On October 28, 2026, PsiQuantum unveiled its first publicly disclosed optical circuit switch: an 8×8-radix demonstration vehicle built on the low-loss silicon photonics platform the company has spent nearly a decade developing for fault-tolerant quantum computing. This publication covered the device itself on November 2 — the 1.2 dB typical insertion loss, the 40 dB crosstalk, the sub-3-millisecond reconfiguration, all measured in PsiQuantum’s own labs, without optical amplification. This article is about something else: what it means for a quantum-computing company to become a switch vendor.

It is not a product launch — no product, no price, no named customer, no commercial port count. It is also not a pivot: co-founder and chief scientific officer Pete Shadbolt has said the company’s life’s work remains building a quantum computer. What it is, is a platform-redeployment story: a company with a mature, foundry-qualified photonics platform and no near-term quantum revenue putting that platform to work in the nearest paying market. The question is whether a platform built for single photons can become a vendor’s product line for AI clusters.

The announcement, precisely

The facts on the record are narrow, and the narrowness is the story. PsiQuantum’s first publicly disclosed optical circuit switch (OCS) is an 8×8 demonstration vehicle: eight inputs, eight outputs, strictly non-blocking, with no moving parts. Company lab testing reports typical results of 1.2 dB insertion loss, 40 dB crosstalk, and under 3 milliseconds of reconfiguration time, all without amplification and beating internal targets. Samples are going to select prospective customers; commercial products will be announced in the coming months. No customer is named, no power figure is published, no price is given, and no deployment is reported.

Two details in the release carry the commercial argument. The first is manufacturing: the chips come out of GlobalFoundries in New York — the same foundry supply chain behind PsiQuantum’s quantum photonic wafers — and the initial products use thermally tuned silicon nitride, a mature, foundry-compatible approach in which heaters adjust the optical path. The switch rides the same 300-millimeter wafer infrastructure the company has qualified for its quantum program: a genuine supply-chain advantage over a startup that must build its foundry relationship from scratch, and the strongest sentence in the announcement for anyone who has watched photonics companies fail on manufacturing rather than physics. There is a policy leg to the same fact: in September 2026, PsiQuantum signed definitive documentation with the U.S. Department of Commerce covering research, domestic manufacturability, and performance of its critical components, including high-performance optical switches. The agreement guarantees neither capacity nor customers, but it makes the New York production a policy credential as well as a supply-chain fact.

The second is the barium-titanate reservation. PsiQuantum has integrated barium titanate (BTO) at 300-millimeter wafer scale and uses a 64×64 BTO switch inside its quantum systems; applying a voltage changes BTO’s refractive index, steering light far faster than heating a waveguide can. It is explicitly not required for the demonstration vehicle or the initial products. The demo is a thermal device; BTO is the faster route held in reserve. Sensible sequencing — ship the manufacturable thing first — but it means the first products will switch at thermal speeds, in a market where competitors already claim sub-300-microsecond reconfiguration.

The leadership is the product strategy

The announcement’s most revealing sentence is not about the device. It is the framing from chief executive Victor Peng, the former Xilinx chief executive who joined PsiQuantum earlier this year: “This opportunity exists because we chose to solve the hardest scaling problems at the beginning of our roadmap to utility-scale quantum computing. As AI computing pushes infrastructure to its limits, we now have an opportunity to put our state-of-the-art technology to work today while continuing toward our ultimate goal” — the world’s first useful, fault-tolerant quantum computers.

Peng’s biography is part of the pitch: he ran Xilinx, which AMD acquired in 2022 for about $50 billion, and was then deeply involved in AMD’s AI and data-center business, before becoming PsiQuantum’s interim chief executive on February 10, 2026, and chief executive on July 15. In an interview with Semafor’s QuantumJane around the announcement, he called the switch “a multi-billion dollar opportunity” and said the company is “absolutely going after the opportunity in optical circuit switching.” That is a CEO talking about a market, not a scientist talking about a device — the tone of a company recruiting customers and partners, not merely publishing results.

The organizational move is equally telling: Rob Soderbery, executive vice president of the photonics and technologies business, was brought on about four months before the unveiling to run the optical circuit switching business. Soderbery is a Cisco and Western Digital veteran, hired with a mandate to accelerate the platform’s application across AI networking. A quantum-computing company does not hire a networking-industry executive to run a side project. It hires one because it intends to sell into a market where the buyers are hyperscalers and network-equipment vendors, and where the sales cycle, qualification process, and support model are nothing like a research program.

Soderbery has described Broadcom, Nvidia, Cisco, Marvell, and Arista as partners rather than rivals, on the grounds that circuit switching links their existing packet switches into larger systems rather than displacing them. That is a sensible framing — and the framing of a company that knows it cannot build the whole fabric itself. An 8×8 switch chip is a component; a data-center fabric is a system, and PsiQuantum owns exactly one layer of it. The partner framing is an admission of that fact, dressed as strategy.

The crowded field it enters

PsiQuantum is entering a market that spent the autumn of 2026 filling up. Three weeks before its announcement, on October 13, Voyant Photonics said it would develop a chip-scale OCS engine for AI data-center networks, adapting the silicon photonics switching architecture it industrialized for its FMCW LiDAR platform. Voyant’s model is deliberately narrower: rather than building a complete system, it plans to supply its OCS technology as an engine that optical networking companies incorporate into their own platforms — the component-supplier play, from a company whose switching architecture is already in production in a different application.

The two dedicated OCS startups moved faster and raised more. On October 14, Oxford-based Salience Labs announced an oversubscribed $125 million Series B, led by Bullhound Capital, with participation from AMD Ventures and Intel Capital. Salience has shown hardware at named milestones: a 32-port all-optical switch debuted in March 2026, and a 64-port version unveiled in September at ECOC 2026, with 128- and 256-port configurations in development. On September 2, iPronics announced its own $125 million Series B, co-led by Maverick Silicon and Light Street Capital, with participation from NVIDIA — and on October 26, three days before PsiQuantum’s announcement, iPronics put its co-founder and CTO, Dr. Daniel Pérez-López, in the CEO’s chair, framing the move as the shift from proving optical circuit switching to shipping it at scale.

The symmetry is instructive: two companies raised $250 million between them in two months, and both are further along the product curve than PsiQuantum — port counts in the dozens, reconfiguration claimed under 300 microseconds, an order of magnitude faster than PsiQuantum’s sub-3-millisecond figure, and a public roadmap to 128 and 256 ports. What PsiQuantum brings that neither startup has is a platform that already exists at manufacturing scale, qualified at a tier-one foundry, with a decade of process development behind it. The startups had to build their photonics and foundry relationships for this market. PsiQuantum spent a decade building both for a different market and is now redeploying them.

That advantage cuts both ways: a platform optimized for quantum computing may carry constraints — wavelength choices, process steps, packaging — that a purpose-built data-center switch would not. And the scaling path is the open question. According to Tom’s Hardware’s coverage, a 64×64 version has already taped out and a 72×72 design is underway, with the company conceding loss will rise somewhat on the larger parts. That concession is honest: loss grows with port count, because larger switches mean longer paths, more waveguide crossings, and more elements in series. The question is not whether 1.2 dB survives to 64 ports — it almost certainly does not, exactly — but whether the loss at 64 or 72 ports stays inside the 2 to 3 dB window that, as PsiQuantum told Semafor, a switch for AI data centers needs to meet. Salience is already at 64 ports. PsiQuantum is taping out its first one.

Above all of them sits the incumbent, and it is not a startup. Google is the only major hyperscaler currently using optical circuit switching in an AI data center, and it has done so for years. Its Apollo switches, the optical layer of the Jupiter network, steer light with MEMS mirrors across 136-port fiber arrays. Google’s own 2022 account credits the optical layer with a fivefold increase in network speed and capacity, a 30 percent reduction in capital expenditure, and a 41 percent reduction in power consumption while serving live production traffic. Google’s switches have 136 ports each, deployed in the tens of thousands. Every silicon-photonics entrant, PsiQuantum included, is arguing that a chip-scale switch can displace a proven MEMS system that already works at hyperscale — a high bar.

The market-size fog

Peng’s “multi-billion dollar opportunity” sounds precise and is not. The market-size figures circulating around this sector disagree by enough to matter, and the disagreement is worth reporting rather than resolving by picking a winner.

Citi recently forecast the OCS market could reach $11 billion by 2030, driven initially by Google deployments and potentially by growing adoption in NVIDIA-based AI infrastructure — a figure cited in Voyant’s October 13 announcement. Coverage of Salience’s round cited Cignal AI projecting the market above $8 billion by decade’s end. iPronics’ October 26 release claimed OCS is “projected to grow to tens of billions by 2030,” attributing the figure to no source at all. The range is not a contradiction to be settled by choosing the largest number: the figures may define the market differently, count different product categories, or simply reflect different optimism. What a reader should take away is the range and the uncertainty — OCS is a market credible observers size in the billions by 2030, with wide disagreement about how many. Any article that repeats “tens of billions” without its attribution, or Peng’s “multi-billion dollar opportunity” without noting it is a CEO describing his own company’s addressable market, is doing the company’s marketing for it.

What “becoming a vendor” actually requires

First, port count: an 8×8 device cannot be dropped into a fabric that rewires hundreds or thousands of endpoints, however good its loss figure; the 64×64 tape-out and the 72×72 design are the first steps, and their loss figures — not the demo’s — are the ones that will matter. Second, packaging: fiber attach, port density, and testability at volume have ended more photonics companies than any physics failure, and PsiQuantum has disclosed no yield, packaging throughput, or cost. Third, control software: a switch a hyperscaler’s network control plane cannot manage is a lab device, and no control or telemetry story accompanies the announcement. Fourth, qualification: hyperscalers qualify optical components over months, with reliability data — none of which is on the record. Fifth, the commercial apparatus itself: pricing, a sales channel, field support, none of which a research organization has in place on day one.

None of that is a prediction of failure. PsiQuantum’s manufacturing position is genuinely unusual for a new entrant: a decade of foundry-qualified, low-loss silicon photonics, built at GlobalFoundries, with a faster electro-optic material already operating in its quantum systems. That is more than a startup with a prototype and a slide deck. But it is also true that the company’s competitors have spent the same autumn demonstrating larger devices, raising larger rounds, and — in iPronics’ case — reorganizing their leadership explicitly around shipping. The honest summary is narrower than the release’s rhetoric and more interesting than a routine product note: a quantum-computing company with a real manufacturing platform has decided that platform’s nearest paying customer is the AI data-center network, and has hired the people and signed the agreements to go after it. The next announcement that matters will be the one with a port count, a loss figure at that port count, and a customer willing to be named.

What to watch

First, the commercial product announcement promised in the coming months: its port count, its loss figure at that port count, and its price will tell us how much of the demonstration survives contact with a product specification. Second, the first named customer or deployment: a switch in a hyperscaler’s lab is a different fact from a switch in a hyperscaler’s network, and the difference is the whole story. Third, the 64×64 device: whether its loss stays inside the 2 to 3 dB window practical deployment requires.

If those things arrive, PsiQuantum will have turned a quantum-computing supply chain into an AI infrastructure business. If they do not, the 8×8 demo will stand as a familiar pattern in photonics — a beautiful device, measured in a lab, waiting for a product to carry it. The physics, at 1.2 dB, is not the problem. Everything after the physics is.

Sources and further reading

PsiQuantum, “PsiQuantum Unveils Optical Circuit Switch Technology Built on Advanced, Low-Loss Silicon Photonics Platform,” October 28, 2026 — the primary source for the 8×8 demonstration vehicle, the 1.2 dB typical insertion loss, 40 dB crosstalk, sub-3-millisecond reconfiguration, the sampling status, the BTO distinction, and the September 2026 Department of Commerce agreement. Semafor’s reporting on the announcement, as summarized in QuantumJane’s October 28, 2026 coverage — for Victor Peng’s “multi-billion dollar opportunity” remark, Rob Soderbery’s role and partner framing, Pete Shadbolt’s statement that the data-center product is not a pivot, and Google as the only major hyperscaler currently using optical circuit switching in an AI data center. Tom’s Hardware coverage of the announcement, October 30, 2026, as reported in secondary coverage — for the 64×64 tape-out, the 72×72 design in progress, the GlobalFoundries manufacturing, and the thermally tuned silicon nitride construction. This publication’s own coverage: Arjun Mehta, “PsiQuantum Turns Its Quantum Photonics Into an AI Network Switch,” November 2, 2026 — the device-level account of the same announcement, including the 3 dB link-budget context and the Google Jupiter precedent; Marcus Chen, “iPronics Hands the Wheel to Its Photonics Founder as Optical Switching Goes to Production,” October 27, 2026 — for the iPronics $125 million Series B, the NVIDIA participation, the $177 million total funding, and the Salience Labs $125 million round with its 32-port and 64-port milestones. Voyant Photonics, “Voyant Photonics Advances into AI Data Center Optical Circuit Switching (OCS) with Production-Proven Silicon Photonics Technology,” GlobeNewswire, October 13, 2026 — for the chip-scale OCS engine plan, the FMCW LiDAR platform origin, the engine-supplier model, and the Citi $11 billion by 2030 forecast. Google, “Jupiter: Evolving a Datacenter Network” (ACM SIGCOMM 2022) — for the Apollo MEMS precedent, the 136-port design, and the fivefold speed and capacity, 30 percent capital-expenditure, and 41 percent power-consumption figures. The “above $8 billion by decade’s end” figure is attributed to Cignal AI as cited in coverage of Salience’s round; the “tens of billions by 2030” figure is iPronics’ own projection as stated in its October 26 release, reported here as a company claim. No customer names, deployment figures, revenue figures, power figures, pricing, or product port counts are asserted here, because none is disclosed in the cited sources. This article is analysis, not investment advice.