Dylan PatelFounder and Chief Analyst, SemiAnalysis · The semiconductor-AI stack

Positron Targets 16 Trillion Parameter Models

18 Apr 2026 · 1 min

Watch on YouTube

0:44

Positron is building hardware to run 16 trillion parameter models on a single server. The key is scaling memory capacity and bandwidth so chips don't need to talk to each other as much. Their OSMOvish product, coming next year, will support context windows in the millions of tokens.

Chapters

We really think that scaling that memory capacity is critical for longer context lengths, being able to support large models. We want to be able to support 16 trillion parameter models running on a single box. If you are able to have that much memory capacity and that high realized memory bandwidth per server, you know, we still have great chip-to-chip networking. We've got 72 112-gig lanes dedicated for inter-chip communication, but you rely less on that if you can do more per chip. When OSMOvish shipping next year, being able to support 16 trillion parameter models, you know, millions of tokens context length, is a very exciting differentiated capability.