AI startup Neurophos raises $110M to develop optical AI chips

What happened Neurophos announced a $110 million Series A on January 22, 2026, to advance photonic AI accelerators, according to the company press release and coverage in TechCrunch and SiliconANGLE. The round will fund R&D to move lab prototypes toward production optical processing units (OPUs).

Technical details Neurophos describes an OPU that places more than one million photonic processing elements—modulators and detectors—on a single chip. The company aims to deliver up to 100× higher inference throughput compared with today's electronic AI processors while reducing energy per operation. The architecture targets exaflop-scale photonic performance through dense on-chip optics and parallel analog computation.

Implications If realized, these chips could move inference-heavy workloads—recommendation systems and multimodal pipelines—to specialized photonic accelerators. Systems integration, software toolchains, and fabrication maturity remain gating factors. Watch for prototype timelines, partner integrations, and independent benchmarks before procurement decisions.

Why It Matters

• OPUs promise much higher inference throughput and lower energy per inference—plan pilot tests for latency-sensitive and production models to measure real gains.
• Expect new software and driver layers: budget integration work for model quantization, analog-to-digital conversions, and updated toolchains.
• Early adopters can offload GPU clusters for inference, but require independent benchmarks, partner integrations, and proven fabrication readiness before large-scale migration.