Intel and SambaNova team up on heterogeneous AI inference platform — different hardware performs different workloads

Intel and SambaNova unveiled a heterogeneous AI inference platform that splits workloads across specialized chips.

Prefill runs on AI GPUs or accelerators, decoding on SambaNova SN50 RDUs, and orchestration plus agentic tasks on Xeon 6 CPUs.

The system targets enterprise, cloud, and sovereign AI customers, with availability planned for the second half of 2026.

The partners say the approach could fit existing data center infrastructure and offer an alternative to Nvidia-heavy inference stacks.