Case Study: Ricoh achieves 10x faster Japanese AI inference with SambaNova SambaCloud

Ricoh runs Japanese AI models 10x faster with SambaNova Systems

Ricoh, a global Japanese digital services and electronics company, faced a challenge running its custom 70B-class AI models for Japanese businesses on existing GPU infrastructure. The infrastructure was too slow for production agentic workloads, achieving only tens of tokens per second. To overcome this, Ricoh turned to SambaNova Systems and its SambaCloud fully-managed cloud inference service.

By implementing SambaNova's SambaCloud, Ricoh achieved a 10x faster inference speed, going from tens to over 700 tokens per second on its 70B-class models. This performance boost reduced the time for complex agentic workflows from one minute down to just ten seconds, enabling Ricoh to efficiently scale its AI operations and deliver cutting-edge solutions.

Ricoh