TechCrunch Startup News From invisibility cloaks to AI chips: Neurophos raises $110M to build tiny optical processors for inferencing
Jan 22, 2026
Neurophos is tackling energy-hungry AI with an optical chip designed for efficient inferencing. They discuss their origin from Duke University and the metamaterials research that fuels their innovations. With a $110M Series A funding led by Gates Frontier, they aim to enhance parallelism using tiny modulators, claiming significant performance gains over traditional GPUs. The podcast also highlights the challenges of manufacturing and the competitive landscape as Neurophos prepares to deliver its first chips by 2028.
AI Snips
Chapters
Transcript
Episode notes
Metasurface Optical Tensor Core
- Neurofos uses a metasurface modulator as an optical tensor core to perform matrix-vector multiplication for AI inference.
- Shrinking optical components lets thousands fit on a chip, increasing parallelism and efficiency versus silicon GPUs.
Big Series A And Early Customer Interest
- Neurofos raised $110 million in a Series A led by Gates Frontier with participation from Microsoft M12 and others.
- The company says it has signed multiple (unnamed) customers and attracted strong interest from Microsoft.
Photonics Tradeoffs And Claims
- Photonic chips can outperform silicon because light generates less heat and is less affected by temperature and electromagnetic fields.
- Traditional optics faced size and mass-production limits, but Neurofos argues its approach overcomes those barriers.