PsiQuantum shows an optical switch for AI infrastructure
A component developed from its photonics platform enters selected customer sampling. Commercial system performance remains to be established.
PsiQuantum has announced an 8-by-8 optical circuit switch demonstrator aimed at AI infrastructure. The company reports typical laboratory insertion loss of 1.2 dB, crosstalk performance of 40 dB and reconfiguration below 3 milliseconds without amplification. Samples are being shared with selected prospective customers; commercial products remain planned.[1]
The announced demonstrator should also be distinguished from the company's separate 64-by-64 barium-titanate switch for quantum systems. PsiQuantum says BTO is not needed for this demonstration vehicle or its initial products.[1]
The appealing question is whether difficult component engineering for quantum machines can find useful work sooner in conventional computing. An optical switch can be interesting on that basis without being a quantum processor or proving a quantum computational advantage.
A production evaluation would need to look beyond a few headline component numbers. Which connections were measured? How consistent is performance across manufactured devices? What happens during reconfiguration, and which workloads tolerate that interval? How do control, packaging, reliability and total cost change the answer?
Those questions frame the next stage of the story. A promising demonstration can justify customer evaluation. Evidence from that evaluation would determine whether the component improves the systems it is intended to serve.