Observed arrival · 2026-11-21
Neurophos’s T100 OPU: photonics at AI-rack scale
Credibility concern recorded. The source reference remains available for verification and correction.
Neurophos presents a photonic AI processor and claims it can deliver substantially more inference compute with less power than a GPU.
- For
- AI hardware engineers and infrastructure planners
- Worth noticing
- The comparison specifies sustained prefill inference at INT8; the captured page also displays a “0× SMALLER” figure.
Field notes
The comparison labels its workload as sustained prefill inference at INT8 and says it draws on public workloads and NVIDIA B200 datasheet values. The FAQ describes an optical core paired with an electronic die for memory and I/O, and says the design is intended for standard server-rack integration. The page lists a 2028 shipping target; its performance figures remain company claims in the supplied evidence.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue