PsiQuantum is sharing samples of an 8×8 optical circuit switch with selected prospective customers, opening an AI-networking application for photonics developed for quantum computing. For infrastructure teams, this is a component-evaluation milestone: the announcement does not provide a measured improvement in AI training time.
The primary announcement is dated October 28, 2026. PsiQuantum says commercial products will be announced in the coming months; that statement does not establish a shipping date.
Read the optical measurements in their own units
PsiQuantum reports typical insertion loss of 1.2 dB and reconfiguration time below 3 milliseconds in its own laboratory tests, without amplification. Insertion loss describes optical power lost through a component. Reconfiguration time describes how long it takes to change a connection path; it is not the speed at which data travels along an established connection.
Lower loss can leave more optical power available for the rest of a link. Whether that is enough for a particular network depends on the other components and paths involved. Neither figure, on its own, establishes GPU utilization, application throughput or a reduction in a facility's electricity bill.
The company also says barium titanate integration is unnecessary for this demonstrator and its initial products. Readers should keep those results separate from the company's other photonic-switch technologies.
The network has to decide when to change its paths
Optical circuit switching already has a history in data centers. Google's Jupiter Evolving paper from SIGCOMM 2022 describes optical switches working alongside centralized software-defined control, traffic engineering and automated operations.
That context explains the integration question. A switch can make a path available, while the network's control system must choose a topology and manage traffic through changes. Google's experience is evidence about its own architecture, not an independent test of PsiQuantum's hardware.
The related Mission Apollo paper describes a balance among port count, optical performance, switching time and cost, together with transceiver design. Improving one component metric leaves those system requirements to be evaluated.
What a customer sample can help establish
A useful evaluation would put the sample into the intended optical link and examine the behavior of the network during path changes. It would also ask what commercial configuration, support and supply terms the eventual product offers.
PsiQuantum's announcement creates an opportunity to begin that work. Its laboratory figures and sampling status are concrete developments; a deployment decision still needs evidence from the intended system.
Sources
Related Articles
Artificial Analysis v5 changes the yardstick for choosing AI models
The updated Intelligence Index changes its tasks and weights. Teams choosing a model need version-aware comparisons, matching effort settings and their own cost evidence.
Anthropic's IPO filing ties much of its $80.7 billion loss to valuation changes
The public S-1 reports a much smaller operating loss and positive operating cash flow for the same six months. These measures explain different parts of the business.
Apollo's AI payoff claim hinges on how productivity is measured
Torsten Slok's October 24 note questions broad AI returns. Its index comparisons and productivity charts are useful context, but they do not measure a particular company's deployment.
Google Personal Intelligence can carry connected-app context into ads
Google's October policy and Search help allow derived context from connected apps to inform ads. Reviewing connections, ad settings and saved history involves separate choices.