What Is a Quantum Processing Unit?

Definition

A Quantum Processing Unit (QPU) is a specialized processor that exploits quantum mechanical phenomena — superposition, entanglement, and interference — to perform computations. Unlike classical processors that represent information as binary bits (0 or 1), QPUs use qubits that can exist in superpositions of 0 and 1 simultaneously, enabling fundamentally different computational approaches.

A QPU is not a replacement for a CPU or GPU. It is a specialized accelerator designed for a specific class of problems — primarily those with quantum-native structure, such as molecular simulation, certain optimization tasks, and quantum cryptography. Most near-term quantum computations use QPUs alongside classical processors in a hybrid architecture.

How QPUs Work

Classical computers process information as bits — binary switches that are deterministically 0 or 1. A QPU processes information as quantum bits (qubits) that obey the laws of quantum mechanics. Three quantum phenomena make QPUs computationally interesting:

• Superposition: A qubit can exist in a combination (superposition) of |0⟩ and |1⟩ simultaneously. Mathematically, the qubit state is α|0⟩ + β|1⟩, where α and β are complex probability amplitudes satisfying |α|² + |β|² = 1.
• Entanglement: Multiple qubits can become entangled — their quantum states become correlated in ways that have no classical analogue. Measuring one qubit instantly determines information about its entangled partner, regardless of distance.
• Interference: Quantum gates manipulate the probability amplitudes of qubit states, causing computational paths leading to wrong answers to destructively interfere (cancel out) while paths leading to correct answers constructively interfere (reinforce). This is what gives quantum algorithms their potential speedup.

Qubits

A qubit is any quantum two-level system that can be controlled with sufficient precision. Physical implementations include: superconducting Josephson junction circuits (cooled to ~15 mK), trapped atomic ions held in electromagnetic traps, neutral atoms captured in optical tweezer arrays, photons encoding quantum information in polarization or path, and semiconductor spin states. Each implementation has different trade-offs in coherence time, gate speed, and scalability.

The qubit count alone is a poor measure of a QPU's computational power. A 10-qubit system with high fidelity and long coherence may outperform a 1,000-qubit system with poor error rates for practical algorithms. Metrics like algorithmic qubits, quantum volume, and CLOPS attempt to capture overall system performance more accurately.

Quantum Gates

Quantum gates are the QPU's computational primitives — the quantum equivalent of logic gates (AND, OR, NOT). Unlike classical logic gates, quantum gates are reversible unitary operations: they rotate the qubit's state on the Bloch sphere without destroying quantum information. Common gates include: the Hadamard gate (H) — creates a superposition from a basis state; the CNOT gate — a two-qubit controlled operation that creates entanglement; the T gate and rotation gates for precise phase control; and native gates specific to each hardware platform (e.g., IBM's ECR gate, IonQ's native Mølmer-Sørensen gate).

Gate fidelity — how accurately a gate operation is implemented on real hardware — is one of the most important hardware metrics. A two-qubit gate fidelity of 99.5% means 0.5% error per gate; errors accumulate with circuit depth, limiting how many sequential operations a QPU can reliably perform.

Measurement

Quantum measurement is the process of extracting classical information from a qubit. When measured, a qubit in superposition α|0⟩ + β|1⟩ collapses to either 0 (with probability |α|²) or 1 (with probability |β|²) — the superposition is destroyed. This is why quantum circuits must be run many times ("shots") to build up statistical estimates of the output probability distribution. Readout fidelity — how accurately the measurement reflects the true qubit state — is a distinct error source from gate fidelity.

QPU Architectures

All QPUs manipulate quantum states, but the physical medium used for qubits — and the method of control — varies significantly across platforms:

See the full architecture comparison for detailed trade-offs.

QPU vs CPU

A CPU (Central Processing Unit) is a general-purpose sequential processor optimized for the workloads that make up the vast majority of computing: operating systems, databases, web services, and business logic. CPUs are extremely fast (GHz clock speeds), highly reliable, and can execute arbitrary programs.

A QPU is not a general-purpose processor. It is a specialized accelerator for a specific class of problems with quantum-native structure. QPUs cannot run web browsers, databases, or most software. They are designed to solve problems where quantum parallelism — exploring exponentially many computational paths simultaneously — provides a theoretical advantage.

The key distinction: CPUs execute classical algorithms deterministically; QPUs execute quantum circuits probabilistically, requiring multiple shots to build up statistical estimates. For the foreseeable future, QPUs will be used alongside CPUs, not instead of them.

For the GPU comparison, see QPU vs GPU vs CPU →

Current State of QPU Technology

We are in the NISQ era — Noisy Intermediate-Scale Quantum. Today's QPUs have 10–1,000+ physical qubits and can execute quantum circuits of modest depth, but lack the error correction needed for fault-tolerant computation. Key current limitations:

• Gate errors accumulate: circuit depth is limited before noise dominates results
• No error correction: physical qubit errors are not corrected in real time (most systems)
• Coherence times: qubits lose their quantum state after microseconds to milliseconds
• Connectivity: most QPUs can only directly interact neighboring qubits, requiring SWAP overhead for non-adjacent operations
• Queue times: cloud QPU access involves wait times of minutes to hours
• Classical simulation: problems up to ~50 qubits can often be simulated classically, limiting verifiable advantage

Progress is real and accelerating. Two-qubit gate fidelities have improved from ~90% to 99.5%+ over the past decade. Qubit counts are scaling. Error correction demonstrations are moving from theory to laboratory results. The roadmap toward fault-tolerant quantum computing is now an engineering challenge rather than a pure research question — but meaningful scale is likely 5–15 years out.

Major QPU Manufacturers

The QPU hardware industry includes a mix of large technology companies, well-funded startups, and national laboratory efforts. Major commercial QPU manufacturers:

Full list: QPU Provider Directory →

How to Access a QPU

No one buys a QPU the way they buy a GPU. Access is primarily cloud-based:

• IBM Quantum Platform

  IBM QPUs via Qiskit. Free tier available. Best for superconducting circuit research.

• Amazon Braket

  IonQ, Rigetti, QuEra, IQM. Pay-per-shot. Managed service, AWS integration.

• Azure Quantum

  IonQ, Quantinuum, Rigetti. Credits for new users. Azure ecosystem integration.

• Provider Direct

  IonQ Cloud, D-Wave Leap, Quantinuum Nexus offer direct API access to proprietary hardware.

Full cloud access guide: QPU Availability & Cloud Access →

Frequently Asked Questions

What does QPU stand for?

QPU stands for Quantum Processing Unit — a specialized processor that uses quantum mechanical phenomena to perform computation.

Is a QPU the same as a quantum computer?

Not exactly. A quantum computer is the full system: the QPU (quantum chip), the cryogenic cooling infrastructure (for superconducting systems), classical control electronics, and the software stack. The QPU is specifically the quantum processor — the chip where qubits are stored and quantum gates are applied.

How many qubits does a QPU need to be useful?

It depends on the application. Today's QPUs range from 5 to over 1,000 physical qubits, but the number of qubits is not the only metric. Gate fidelity, connectivity, and error rates determine actual computational capability. For near-term algorithms, effective performance is often measured in "algorithmic qubits" — the number of qubits that can participate in useful, error-limited computation.

What is the difference between physical qubits and logical qubits?

A physical qubit is the raw quantum two-level system (transmon, trapped ion, etc.) — it is noisy and error-prone. A logical qubit is an error-corrected virtual qubit encoded across many physical qubits using quantum error correction codes. A fault-tolerant quantum computer needs thousands of physical qubits per logical qubit. Today's systems operate at the physical qubit level.

Can a QPU replace a GPU for AI/ML workloads?

No — not in the foreseeable future. GPUs excel at massively parallel classical matrix operations, which is the foundation of modern machine learning. QPUs operate on fundamentally different computational principles and are not a replacement for GPU-based ML. Quantum computing may eventually accelerate specific ML subroutines (quantum linear algebra, sampling), but the comparison is not apples-to-apples.

What temperature does a QPU operate at?

Superconducting QPUs must be cooled to approximately 15 millikelvin (mK) — colder than outer space — using a dilution refrigerator. This is necessary to maintain superconductivity and reduce thermal noise that would destroy qubit coherence. Trapped-ion and neutral-atom systems can operate at room temperature for the ions themselves, though they still require sophisticated vacuum systems and laser equipment.

What programming languages are used to program a QPU?

Quantum circuits are typically written using domain-specific frameworks: Qiskit (Python, IBM), Cirq (Python, Google), PennyLane (Python, Xanadu), CUDA-Q (C++/Python, NVIDIA), and OpenQASM (assembly-level quantum IR). These compile quantum programs into gate sequences that run on the hardware.

What is quantum advantage?

Quantum advantage (sometimes called quantum supremacy for specific cases) refers to demonstrating that a quantum processor can solve a specific problem faster than any classical computer — or solve a problem that is intractable classically. Demonstrating quantum advantage for practically useful problems remains an open research challenge.

Can I access a QPU without buying one?

Yes. Multiple cloud providers offer QPU access: IBM Quantum Platform (IBM QPUs via Qiskit), Amazon Braket (IonQ, Rigetti, QuEra, IQM), Azure Quantum (IonQ, Quantinuum, Rigetti), and provider-direct cloud systems. Access is pay-per-shot or subscription-based. Some providers offer free tiers for research.

What is NISQ?

NISQ stands for Noisy Intermediate-Scale Quantum — a term coined by physicist John Preskill to describe today's quantum processors. NISQ devices have 50–1,000+ qubits but lack error correction, making them error-prone and limited in circuit depth. NISQ algorithms (VQE, QAOA) are designed to work within these constraints.

Key Terms

Quick reference for terms used throughout this guide. Full definitions in the Quantum Computing Glossary.