Quantum computing developer PsiQuantum has announced a major technological pivot with potential massive implications for the artificial intelligence industry, revealing a new optical circuit switch built on its advanced low-loss silicon photonics platform. The breakthrough aims to address one of the most persistent bottlenecks in modern AI data centers: the energy-intensive and latency-inducing conversion of data between optical signals and electrical signals as information moves across massive computing clusters.

In a formal release issued earlier this week, the company showcased its novel approach to managing light within data centers. Traditional optical switches typically suffer from optical loss—a phenomenon where the intensity of light degrades as it traverses the switching mechanism. The more light a switch receives, the less ultimately reaches the receiving side, with signal degradation conventionally measured in decibels (dB). Standard data center optics can generally tolerate a signal loss of up to 3 dB, which equates to losing roughly half of the light passing through the hardware.

PsiQuantum’s newly debuted switch significantly outperforms this baseline. During initial laboratory demonstrations, the prototype registered a signal loss of only 1.2 dB, translating to a loss of approximately just one-quarter of the incoming light. Crucially, this high-efficiency performance was achieved entirely without the use of signal amplifiers, a component that typically adds complexity, power consumption, and potential points of failure to traditional data center architectures.

Despite the promising performance metrics, the initial prototype remains relatively modest in scale. The current iteration features an 8×8 configuration—meaning it handles eight inputs and eight outputs. In comparison, competing commercial switches utilized in massive hyperscale data centers frequently feature capacities ranging from 32 to upwards of 300 ports. However, PsiQuantum representatives have indicated that development teams are already actively working on scaling up the technology to build larger, enterprise-grade devices capable of meeting the rigorous demands of modern cloud providers and AI infrastructure operators.

PsiQuantum turns its photonics know-how into an optical switch for supercharging AI data center GPU networks — 8x8…

The core advantage of the new technology lies in the fundamental way optical circuit switches (OCS) handle data transmission. Unlike standard packet switches that require incoming optical signals from fiber-optic cables to be converted into electrical signals at every single routing hop—and then converted back into light before traveling to the next destination—an optical circuit switch passes light straight through from fiber to fiber. This continuous optical pathway eliminates the energy overhead and processing latency inherent in continuous optical-to-electrical-to-optical (O/O) conversions.

An OCS maintains a fixed, direct light path across the network fabric until a routing change is explicitly commanded by the system. Consequently, the primary engineering challenge and performance penalty of an optical circuit switch come down strictly to insertion loss—how much light manages to survive the transit through the microscopic pathways of the switch itself.

PsiQuantum’s proprietary design tackles this challenge by leveraging circuits that are precisely etched directly onto a silicon chip. Unlike alternative optical switching technologies—such as the microelectromechanical systems (MEMS) utilized by companies like Google and Lumentum, which rely on arrays of tiny, physical moving mirrors to redirect beams of light—PsiQuantum’s all-optical silicon photonics architecture features no moving parts whatsoever. This solid-state approach is designed to be integrated directly alongside conventional packet switches within existing data center topologies, acting as a high-speed, low-loss highway for massive blocks of data while leaving granular packet-level routing to traditional electronics.

The development represents a notable cross-pollination of quantum engineering and classical high-performance computing infrastructure. While PsiQuantum’s primary long-term corporate mission is the commercialization of a fault-tolerant, utility-scale quantum computer—a endeavor that inherently relies on advanced photonics to route and manipulate quantum bits, or qubits, through beams of light—the foundational manufacturing techniques and low-loss material science developed for quantum hardware are increasingly finding lucrative, near-term applications in the booming artificial intelligence sector.

As hyperscale data centers continue to struggle with soaring power densities, thermal management constraints, and physical networking bottlenecks driven by the explosive growth of large language models and distributed AI training workloads, innovations that reduce the energy cost of moving data from rack to rack are commanding intense industry attention. By translating its expertise in silicon photonics from the quantum realm into classical data center switching, PsiQuantum is positioning its technology to capture a share of the rapidly expanding market for high-speed, low-latency optical infrastructure.