NISQ
Also known as: noisy intermediate-scale quantum, NISQ era, NISQ devices, NISQ computers
NISQ, short for 'noisy intermediate-scale quantum,' describes the current era of quantum computers: machines with tens to around a thousand qubits that are too error-prone for full error correction.
The term was coined to set realistic expectations. While NISQ machines are valuable research tools, the hope that they would deliver broad commercial advantage has largely faded, and the field's focus has shifted to error-corrected systems.
NISQ, explained at three levels
At home
Think of the first airplanes: they really flew, but only short hops, and nobody would book a transatlantic trip on one. NISQ, short for noisy intermediate-scale quantum, describes quantum computers at that stage. They have enough qubits to do things ordinary computers find hard to simulate, but errors creep in so fast that programs must stay short. They're research tools, not yet workhorses.
At work
NISQ, or noisy intermediate-scale quantum, describes the quantum computers available today: tens to around a thousand qubits, without full error correction. If a vendor sells quantum computing services now, it's almost certainly NISQ hardware. For businesses, it's useful for learning and research, but it isn't expected to outperform classical computers on everyday business problems.
At home
Each part of the name means something. 'Noisy' means every gate has a small chance of error, roughly 0.1% to 1% for two-qubit gates, and those errors pile up. 'Intermediate-scale' means roughly 50 to around a thousand qubits: enough that classical computers struggle to simulate them exactly, but too few for full error correction. Together, they limit programs to somewhere between hundreds and a few thousand operations before results dissolve into random noise.
At work
Most NISQ algorithms are variational. A short quantum circuit with adjustable settings runs on the hardware, a classical computer measures the result, adjusts the settings, and repeats, much like training a machine learning model. Examples include the variational quantum eigensolver for chemistry and the quantum approximate optimization algorithm for optimization. Error mitigation techniques, which estimate and subtract noise without full correction, can sharpen results, but they need many repeated runs, raising cost and time.
At home
Variational algorithms rely on tuning circuit settings, and as circuits grow, many hit 'barren plateaus', regions where the landscape is so flat that the optimizer can't tell which way to move. Hardware noise flattens it further. Meanwhile, classical methods keep improving. In my opinion, those challenges are why NISQ commercial claims rarely survived careful comparison with the best classical methods, and why attention has shifted toward error-corrected machines.
At work
NISQ results are often matched by improved classical simulations, such as tensor network methods, soon after publication, as happened with parts of IBM's 2023 utility experiment. That moving baseline makes advantage claims fragile. The field now describes an 'early fault-tolerant' era, with small numbers of logical qubits running short error-corrected circuits, as the next step. Organizations investing in quantum skills should treat NISQ experiments as preparation for that hardware, not a source of near-term return on investment.
Real-world examples
At home
Marcus took a quantum computing course and ran a small algorithm on a real NISQ machine through the cloud. The results were noisy, and he had to run the program thousands of times to get a reliable answer.
At work
Nadia leads innovation at a chemical company. After a year of NISQ experiments, her team concluded that the hardware couldn't yet beat their classical simulations. They shifted their goal to building skills and preparing problems for future error-corrected machines.
How it compares
| feature | NISQ | Fault-tolerant quantum computing |
|---|---|---|
| Error correction | None or limited | Full error correction |
| Qubits | Tens to around a thousand noisy physical qubits | Many logical qubits built from many physical qubits |
| Program length | Short circuits | Long, complex computations |
| Can run Shor's algorithm at useful scale | No | Yes |
| Status | Current era | Targeted for the late 2020s and beyond |
NISQ machines are today's learning tools; fault-tolerant machines are the goal for practical advantage.
For analysts
Treat NISQ pilots as education, not return on investment. I'd fund small teams to learn the tools and identify future use cases, while being sceptical of any claim that today's hardware beats classical methods on business problems.
History & origin
Physicist John Preskill of Caltech introduced the term NISQ in a December 2017 talk, published in 2018. He argued that machines with 50 to a few hundred noisy qubits could explore new physics but wouldn't run algorithms like Shor's. Algorithms designed for NISQ hardware, such as the variational quantum eigensolver (proposed in 2014) and the quantum approximate optimization algorithm (2014), attracted heavy interest. Google's 2019 quantum supremacy experiment ran on NISQ hardware, and IBM's 2023 'quantum utility' experiment showed useful-looking results on a 127-qubit processor, though classical methods later matched some of them. In my view, NISQ was the right label at the right time, and its gradual retirement shows how fast expectations have shifted toward fault tolerance.