Il n'existe pas « un » ordinateur quantique : il y a sept machines très différentes, construites par des équipes très différentes. Voici chacune d'elles en termes simples : comment elle fonctionne, quel est son pari, et qui la construit.
Why are there different kinds at all?
A qubit can be built from anything quantum you can control. A loop of superconducting wire, a single trapped atom, a particle of light. Each choice trades something for something else: speed versus quality, easy manufacturing versus easy scaling, working today versus working best someday. Nobody knows which recipe wins, so all of them are seriously funded. It's genuinely like the early days of flight: balloons, gliders and flapping machines, all in the air at once.
As you read the cards below, keep three questions in mind: how good are the qubits (fewer errors = less overhead), how fast can it grow (ten qubits is a demo; millions is the dream), and who's actually betting on it.
Superconducting circuits
Qubits made from tiny loops of superconducting wire, run in chandelier fridges colder than space.
Le pari: The most mature path (fast gates and chip-fab manufacturing) but needs extreme cold and heroic error correction at scale.
Qui le construit:
IBM IBMGoogle GOOGLRigetti RGTIAmazon (Ocelot chip) AMZNIQM privé
Most machines on the cloud today; error-corrected logical qubits improving.
Trapped ions
Individual charged atoms held by electromagnetic fields and steered with lasers.
Le pari: The highest-quality qubits and any-to-any connections. Gates run slower, and scaling means linking many traps together.
Qui le construit:
IonQ IONQQuantinuum majority-owned by Honeywell (HON)Alpine Quantum privé
Records for qubit fidelity; scaling architectures in progress.
Photonics
Qubits as particles of light traveling through optical chips.
Le pari: Light barely needs cooling and networks naturally. Photons hate interacting, so it leans hardest on error correction from day one.
Qui le construit:
PsiQuantum privéXanadu privéQuandela privéQuantum Computing Inc. QUBT
Betting on a leap straight to large fault-tolerant machines.
Neutral atoms
Uncharged atoms held in place by arrays of laser 'tweezers'.
Le pari: Scales to big, rearrangeable qubit arrays quickly. The race is pushing gate quality to match the leaders.
Qui le construit:
QuEra privéPasqal privéInfleqtion privéAtom Computing privé
Rapid array growth; strong error-correction demos since 2023.
Topological qubits
Exotic qubits designed to be error-resistant by their very physics.
Le pari: The hardest science with the biggest payoff if it works: potentially far fewer qubits wasted on error correction.
Qui le construit:
Microsoft (Majorana program) MSFT
Early but advancing; a deliberate high-risk, high-reward outlier.
Quantum annealing
Special-purpose machines that settle into low-energy answers for optimization problems.
Le pari: Thousands of qubits shipping today, but purpose-built for optimization, not general quantum computing.
Qui le construit:
D-Wave QBTS
Commercial for years; the 'useful now, narrower promise' bet.
Software & enablers
The picks-and-shovels layer: control systems, error suppression, algorithms, cloud access and quantum-safe security.
Le pari: Hardware-agnostic: this layer can win whichever qubit recipe prevails.
Qui le construit:
Nvidia (CUDA-Q) NVDAQ-CTRL privéClassiq privéStrangeworks privéArqit ARQQ
Grows with every machine shipped, regardless of the winner.
So… which one wins?
Honestly: nobody knows, and anyone who claims certainty is selling something. The scoreboard that matters isn't raw qubit counts. It's error-corrected logical qubits getting better, and real problems (especially chemistry) being solved. Watch those two things and you'll read every quantum headline like a pro.
Regard investisseur: chaque type de machine représente un profil de risque différent. La couche logicielle/facilitateurs est un pari sur l'ensemble du domaine plutôt que sur un seul gagnant. La même carte avec le cadre d'un investisseur se trouve sur le/la page investissement.
