# PsiQuantum Turns Its Quantum Photonics Into an AI Network Switch, Starting With a 1.2 dB 8×8 Prototype > Source: > Published: 2026-10-30 16:00:00+00:00 **PsiQuantum has built an 8×8 optical circuit switch on the same silicon photonics it uses for its quantum computers, and its lab numbers already clear the loss budget Nvidia asked for. The catch is that eight ports is a long way from what a GPU cluster needs.** PsiQuantum has spent close to a decade building low-loss silicon photonics for a fault-tolerant quantum computer that does not exist yet. This week it found a nearer customer for that work: the networks inside AI data centers. On October 28 the company showed an optical circuit switch (OCS) demonstration chip with eight inputs and eight outputs, and says commercial products will be announced [in the coming months](https://thequantuminsider.com/2026/10/28/psiquantum-optical-circuit-switch-ai-data-centers/). ## What the prototype does An optical circuit switch steers light from one fiber to another without turning it into an electrical signal first, which saves the power and latency of an extra conversion. PsiQuantum’s part is strictly non-blocking, so any input can be routed to any output, and it has no moving parts. The company’s lab figures: - **Insertion loss:** 1.2 dB typical, fiber-to-fiber, with no optical amplification - **Crosstalk:** 40 dB, meaning roughly 0.01% of the light leaks into other ports - **Reconfiguration:** under 3 ms Loss is the number that decides whether anyone buys this. At 1.2 dB, about a quarter of the light disappears on the way through. Coherent and Lumentum both say standard data center transceivers can absorb around 3 dB of extra loss from a switch, and [Tom’s Hardware reports](https://www.tomshardware.com/tech-industry/photonics/psiquantum-turns-its-photonics-know-how-into-an-optical-switch-for-supercharging-ai-data-center-gpu-networks-8x8-lab-prototype-loses-less-light-than-rivals-with-plans-to-develop-larger-devices) that PsiQuantum lands well under the 3 dB target set in an Nvidia request for proposals. Nvidia also happens to be an investor, through its NVentures arm. ## Up against mirrors and amplifiers Google is the reference customer for this whole category. It runs MEMS optical switches, essentially arrays of tiny tilting mirrors, in the tens of thousands, and uses them to reshape TPU pods and route around failed hardware. Its Palomar switch handles 136×136 ports at typically under 2 dB, while Lumentum’s R300 MEMS unit reaches 300×300 at about 1.5 dB. Silicon photonics rivals Salience and iPronics offer 32-port parts that reconfigure in under 300 µs, roughly ten times faster than PsiQuantum’s demo, but they lean on optical amplifiers to make up for their losses. PsiQuantum’s pitch, then, is MEMS-class loss on a monolithic chip with no mirrors and no amplifier. Port count is the open question. Losses stack up as a switch grows, and an 8×8 die says little about how a 64-port device will behave. According to Tom’s Hardware, a 64×64 version has already taped out and a 72×72 design is underway, and PsiQuantum concedes loss will rise somewhat on the larger parts. ## A quantum spin-off, made at GlobalFoundries The chips come out of [GlobalFoundries in New York](https://techblog.comsoc.org/2026/10/28/psiquantum-targets-ai-data-center-networks-with-silicon-photonic-optical-switching/), the same supply chain behind PsiQuantum’s quantum photonic wafers. The first switches use thermally tuned silicon nitride. Further out, the company has barium titanate (BTO) integration running on 300 mm wafers and already uses a 64×64 BTO switch in its quantum systems. BTO’s strong electro-optic effect is the route to switching far faster than today’s millisecond-class demo, though PsiQuantum says the initial products will not need it. “This platform can deliver exceptional performance for optical circuit switching for AI computing,” said executive vice president Rob Soderbery. CEO Victor Peng tied it back to the quantum program, saying the company “chose to solve the hardest scaling problems at the beginning of our roadmap.” The market timing makes sense. Analyst firm Cignal AI roughly tripled its 2026 OCS estimate earlier this year, even as it expects most deployments to stay inside Google through the end of the decade, with everyone else still at the trial stage. PsiQuantum has sent samples to GPU vendors and hyperscalers. Whether a quantum computing company can turn a lab-grade 8×8 into a 64-port product that survives a production data center is what its next announcement has to show.