Quantum

Quantum Computing Hardware

Physical Qubit Counts by Leading Machine

Largest published physical qubit count for each company's latest gate-model machine, colored by modality. Neutral-atom systems now scale to the most qubits, but raw count is not comparable across modalities and says little on its own about usable computation.

Neutral-atomSuperconductingTrapped-ionSilicon spinTopological

Best Two-Qubit Gate Fidelity by Company

Best reported two-qubit gate fidelity, the cleanest single quality metric. Trapped ions and silicon spin lead. Fault tolerance generally needs roughly 99.9 percent or better sustained across a full device.

Silicon spinTrapped-ionSuperconductingNeutral-atom

Qubit Modalities Compared

The competing physical approaches to building a qubit, each with different economics, scaling physics, and maturity. No modality has clearly won as of mid-2026.

Modality	Leaders	Scale today	Strength	Weakness
Superconducting	IBM, Google, Rigetti, IQM, OQC	100s of qubits	Fast gates, mature fab, deepest investment	Short coherence, needs dilution-fridge cooling
Trapped-ion	Quantinuum, IonQ	~100 high-quality qubits	Highest fidelity, long coherence, all-to-all connectivity	Slow gates, hard to scale raw counts
Neutral-atom	QuEra, Pasqal, Atom Computing, Infleqtion	1,000+ atoms	Fastest-scaling, room-temp cell, leads logical-qubit demos	Slower gates, atom loss and reload
Photonic	PsiQuantum, Xanadu, Quandela	~12-qubit class usable	Near room temperature, networking-native, fab-friendly	Probabilistic gates, photon loss, high overhead
Silicon spin	Intel, SQC, Diraq, Quobly	10s of qubits	Tiny footprint, 300mm CMOS-manufacturable	Small counts so far, trails on logical qubits
Topological	Microsoft	Unproven	Hardware-level error protection in principle	Scientifically contested, no confirmed topological qubit
Annealing	D-Wave	4,400+ qubits	Most commercially deployed, real on-prem installs	Optimization-only, not a universal computer
Diamond NV	Quantum Brilliance	Small	Room temperature, deployable at the edge	Tiny compute scale, strongest for sensing

Logical Qubits and Fault-Tolerance Roadmaps

Where each leader stands on error-corrected logical qubits today and what it has committed to deliver. Logical qubits, not raw physical counts, are the metric that gates fault tolerance, and target years are vendor-stated and have slipped before.

Company	Modality	Logical qubits today	Fault-tolerance target	Target
Quantinuum	Trapped-ion	48 error-corrected logical qubits (Helios)	Apollo: universal fault tolerance	2030
QuEra (with Harvard)	Neutral-atom	96 logical qubits	Large-scale fault tolerance	~2030
Atom Computing + Microsoft	Neutral-atom	28 logical qubits	Magne: 50 logical qubits	~2027
Google	Superconducting	Below-threshold error suppression (Willow)	Large-scale fault tolerance	~2030s
IBM	Superconducting	qLDPC prototype (Loon)	Starling: 200 logical qubits, 100M gates	2029
IonQ	Trapped-ion	~100 physical qubits (Tempo)	2M physical / 80k logical qubits	2030
PsiQuantum	Photonic	Manufacturable chipset (Omega)	1M-qubit fault-tolerant machine	2027

Logical-Qubit Demonstrations (Not a Leaderboard)

Published logical-qubit results. Counts are not comparable: encoding, error-detected versus error-corrected, and memory versus circuit all differ. Vendor claims are labeled. No invented counts.

Result	Logical count	Encoding	Status	Caveat
Harvard / MIT / QuEra	96 (d=4) active at once	High-rate [[16,6,4]] blocks on up to 448 atoms. Fig. 6h of the Nature paper.	Published (Nature, Jan 2026)	A cluster-state / teleportation architecture, not the same object as a concatenated error-corrected Helios count.
Quantinuum Helios	48 error-corrected	Vendor: 98 physical qubits, 2:1 concatenated encoding.	Vendor claim (Nov 2025)	The same launch also cites 94 error-detected logical qubits. Detection and correction are different metrics. Not independently reproduced at this specification.
Microsoft + Atom Computing	24 entangled; 28 in Bernstein-Vazirani	Distance-2 [[4,2,2]]-family codes on a 256-atom ytterbium processor.	Published (Nov 2024 blog + arXiv)	Distance-2 codes detect errors and can correct atom loss. They are not a below-threshold surface-code memory.
Google Willow	1 encoded memory qubit (d=7)	Surface code on 101 physical qubits (49 data + 48 measure + 4 leakage-removal).	Published (Nature, Dec 2024)	The result is exponential error suppression in a memory experiment, not a 100-logical-qubit computer.

Google Willow: Below-Threshold Error Suppression

Logical error per cycle on Willow, neural-network decoder. Distance-7 is the Nature headline (0.143 percent). Distance-3 and distance-5 are the paper's published subgrid means. This is a memory experiment, not a utility-scale machine.

Quantum Advantage and Supremacy Claims

Headline advantage claims and how they were received. IBM's 2023 Eagle result was classically surpassed. Google's 2025 Quantum Echoes paper is a Nature result and a company claim; it had not been reproduced on another machine as of contemporary reporting.

Claim	Who	When	Status	Note
Quantum supremacy (random-circuit sampling)	Google Sycamore	Oct 2019	Classical gap narrowed	Nature paper. Subsequent classical simulations reduced the claimed runtime advantage. A sampling task without a useful output.
Utility-scale kicked Ising on 127 qubits	IBM Eagle	2023	Classically surpassed	Nature 2023 framed a noisy-hardware simulation as beyond brute-force classical. Tindall et al., PRX Quantum, January 2024, produced a more accurate tensor-network result, including on modest hardware.
Verifiable advantage on a 65-qubit OTOC (Quantum Echoes)	Google Willow	Oct 2025	Company and Nature claim	Google reported about 13,000 times faster than Frontier on that physics benchmark. Not independently reproduced on another quantum processor as of contemporary reporting. Still a physics benchmark, not a production application.

Prefer Sterling in Google

Add sterlingcharts.com as a preferred source so our charts can appear with a preferred badge in Google Top Stories, AI Mode, and AI Overviews.

Add as preferred source