Quantum Academy
No. Every quantum computer built so far depends on classical computers to run. Designs for larger machines need more classical computing to correct their errors. Classical machines compile the quantum program, generate the signals that operate the qubits (the quantum counterparts of bits), turn measurements into data and, in machines that correct their own errors, work out which corrections to apply. They also run most of the workload. The quantum processor is an accelerator, a specialized unit that the classical system calls on for the few steps it does faster, as a graphics processor does for AI training.
The US National Academies of Sciences, Engineering, and Medicine reached the same conclusion in its 2019 report on quantum computing. Quantum computers, it found, are “unlikely to be useful as a direct replacement for classical computers” and “require a number of classical computers to control their operations.” It described them as special-purpose devices “analogous to a co-processor or an accelerator.”
What a quantum processor is good at
A quantum processing unit, or QPU, is the chip that contains the qubits. A QPU is faster than a classical processor only on problems with a mathematical structure that a quantum algorithm can exploit. Those problems belong to four families, ranked by the strength of the evidence:
- Hidden periodic structure, which includes factoring large numbers. Shor’s algorithm, published in 1994, factors far faster than any known classical method.
- Simulation of quantum systems, such as molecules and materials.
- Unstructured search, where Grover’s algorithm, published in 1996, finds one item among N in about √N steps, against N/2 for a classical search.
- Optimization and machine learning, where the evidence for a practical advantage is thinnest.
In a 2023 analysis in Communications of the ACM, Torsten Hoefler, Thomas Häner and Matthias Troyer concluded that square-root savings like Grover’s, called quadratic speedups, “are insufficient for practical quantum advantage.” They named chemistry and materials science as the most promising application.
Their numbers also show why a QPU makes a poor general-purpose computer. They assumed a future error-corrected machine, one that detects and fixes its own errors as it runs, with each operation taking 10 microseconds, against 0.7 nanoseconds for a classical chip’s cycle, more than 14,000 times shorter. The machine’s input and output would be 10,000 times slower than an existing classical chip’s. On that basis, they concluded that quantum computers “will be practical for ‘big compute’ problems on small data, not big data problems.” Web servers, payments databases and video editing are data-heavy jobs that classical hardware will keep doing.
The first family is why the quantum threat to cryptography is real. A large enough error-corrected machine running Shor’s algorithm would break RSA and elliptic-curve cryptography, the public-key methods used for key exchange and digital signatures. NIST published its first post-quantum standards in August 2024 to replace them, including ML-KEM (Kyber) for key establishment and ML-DSA (Dilithium) for digital signatures. That machine would be an accelerator too, with classical computers doing every step but one.
Factoring 15, step by step
Shor’s algorithm turns factoring into period finding. Pick a number a, divide a, a², a³ and so on by N, and note the remainders. The period, r, is the number of steps before the remainders start to repeat. Here are the steps for N = 15 and a = 7:
| Step | Runs on | N = 15, a = 7 |
|---|---|---|
| Check that N is odd and not a power of a prime | Classical | 15 passes |
| Check that a and N share no factor, using their greatest common divisor (gcd) | Classical | gcd(7, 15) = 1 |
| Find the period r of the remainders | Quantum | 7, 4, 13, 1, then repeating, so r = 4 |
| Recover r from the measurement, then check that r is even and a^(r/2) + 1 isn’t a multiple of N | Classical | r = 4 is even; 7² + 1 = 50 isn’t a multiple of 15 |
| Compute gcd(a^(r/2) − 1, N) and gcd(a^(r/2) + 1, N) | Classical | gcd(48, 15) = 3 and gcd(50, 15) = 5 |
| If a check fails, pick a new a and start again | Classical | Not needed |
At N = 15 you can find the period by hand. For a 2,048-bit RSA key the period can be hundreds of digits long, and that one row is the job a QPU would accelerate. The other five rows run on classical hardware at every key size. A machine able to break real keys is more likely to be assembled from several qubit technologies and run by a large classical control system than built as one chip.
The classical work in every quantum run
The National Academies report describes a quantum computer as four layers, and only one of them, the quantum data plane that contains the qubits, is quantum. The other three are classical electronics and computers that do four jobs on every run.
Compilation
A compiler, software running on an ordinary computer, rewrites the quantum program into the small set of operations a particular device supports. It then maps each of the program’s qubits to a physical location on the chip, accounting for which qubits can interact directly.
Control
Qubits respond to precisely timed analog signals: microwave and low-frequency electrical signals for superconducting qubits, and laser light or microwaves for trapped ions. Classical electronics generate those signals from the compiled instructions.
Readout
Measuring a qubit returns a single 0 or 1 and destroys the rest of its quantum state. Measurement electronics convert the analog signal into classical bits. Each run yields one outcome drawn from a probability distribution, so a program runs many times and classical software tallies the outcomes into a result.
Error-correction decoding
Physical qubits, the actual devices on the chip, are error-prone. Error correction encodes a logical qubit, the reliable unit an algorithm works with, across many physical qubits. Some of the physical qubits that make up a logical qubit are measured in every cycle to produce syndrome bits, which flag where errors probably occurred without revealing the protected information. A classical program called a decoder reads the syndrome bits and works out which corrections to apply.
The decoder has to process syndrome data as fast as the chip produces it. In one superconducting experiment published in Nature, the processor completed a cycle every 1.1 microseconds, roughly 900,000 cycles a second. The decoder processed syndrome data in real time for up to a million cycles, returning results with an average delay of 63 microseconds. The authors warn that “an increasing backlog of syndrome information can cause an exponential blow-up in computation time.” The National Academies report adds that this classical task grows with the size of the machine.
Quantum processors beside supercomputers
Supercomputing centers already install quantum hardware this way. In 2023, the EuroHPC Joint Undertaking, the European partnership that funds shared supercomputers, signed agreements with six European centers to host quantum computers integrated into their existing supercomputers. In June 2025, the Japanese research institute RIKEN began operating a superconducting quantum computer with 156 physical qubits together with its Fugaku supercomputer in Kobe.
A 2025 reference design for a widely used open-source scheduler, the software that assigns machines to users’ jobs, treats quantum devices as “schedulable first-class resources” and lets one job reserve classical computers and a QPU together. Across each step of a hybrid quantum-supercomputer job, from the queue to the decoder, only the quantum circuit itself runs on the QPU.
What buyers get
What you can buy is access to an accelerator: time on a QPU through a cloud service, or an allocation at a supercomputing center that hosts one. Nothing in your data center gets replaced, and your own classical code still prepares each problem and interprets each result.
You’ll need skills on both sides of that interface: someone who can judge which part of a problem fits one of the four families, and engineers who write the classical code around it and manage the data. A team with only one of those skills can’t tell whether the quantum step beats the best classical method.
Our Quantum Computing for Executives course covers what quantum computers can and can’t do, hardware approaches, realistic timelines for commercial utility, industry applications and how to evaluate vendor claims. Introduction to Quantum Computing is our introductory course on the technical basics.
When a pitch or a headline says quantum computers will replace classical ones, ask for the workload written out like the factoring table above, with every step marked quantum or classical. Then ask which classical method the quantum step was measured against, and by what margin it won.