Story details

Japan’s NEDO selected Preferred Networks and Tenstorrent for a joint xPU design project incorporating photonics-electronics integration. Their five-year research programme targets processor systems for training and inference across scientific, physical and agentic AI workloads.

The announced work divides into system integration and packaging, AI chiplets using stacked memory (led by PFN), and open-architecture CPU chiplets supporting confidential computing (led by Tenstorrent). Commercialization is a later aim, not a current product launch.