Single-photon qubits versus squeezed states

The real split is not photonics versus non photonics, it is what each company treats as the basic unit of computation. PsiQuantum is trying to make light behave like standard digital bits, where each qubit is encoded in individual photons moving through waveguides, switches, sources, and detectors on silicon chips. Xanadu starts from squeezed light, where information lives in continuous properties of the optical field, then uses measurements and error correcting encodings to turn that analog resource into computable qubits.1 2 3 4 5

• PsiQuantum’s bet is manufacturing discipline. Its platform integrates single photon sources, superconducting single photon detectors, and photonic circuits on 300 mm silicon photonics wafers made with GlobalFoundries. That fits its goal of scaling to very large fault tolerant systems by using existing semiconductor fabs and telecom style fiber links between chips.1 2 3 6
• Xanadu’s workflow looks different at the hardware level. Lasers generate squeezed states, interferometers mix them, and detectors read out continuous variable measurements. In Xanadu’s own roadmap, those continuous variable states are the raw material for GKP style encoded qubits, which is a more layered path from optical physics to logical computation.4 5 7
• That makes the competition less about qubit counts and more about where each team places its hardest engineering problem. PsiQuantum concentrates difficulty in deterministic chip manufacturing and low loss photonic interconnects. Xanadu concentrates difficulty in generating high quality squeezed states, controlling large interferometers, and converting those states into fault tolerant logical qubits.1 2 4 5 7

Going forward, photonic quantum computing is likely to divide into two winning tests. PsiQuantum needs to prove that foundry built single photon modules can be assembled into a large error corrected machine. Xanadu needs to prove that continuous variable optics can be promoted into logical qubits efficiently enough to support broad fault tolerant workloads. The company that closes that translation gap first will define the category.1 2 3 5 7