PsiQuantum's 8x8 optical switch loses 1.2 dB in AI data center demo

PsiQuantum's 8x8 optical circuit switch loses a typical 1.2 dB of light in lab tests — about 24% of the signal — without any amplification, the quantum computing company announced in a release earlier this week. The figure sits well under the 3 dB that standard data center optics tolerate and under the 2–3 dB ceiling the company told Semafor its switch needed to be workable.

The prototype passes light straight through, fiber to fiber, without converting it to or from electrical signals. Regular data center switches make that conversion at every hop, consuming power and adding latency. PsiQuantum's switch uses circuits etched on a silicon chip with no moving parts, unlike the MEMS mirror designs from Google and Lumentum, and would run alongside regular packet switches.

How does it compare with rival switches?

The port count is the weak point. Eight inputs and eight outputs trail competitors' 32 to 300 ports:

• PsiQuantum 8x8 demo: 1.2 dB typical loss (~24% of light), under 3 ms reconfiguration, silicon photonics, no amplification, no moving parts.
• Google Palomar: 136x136, typically under 2 dB (under 37% lost), millisecond MEMS mirrors.
• Lumentum R300: 300x300, close to 1.5 dB (~29% lost) typical at start of life, MEMS mirrors.
• Salience Labs OC-32M: 32x32, under 2 dB after built-in amplification, under 300 µs, silicon photonics.
• iPronics ONE-32: 32 ports, loss not published, under 300 µs, silicon photonics with optical amplifiers.

The PsiQuantum chip is "strictly non-blocking," so a new connection never disturbs existing ones, and "every input can reach every output." Lab measurements also showed 40 dB of crosstalk — about 0.01% of light leaking to other ports — and under 3 ms reconfiguration, all exceeding "initial targets set for these devices." The loss figure is measured fiber to fiber, including getting light onto and off the chip but not connector losses, PsiQuantum told Tom's Hardware Premium.

Where do the low losses come from?

The platform descends from quantum computing hardware, which tolerates only limited light loss. PsiQuantum's 2025 Nature paper reports silicon nitride waveguides losing as little as 0.5 dB per meter (~11% of light per meter), splitters at 0.5 millidecibels and crossings at 1.2 millidecibels — the last two a few hundredths of a percent of light or less. Chip-to-fiber coupling has reached about 60 millidecibels, roughly 1.4% of light lost, in the best published lab results, though without a full fiber array attached. Those numbers set a best case, not a worst case, for a 64-port switch, since losses add up as ports grow.

Acceptable loss depends on the optics at each link's end. Coherent says an OCS can add "up to 3 dB additional insertion loss" and its transceivers still work; Lumentum told Gazettabyte that 3 dB suffices with off-the-shelf FR4, DR4 and DR8 modules covering 500 meters to 2 kilometers. PsiQuantum's 8x8 targets the O-band those modules use. The prototype comes in below the 3 dB target in a request for proposals from Nvidia, the company says.

What comes next?

A 64x64 version is taped out and a 72x72 is underway, with PsiQuantum "actively developing larger devices." It declined to share future product details but says loss, possibly slightly higher, will stay very low.

Speed matters less than loss today. Google, with deployments in the "tens of thousands," uses its switches to resize TPU pods or route around failed chips on millisecond timescales. PsiQuantum's under 3 ms fits that class; Salience and iPronics claim under 300 microseconds. For speed, PsiQuantum holds barium titanate (BTO), a material in an existing 64-port quantum-system switch needing gigahertz switching — roughly a million times faster than the demo. BTO bends light under an applied voltage; PsiQuantum chose it over lithium niobate for the strongest effect. It is "not required for the demonstration vehicle or initial products," which use thermally switched silicon nitride, but is "an additional path to faster, more advanced future products."

Building thousands of switches for conventional applications reduces packaging, assembly and supply-chain risk for the company's quantum parts. AI and quantum computing are "complementary technologies," PsiQuantum says. The July hire of Rob Soderbery as executive vice president was framed as "accelerating its application across AI networking and other next-generation computing infrastructure," and the company's CHIPS award, finalized in September, covers "high-performance optical switches" among other components.

Co-founder Pete Shadbolt praised Google's MEMS switch but said he would prefer similar-quality switching on a monolithic chip with no moving parts. Nvidia holds a stake through NVentures and collaborates on "GPU-QPU integration" and silicon photonics; it also joined rival iPronics' $125 million round in September for a second-generation switch.

Is the market ready?

Cignal AI expects most OCS deployments to stay inside Google through the end of the decade; outside Google, the technology remains at "proof-of-concept and early trial." The firm raised its market estimates this year roughly threefold for 2026 and by more than 40% for 2029. PsiQuantum says samples and testing kits have reached several computing companies, including GPU vendors and hyperscalers, with commercial products to be "announced in the coming months."

Salience has already demonstrated 64 ports at this year's European Conference on Optical Communication, with 128- and 256-port products on its roadmap. PsiQuantum's figures remain unverified. If they hold at 64 ports without amplifiers, larger monolithic optical circuit switches are on the way.