Pleco
Photonics

GPUs are extraordinary at training and high-throughput batch inference. They are structurally limited at three things: latency floors below one millisecond, deployment outside climate-controlled facilities, and physical-system workloads where the inference loop has to close in real time.

The next decade of machine learning deployment runs into all three.

Pleco's processor uses photonic computation — light moving through configurable optical paths — to deliver inference latencies and deployment profiles that are physically unavailable to electronic accelerators.

The architecture is validated across the full target workload set in simulation. The first hardware prototype is in build.

• Provisional patent filed April 14, 2026 (USPTO).
• Four computational modes validated in simulation: GKSL dynamics, analog matrix multiply, Turing-equivalent sequential computation, reservoir computing.
• Full neural network inference validated on the mesh substrate (8/8 scorecard).
• Fiber breadboard prototype in build — first physics validator.
• Silicon MPW (AIM Photonics) planned for Q4 2026 / Q1 2027.

• 2026: fiber breadboard operational. First commons-node deployment.
• 2027: silicon MPW tapeout. Non-provisional patent filing.
• 2028–2030: flagship photonic compute card. Federated mesh deployments.