PsiQuantum Unveils Optical Circuit Switch for AI Data Centers

Introduction

PsiQuantum, known for its utility‑scale photonic quantum computer efforts, has unveiled an 8×8 optical circuit switching (OCS) demonstrator built on its low‑loss silicon photonics platform. The lab results—1.2 dB insertion loss, 40 dB crosstalk, and reconfiguration times under 3 ms—exceed the company’s own targets and do not require optical amplification. By repurposing technology originally engineered for fault‑tolerant quantum computing, PsiQuantum aims to address the mounting bandwidth and latency challenges of modern AI data‑center networks.

Why AI Data Centers Need Optical Switching

AI workloads are growing faster than the underlying compute infrastructure, creating massive data‑movement demands across large clusters of GPUs, TPUs, and custom ASICs. Traditional electrical interconnects struggle with power, latency, and scaling limits, while optics offers:

• Higher bandwidth density – multiple terabits per second per fiber.
• Lower power per bit – reduced electrical conversion losses.
• Scalable topology – circuit‑switched fabrics can reconfigure paths quickly to match dynamic traffic patterns.

PsiQuantum’s OCS leverages these advantages, promising a more efficient backbone for AI clusters that must shuffle petabytes of training data and model parameters.

Technical Highlights of the Demonstrator

| Metric | Demonstrated Value | Significance | |--------|-------------------|--------------| | Insertion loss | 1.2 dB | Keeps signal power high, minimizing the need for repeaters or amplifiers. | Crosstalk | 40 dB | Ensures clean channel separation, critical for error‑sensitive AI workloads. | Reconfiguration time | < 3 ms | Allows near‑real‑time traffic engineering, supporting bursty AI training phases. | Platform | Low‑loss silicon photonics (300 mm wafer) | Benefits from mature semiconductor manufacturing, enabling volume production.

The switch does not rely on barium titanate (BTO) integration for this version, highlighting the robustness of the silicon photonics core while preserving a path to even faster future devices.

Path to Commercialization

PsiQuantum is currently distributing demonstration samples to a select group of prospective customers. The company plans to evolve the 8×8 demonstrator into larger‑scale, commercially viable OCS products for AI data centers. Key steps include:

1. Scaling radix – moving from 8×8 to higher‑port counts (e.g., 16×16 or 32×32) to match larger fabric requirements.
2. Integration with existing data‑center fabrics – providing interfaces compatible with common Ethernet and InfiniBand standards.
3. Reliability testing – validating long‑term operation under data‑center temperature and power conditions.
4. Supply‑chain development – leveraging the recent U.S. Department of Commerce agreement to secure domestic manufacturing of critical photonic components.

By building on a platform originally designed for quantum‑computing error correction, PsiQuantum can offer performance margins that exceed typical AI‑centric optical switches, potentially reducing the need for over‑provisioning and lowering total cost of ownership.

Practical Takeaways

• For data‑center architects: PsiQuantum’s OCS could become a drop‑in solution for high‑speed, low‑latency interconnects, especially where traffic patterns shift rapidly during model training.
• For AI developers: Faster, more reliable data movement may shorten training cycles and enable larger model experiments without hitting network bottlenecks.
• For investors and industry watchers: The move illustrates how quantum‑computing R&D can generate near‑term revenue streams, diversifying PsiQuantum’s business while it pursues its long‑term quantum‑computer goal.

--- PsiQuantum’s optical circuit switch showcases how a quantum‑focused silicon photonics platform can immediately benefit AI infrastructure, offering low loss, high isolation, and rapid reconfiguration—key attributes for the next generation of data‑center networks.