PsiQuantum optical circuit switch: what was announced
PsiQuantum wants its quantum photonics to route traffic inside AI data centers. On October 28, the Palo Alto company unveiled an 8x8 optical circuit switch built on the silicon photonics platform it developed for utility-scale quantum computers, according to its press release.
The product is not for sale yet. PsiQuantum calls the device an 8x8-radix demonstration vehicle, and it is sending samples to prospective customers for further evaluation. The company says commercial products will be announced in subsequent months, but it has not disclosed pricing, volume or a launch date. Quantum Computing Report describes the work as a demonstration derived from the company's silicon photonics quantum architecture.
Supercomputing.news, citing Semafor reporting, says the switch is the first product PsiQuantum is developing. That report is one of the few pieces that frames the device as a product bet rather than a research milestone.
The company's pitch centers on bottlenecks. CEO Victor Peng tied the work to the quantum roadmap, saying PsiQuantum aims to "solve the hardest scaling problems at the beginning of our roadmap to utility-scale quantum computing." Rob Soderbery, an executive vice president at the company, made the networking case more directly: "optics will play an increasingly important role in enabling that growth."
How the 1.2 dB loss and 3 ms switching time compare
Company lab tests put the headline numbers at 1.2 dB typical insertion loss, 40 dB crosstalk and reconfiguration in under 3 milliseconds, all without optical amplification. The Quantum Insider and Dealroom repeat those figures. Insertion loss is the power lost as light passes through the device, so lower is better. Crosstalk measures how much light leaks between channels, and 40 dB means that leakage sits far below the signal.
The rival numbers are the more useful comparison. Hardware Busters reports that silicon photonics rivals Salience and iPronics offer 32-port parts that reconfigure in under 300 microseconds. Those parts depend on optical amplifiers, which PsiQuantum's device does not.
| Device | Ports | Reconfiguration | Optical amplification |
|---|---|---|---|
| PsiQuantum 8x8 demonstration | 8x8 | Under 3 ms | None reported |
| Salience and iPronics parts (per Hardware Busters) | 32-port | Under 300 microseconds | Depends on amplifiers |
The trade-off cuts both ways. PsiQuantum's reconfiguration ceiling is ten times the rival figure, and its switch has eight ports to the rival's 32. Its advantage is that it reports strong loss and crosstalk numbers without amplifiers. Those are company-reported results from lab tests, and the coverage reviewed includes no independent benchmark against the rival parts.
Why AI data centers are looking at optical circuit switches
An optical circuit switch routes light from one fiber to another without first converting it into an electrical signal. That can save power and latency in AI clusters, according to Hardware Busters and Qubitwire. Every conversion adds energy use and delay, so skipping one matters when large clusters need fast, efficient links.
Optical switching has a track record, though not in this exact form. The IEEE ComSoc Technology Blog notes that optical switching has proven useful in specific architectures, including Google's production data-center network. That gives PsiQuantum a precedent to point to, but the same analysis does not treat its new device as proven.
Speed is the limiting factor. Circuit switching changes paths more slowly than packet-level routing, so it suits connections that stay in place for longer intervals, according to Electronics Industry News. Traffic that shifts packet by packet will not fit that model. Buyers will need to map their traffic patterns before assuming an optical circuit switch belongs in the fabric.
Loss matters less than the headline suggests in some designs. Lumentum and Coherent both say standard transceivers can absorb about 3 dB of extra switch loss, per Hardware Busters. If that holds, the gap between 1.2 dB and a higher-loss switch may count for less than the marketing number implies, because the link budget has room to spare. For another sign of how seriously large chipmakers take quantum, see USA 24's coverage of Intel's $2.3 billion agreement to buy quantum software maker Classiq.
The Commerce Department agreement and the unconfirmed Nvidia target
The government angle may prove as important as the hardware. In September 2026, PsiQuantum signed definitive documentation with the US Department of Commerce to speed development and domestic manufacturing of components, including optical switches, according to Dealroom and The Quantum Insider. None of the coverage reviewed gives the value of that agreement.
USA 24 has reported that $389 million in CHIPS Act awards came with government equity stakes. That is a useful comparison for how Washington structures support for domestic manufacturing. Nothing in the PsiQuantum coverage, however, says the Commerce agreement includes an equity stake.
The Nvidia claim needs careful handling. Hardware Busters, citing Tom's Hardware, reports that PsiQuantum's loss figure falls under a 3 dB target in an Nvidia request for proposals. USA 24 could not confirm that request in a primary source, so treat the claim as unverified. If the request exists, a 1.2 dB result would clear that threshold with room to spare, which would matter more to buyers than any press release.
What buyers should verify before testing PsiQuantum samples
Data-center operators and networking buyers who gain access to samples should start with the numbers. The following steps reflect the guidance that matters most for a device at this stage:
- Ask for independent test data on insertion loss, crosstalk and reliability before evaluating any optical circuit switch, not just the company's lab figures.
- Check the loss against your own transceiver budget. The roughly 3 dB tolerance that Lumentum and Coherent describe is a starting point, not a guarantee.
- Measure reconfiguration time against your actual traffic. Under 3 ms may suit long-lived paths and fail for shorter ones.
- Ask what production port counts look like. An 8x8 demonstration vehicle is far from the 32-port parts already on the market.
Investors and analysts have a simpler task. They should track the commercial announcement, which the company says will come in subsequent months without a date.
Open questions and what to watch next
Every performance figure here is company-reported. Independent verification and customer evaluation are still pending, and long-term reliability at scale is unknown. One analysis cautions that the demonstration alone does not establish any benefit to an AI cluster, per Qubitwire.
There is also a strategic question. PsiQuantum is extending its platform into AI networking, but its quantum roadmap remains the stated ultimate goal. The sources reviewed do not quantify how much engineering effort shifts toward networking.
The next signals are customer evaluation results, the timing of the first commercial product and whether the Nvidia request for proposals surfaces in a primary source. The Commerce agreement's terms may also become clearer as manufacturing plans take shape. The real test is less the lab data sheet than a production cluster, and whether a customer reports that the switch held up there.
