SambaNova and Intel Reveal Heterogeneous AI Inference Blueprint Using Xeon 6 and RDUs

SambaNova Systems and Intel have announced a new heterogeneous AI hardware blueprint that integrates GPUs, SambaNova RDUs, and Intel Xeon 6 CPUs to optimize inference for agentic AI workloads, announced in a press release.

The system design uses GPUs for the prefill phase, SambaNova RDUs for decoding, and Intel Xeon 6 processors for agentic tools and system orchestration. SambaNova will standardize Xeon 6 as the host CPU paired with its RDUs, enabling a balanced architecture for high-throughput, low-latency inference. The production-scale system will be available to enterprises, cloud providers, and sovereign AI programs in the second half of 2026.

According to SambaNova, the Xeon 6 CPUs deliver more than 50% faster LLVM compilation times compared to Arm-based servers and up to 70% faster vector database performance than other x86 systems. This configuration aims to accelerate coding agent workflows by improving compilation and execution speeds.

Intel’s Xeon 6 processors will manage task coordination, workload distribution, and tool execution, while SambaNova RDUs will handle token generation. The collaboration marks the next stage in the companies’ partnership, moving from joint engineering to large-scale commercial deployment for enterprise and cloud AI infrastructure.