one. It provides an exponential speedup over the best known classical methods, turning a problem that would take longer than the universe's age into one taking hours. This matters because the difficulty of factoring is what secures a large portion of current encryption. This is the single result that made governments fund the field.

Unstructured search. Finding an item in an unsorted collection of N items in roughly the square root of N steps, rather than N. A quadratic speedup — real, general, and much less dramatic than exponential. For a trillion items it is a million steps instead of a trillion, which sounds enormous until you account for how much slower each quantum step is.

Simulating quantum systems. The application most likely to matter first, and the original motivation. Molecules and materials are quantum systems, and simulating them classically is exponentially hard — which is why computational chemistry approximates heavily. A quantum computer simulating a quantum system does not face that barrier. Realistic targets include catalyst design, battery chemistry and nitrogen fixation, where a better answer has enormous economic value.

Optimisation and linear algebra. The area with the most commercial excitement and the weakest evidence. Various approaches promise speedups on optimisation and machine-learning problems. Several apparent advantages have evaporated when someone found a better classical algorithm, and the honest position is that this remains an open research question rather than a near-term application.

9. Why most problems get no speedup

This is the section that separates informed expectations from marketing.

A quantum speedup requires the problem to have structure the interference pattern can exploit — a periodicity, a symmetry, an algebraic property. Factoring has such structure, which is precisely why it is vulnerable.

Most computation has no such structure. Rendering, databases, web serving, video encoding, most machine learning, most business logic — no known speedup, and for many, provable arguments that none exists. A quantum computer running ordinary software would be dramatically slower than a laptop, because each operation is far slower and the machine has no advantage to compensate.

There is a further practical obstacle: getting data in. Many proposed algorithms assume the ability to load a large classical dataset into quantum states efficiently. No general method exists, and for data-heavy problems the loading cost can exceed the entire computational advantage. Several published speedups quietly depend on this assumption.

The realistic picture is a specialised coprocessor for a narrow set of problems, used alongside classical machines that do everything else — not a replacement for anything.

10. Noise and decoherence

Now the engineering, which dominates the field's actual work.

Quantum states are extraordinarily fragile. A stray magnetic field, a vibration, a photon of heat, or interaction with anything at all disturbs them. The state degrades into ordinary classical randomness — decoherence — and the computation is ruined.

Superconducting qubits maintain coherence for tens to hundreds of microseconds. Operations take tens to hundreds of nanoseconds. That allows perhaps a few thousand operations before the state is gone, and every one of those operations is itself imperfect.

Current error rates run around one in a thousand per operation for the best two-qubit gates. Sound impressive until you compound it: a thousand operations gives roughly even odds of at least one error, and a single error usually invalidates the entire result. Useful algorithms need millions to billions of operations.

This gap — between what hardware sustains and what algorithms require — is roughly six orders of magnitude, and closing it by improving qubits alone is not plausible. Hence error correction.

11. Error correction

The idea that makes large quantum computers conceivable, and the reason they remain years away.

Classical error correction copies data and takes a majority vote. Quantum states cannot be copied, and measuring to check for errors destroys them. Quantum error correction solves both problems with genuine ingenuity: information is spread across many physical qubits so that errors can be detected by measuring relationships between qubits — which reveals whether an error occurred without revealing the encoded value, and therefore without collapsing it.

A group of physical qubits behaving as one reliable qubit is a logical qubit. The threshold theorem shows that if physical error rates fall below a certain level, arbitrarily long computations become possible by adding more physical qubits per logical one.

The overhead is the problem. Depending on the scheme and the target error rate, a logical qubit requires roughly a thousand physical qubits — with estimates ranging from a few hundred to tens of thousands. So a machine needing a few thousand logical qubits to factor a cryptographically relevant number needs millions of physical ones. Current devices have hundreds to low thousands.

Progress is real. Error correction has been demonstrated with logical error rates below physical ones — the crossing point that proves the approach works — and logical qubits have been operated and shown to improve as more physical qubits are added. That is a genuine milestone. It is also several orders of magnitude from a useful machine.

12. How qubits are physically built

No approach has won. Superconducting and trapped-ion systems lead on demonstrated results; neutral atoms have advanced quickly; topological approaches would change the calculus entirely if they work, which remains uncertain.

Worth understanding: the refrigeration requirement for superconducting machines is not incidental. These systems operate near absolute zero, colder than deep space, and the apparatus is large, expensive and power-hungry. This is a substantial reason quantum computing will be accessed as a cloud service rather than owned.

13. The current state, honestly

Present machines are described as noisy intermediate-scale devices — enough qubits to be interesting, too noisy to be reliable, no error correction.

What they can do: run small algorithms; demonstrate principles; support research; and produce results occasionally competitive with classical simulation on carefully chosen problems.

What they cannot do: run any algorithm long enough to matter commercially; break any encryption in use; or outperform classical computers on any problem anyone actually needs solved.

The honest summary is that no organisation is currently getting production value from quantum computing. Many are experimenting, several are building genuine expertise, and a number of announced applications, examined closely, turn out to be small demonstrations that a classical computer handles faster.

14. Quantum advantage and what it proved

Several groups have claimed quantum advantage — a quantum machine outperforming the best classical one on some task.

These claims are genuine and narrower than headlines suggest. The tasks are specifically constructed to be hard classically and natural quantumly, with no practical use whatsoever. They are proof-of-principle experiments demonstrating that a quantum device can do something classically infeasible, which is a real scientific milestone.

Every such claim has also been contested, usually by someone finding a smarter classical algorithm or applying more classical computing power, and the gap has repeatedly narrowed. That back-and-forth is healthy science and it makes the boundary genuinely blurry.

The distinction to hold: quantum advantage means outperforming classical machines at something. Quantum utility means outperforming them at something useful. The first has arguably been achieved. The second has not, and it is the one that matters commercially.

15. The cryptography problem

The one genuinely urgent consequence, and the reason a technology years away needs attention now.

A sufficiently large error-corrected quantum computer would break the public-key cryptography securing most internet traffic, digital signatures and certificate infrastructure. Not weaken it — break it.

Estimates for the machine required run to millions of physical qubits. That is far beyond current capability and, on most credible projections, a decade or more away, with genuine uncertainty in both directions.

The urgency comes from harvest now, decrypt later. An adversary can record encrypted traffic today and decrypt it when the capability arrives. Any data that must remain confidential for ten or twenty years — medical records, state secrets, long-term contracts, intellectual property — is exposed today if it crosses a network protected only by vulnerable cryptography.

The response is post-quantum cryptography: algorithms based on mathematical problems with no known quantum speedup. Standards have been published, implementations exist in major libraries, and migration is a multi-year undertaking for any large organisation. This is not speculative preparation; it is a scheduled infrastructure programme with a deadline nobody knows precisely.

Note that symmetric encryption and hashing are much less affected — the search algorithm halves their effective strength, addressed by using larger keys. The problem is specifically public-key cryptography.

16. What to actually do now

Everyone should do this:

• Inventory your cryptography. What algorithms, where, in what systems, protecting what. Most organisations cannot answer this, and it is the prerequisite for everything else.
• Identify long-lived secrets. Anything needing confidentiality past roughly 2035 is at risk from recorded traffic today.
• Build crypto-agility. The ability to change algorithms without rebuilding systems. Valuable regardless of quantum computing, and hard to retrofit.
• Plan migration to post-quantum standards. Start with anything protecting long-lived data.
• Ask vendors about their roadmaps. Their timelines constrain yours.

Some organisations should also do this: if you work in chemistry, materials, pharmaceuticals or certain areas of finance, build enough expertise to evaluate claims and identify which of your problems might eventually fit. That means a small number of people who understand the field, not a large programme.

Almost nobody should do this: buy hardware, build a large quantum team, or plan a product around near-term capability. Cloud access is available for experimentation at a fraction of the cost, and near-term capability does not yet exist to build on.

17. Twelve misconceptions

1. "They try all possibilities at once." They compute on superpositions and must engineer interference to extract an answer.
2. "They will replace classical computers." Specialised coprocessors for narrow problems.
3. "They make everything faster." Most problems have no known speedup, and some provably none.
4. "Encryption is broken now." The required machines do not exist and are years away.
5. "Encryption is fine for now." Recorded traffic can be decrypted later. Long-lived secrets are exposed today.
6. "More qubits means better." Quality and connectivity matter as much as count.
7. "Quantum advantage means useful." The demonstrated tasks have no application.
8. "Quantum machine learning is here." Mostly unproven, with several claimed speedups withdrawn.
9. "Entanglement transmits information." It does not, and cannot.
10. "They are just very parallel." Parallelism without interference gives nothing.
11. "Simulators prove the hardware works." They run on classical machines and hit a hard wall around fifty qubits.
12. "It is always ten years away." Error correction has crossed a real threshold. The trajectory is genuine even if the date is not.

18. A worked example: searching a list

Take the simplest useful quantum algorithm — finding a marked item in an unsorted list of a million entries — and follow it, because it makes the abstract mechanics concrete.

The classical baseline. With no structure to exploit, you check items one at a time. On average you find it after half a million checks, and in the worst case a million. There is no cleverer classical method; the information simply is not there.

Preparation. Twenty qubits suffice, since twenty bits address a million positions. They are placed into an equal superposition of all million possible addresses. Every position now has the same amplitude, which means measuring immediately would return a random position — exactly as useless as guessing.

The marking step. An operation is applied that recognises the target and flips the sign of its amplitude, leaving all others unchanged. Note what has and has not happened: the target's amplitude is now negative while the rest are positive, but the probability of measuring it is unchanged, because probability depends on the amplitude's magnitude. Measuring here still gives a random answer. The information is present in the state and completely invisible to measurement.

The amplification step. A second operation reflects every amplitude about their average. Because the target's amplitude is negative and the others positive, the average is slightly below the common positive value. Reflecting about it pushes the target's amplitude up substantially and pulls every other one down slightly. The target is now marginally more likely to be measured than any other single position.

Repetition. Marking and amplification are repeated. Each round moves more amplitude onto the target. After roughly a thousand rounds — the square root of a million — nearly all the amplitude sits on the target, and measurement returns it with high probability.

The result. A thousand rounds instead of half a million checks. A quadratic speedup, and one that applies to any search with no structure at all, which is what makes it general.

Three cautions the example makes vivid. First, repeating too many times makes it worse — the amplitude overshoots and starts moving away. You must stop at the right point, which requires knowing the list size in advance. Second, each quantum operation is enormously slower than a classical comparison, so the crossover point where this beats a classical search is far higher than the step counts suggest. Third, and most importantly, the list must be encoded in a way the marking operation can query. If the million items are sitting in an ordinary database, loading them into a quantum state costs at least a million operations, which destroys the advantage entirely. The speedup is real when the "list" is implicitly defined by a function — all possible keys, all possible configurations — and largely illusory when it is actual stored data.

That last point generalises beyond this example, and it is the most common reason a proposed quantum application does not survive scrutiny.

19. Glossary

20. Frequently asked questions

When will quantum computers be useful?

For chemistry and materials simulation, plausibly within the next decade, and that is the application most likely to arrive first. For breaking current cryptography, most credible estimates run beyond ten years with real uncertainty in both directions. For general commercial computing, there is no reason to expect it at all, because most problems have no known speedup.

Should we be worried about our encryption?

Not today, and yes for anything with a long confidentiality lifetime, because traffic recorded now can be decrypted later. Inventory what cryptography you use and where, identify data that must stay secret past the mid-2030s, and plan migration to post-quantum standards starting with that data. Crypto-agility is worth building regardless.

Can we simulate a quantum computer instead?

Up to about fifty qubits, yes, and simulators are genuinely useful for developing and testing algorithms. Beyond that the memory required grows exponentially and exceeds anything buildable. That wall is precisely why quantum computers might be worth having.

Is quantum machine learning real?

It is an active research area with weak evidence so far. Several proposed speedups have evaporated when better classical algorithms were found, and many depend on loading large datasets into quantum states, for which no efficient general method exists. Treat commercial claims in this area with more scepticism than in chemistry simulation.

Why do they need to be so cold?

Superconducting qubits require near absolute zero because thermal energy at ordinary temperatures would destroy the quantum states instantly. Other approaches have different requirements — trapped ions need ultra-high vacuum, photonic systems can operate at room temperature — but every approach needs extreme isolation from its environment in some form.

How many qubits do we need?

It depends entirely on the problem and on error correction overhead. Useful chemistry simulation might need hundreds to thousands of logical qubits; breaking widely used encryption needs a few thousand. At roughly a thousand physical qubits per logical one, those translate to hundreds of thousands or millions of physical qubits, against current devices in the hundreds to low thousands.

Is any of this a scam?

The physics is unambiguously real and the engineering progress is genuine. Some commercial claims about near-term business value are considerably ahead of the evidence, particularly around optimisation and machine learning. A useful filter: ask what specific problem, what size, and how it compares to the best classical method — vague answers to those three questions are the reliable warning sign.

What should a technical leader do this year?

Inventory your cryptography and start planning post-quantum migration — that is real work with a real deadline. Beyond that, keep a small amount of awareness, use cloud access if you want to experiment, and resist building a programme around capability that does not yet exist. The cryptographic response is urgent; almost everything else is not.

Key takeaways

• Interference, not parallelism, is where quantum speedup comes from — wrong answers must cancel.
• Only structured problems benefit. Most computation has no known speedup and never will.
• Error correction is the obstacle, and it costs roughly a thousand physical qubits per reliable one.
• Chemistry simulation is the likeliest first real application, not optimisation or machine learning.
• The cryptography risk is present tense because recorded traffic can be decrypted later.
• Inventory and crypto-agility are the actions worth taking now; hardware is not.

Quantum computing is a genuine scientific achievement working through a difficult engineering phase toward a narrow but valuable set of applications. Holding both halves of that sentence at once — real and narrow, profound and years away — is the whole of an informed position on it.