Why would a quantum computing company build a network switch?
PsiQuantum is best known for its plan to build a large quantum computer. On October 28 it announced something plainer: an optical switch for the networks inside AI data centers.
The link is the technology underneath. PsiQuantum builds its quantum hardware on silicon photonics, which means chips that guide light. The announcement says the new switch is built on the company's low-loss silicon photonics platform.
One piece of its quantum hardware stays out of this device. PsiQuantum uses a 64×64 switch in its quantum systems. The company says that part is not required for the demonstration or the initial products announced.
What does a switch do that a cable doesn't?
A cable carries data from one place to another. A switch decides which place that is. An AI cluster needs both.
Think of a building with thousands of rooms, each holding a processor. The cables are the hallways. The switch is the set of doors that decides which hallway meets which room right now. It's a rough analogy, but it keeps the two jobs apart.
The job matters because AI systems link thousands of processors that trade results constantly. PsiQuantum's pitch is that lower latency keeps GPUs and other accelerators doing useful work. When data waits on the network, expensive chips wait with it.
Many switches take in light, convert it to electricity to decide where it goes, then convert it back to light. An optical circuit switch skips that step. PsiQuantum's product page says its switch "routes light directly from port to port, with no conversion." It describes the 8×8 device as strictly non-blocking, which means any input can reach any output, with no moving parts.
What did PsiQuantum actually show?
PsiQuantum showed a working 8×8 optical switch in its lab. The numbers are the company's own, and none are independently verified.
The announcement calls the device a "demonstration vehicle": eight light inputs and eight outputs. PsiQuantum reports typical results of 1.2 dB of insertion loss, 40 dB of crosstalk and reconfiguration in under 3 milliseconds. It says all three beat its initial targets "without amplification." The announcement's subheadline says loss "as low as 1 dB," so the 1.2 figure is the typical case.
| Measure | PsiQuantum's typical result | Plain-English reading |
|---|---|---|
| Insertion loss | 1.2 dB | About a quarter of the light is lost passing through one switch |
| Crosstalk | 40 dB | How much a signal leaks into the wrong port. As a ratio, 40 dB is 10,000 to 1 |
| Reconfiguration time | Under 3 ms | Changing a connection takes less than three thousandths of a second |
The right-hand column is simple arithmetic, not a PsiQuantum claim. It shows why loss gets so much attention. The product page warns that each switch "may add some loss, which accumulates across a network." If every switch matched the typical figure, a path through four of them would lose about two-thirds of its light.
On speed, IEEE ComSoc's technical blog says a switching time under 3 milliseconds could support some scheduled or recurring AI communication patterns, depending on the workload. The same article calls the numbers company-reported laboratory results, not independently verified benchmarks.
Nexairi Dispatch
Get the next AI move before it turns into a workflow problem.
Join the free newsletter for concise AI news, practical checklists, and the decisions practitioners need to make next.
Free. No spam. You will also get the Nexairi Dispatch.
How far is a lab demonstration from a product?
A long way. The company's own wording separates a lab demonstration from a product, and the table below follows that wording.
| Stage | What is stated | What is missing |
|---|---|---|
| Demonstrated in the lab | An 8×8 device with the typical results above | Independent measurements |
| Sampled | Samples shared with "select prospective customers" | Named customers and their test results |
| Announced as a product | Commercial products "in the coming months" | Price and ship date |
| Larger devices | "Actively developing larger devices" | Port counts and dates |
| Running in a data center | Nothing public | Any deployment at all |
Making the part is its own challenge. PsiQuantum's product page points to GlobalFoundries for high-volume manufacturing, and the announcement says it signed definitive documentation with the U.S. Department of Commerce in September to support research and domestic manufacturing of critical components, including optical switches. Nexairi's reporting on a $2 billion GlobalFoundries deal for AI chip parts shows how long the step from a signed deal to volume can run.
How is this different from Point2's cables?
Earlier this week Nexairi covered Point2 Technology, which sends radio-frequency signals through a plastic cable. It has raised a $136 million Series B and is preparing products for customer testing. The power and cost advantages it cites are the company's own claims.
Point2 is working on the path between equipment. PsiQuantum is working on the choice of path. Both aim at moving data through large AI systems, and neither announcement says it replaces the other.
Two readings of the same announcement
The supported reading is narrow. A company built a research platform for one purpose and has shown a device for a second market. The 8×8 switch works in its lab.
Everything past that is open. PsiQuantum may be seeking a nearer-term market for technology built for a longer-term goal. The switch may also become a niche product that sits beside existing networking. The sources don't separate those readings, and a named customer test would.
What would make this matter more?
Independent testing and a named customer would matter most. Larger devices, a price and a ship date would show whether it can scale.
In full, four pieces of evidence would change the picture:
- Independent measurements that match the figures above
- A named customer that has tested the samples, with results
- Larger devices, since a cluster needs far more than eight ports
- A price and a ship date
PsiQuantum says commercial products and updates on larger devices are coming in the next few months. That is the next point where the story can be checked.
What This Means for You
The upside: a light-based chip platform built for quantum computing may have a second use, and that could add a supplier to the networking layer of the AI buildout.
The risk: a lab result isn't a product. Loss adds up with every switch in a path, so one good number on one small device tells you little about a full network.
The next question: when an AI networking announcement says "optical switch," ask what job it does, how many ports it has shown and who outside the company has tested it.
Here is the plain read. A quantum company showed up with a small switch and strong lab numbers. That's worth knowing about. It is also the first step on a short ladder, and the next steps are the ones that need other people's proof.
Sources
- PsiQuantum: Unveils Optical Circuit Switch Technology Built on Advanced, Low-Loss Silicon Photonics Platform (October 28, 2026)
- PsiQuantum: Optical circuit switch product page
- IEEE ComSoc Technology Blog: PsiQuantum targets AI data center networks with silicon photonic optical switching
- Nexairi on Substack: Nvidia Backed a Startup Making AI Data Center Cables Out of Plastic
Related articles on Nexairi
Jim Smart is the founder and editor in chief of Nexairi. A Business Intelligence Developer with experience building data systems for Verizon, U.S. Army operations, and enterprise finance teams, Jim spent years turning complex data into decisions that executives could act on — dashboards, forecasting models, and automation pipelines across telecom and government contracting. He founded Nexairi to apply that same clarity to AI: making emerging technology understandable and actionable for the operators, accountants, and business owners who need it most. Jim holds GenAI certifications from the University of South Florida Bellini College of AI and completed Springboard's Data Science Career Track.
