Nisq Era Limitations

NISQ stands for Noisy Intermediate-Scale Quantum, the name John Preskill gave in 2018 to every quantum computer built so far: too big to simulate trivially on a classical machine, too error-ridden to run the deep, fault-tolerant algorithms quantum theory promises. The four constraints below are not independent quirks of one bad chip — they compound into a single ceiling that IBM, Google, IonQ, and Quantinuum are all racing to push past from different directions, and understanding how the constraints interact explains why "quantum advantage" claims keep getting walked back the moment a classical team sits down and tries harder.

Noise limits circuit depth: every gate on a NISQ chip has a nonzero error rate, typically 0.1-1% per two-qubit gate even on today's best hardware, and those errors compound multiplicatively rather than canceling out. A circuit with 20 sequential layers of two-qubit gates at 99.5% fidelity each already has roughly a 10% chance of a wrong answer purely from gate noise, before accounting for decoherence while qubits sit idle waiting their turn. That arithmetic is why useful NISQ algorithms such as the variational quantum eigensolver and QAOA keep their circuits shallow by design, trading circuit depth for many repeated short runs whose results get averaged classically — a workaround, not a fix, because the workaround itself caps how large a problem instance is worth attempting.

Qubit count limits problem size: IBM's Condor has 1,121 physical qubits, Google's Willow generation runs just over 100, IonQ's and Quantinuum's trapped-ion systems sit in the tens, and every one of those numbers is counting physical qubits, not the logical qubits a real algorithm needs. Shor's algorithm at a cryptographically relevant scale is estimated to need on the order of 20 million physical qubits once error correction overhead is included — a gap of four to five orders of magnitude from where IBM Condor sits today. Raw qubit count is also not simply additive with usefulness: connectivity, coherence time, and gate fidelity all have to scale together, which is why IBM's own roadmap talks less about qubit count past Condor and more about modular chip-to-chip linking (its Heron generation) to grow capability without each chip's noise floor getting worse.

Error correction is not yet scalable: the surface code, the leading candidate for fault-tolerant quantum computing, needs roughly 1,000 to 1,500 noisy physical qubits to encode one reliable logical qubit, and nobody has run that encoding at the scale a real algorithm requires. Google's Willow chip demonstrated the core result the field needed in 2024 — below-threshold error correction, meaning a bigger surface-code patch actually reduces logical error rate instead of just adding more places for noise to leak in — but that was shown on a few logical qubits, not the hundreds a useful chemistry or cryptography calculation would need simultaneously. Going from "we proved the physics works" to "we have a fault-tolerant machine" is itself a multi-year, multi-generation engineering program, which is why IBM's own public roadmap pushes its first fault-tolerant system, Starling, out to 2029.

Classical simulation can still handle small circuits, and that single fact is the whole reason "quantum advantage" is a contested claim rather than a settled one. Below roughly 50 qubits and shallow depth, a classical supercomputer can often simulate a quantum circuit exactly using tensor-network methods, and even well above 50 qubits, clever classical algorithms have repeatedly closed gaps that looked unbridgeable at first announcement — IBM's own team showed a classical simulation could approximate Google's 2019 Sycamore sampling result far faster than Google's original estimate. That cat-and-mouse dynamic between quantum hardware claims and classical algorithmic counter-moves is exactly the ladder this term's companion entry on quantum advantage walks through, because every NISQ-era limitation above — shallow circuits, modest qubit counts, no scalable error correction — is precisely what keeps leaving room for classical computers to catch up.

The practical reading for anyone evaluating a quantum computing claim today is to ask which of these four walls the result actually got past — a deeper circuit than before, a qubit count that still fits a workstation's simulation budget, or a genuinely error-corrected logical qubit — because headlines rarely specify which wall moved, and the ones that haven't moved are the ones that still make "quantum advantage" a debate instead of a fact.