Intel, SambaNova Unveil Agentic AI Infra Blueprint Combining GPUs, RDUs, CPUs

Intel and SambaNova have announced a joint blueprint for agentic AI infrastructure, combining GPUs, SambaNova RDUs, and Intel Xeon 6 processors to support enterprise and cloud deployments. 

The companies said the design will be available in the second half of 2026 for enterprises, cloud providers, and sovereign AI programmes.

The collaboration addresses limitations of GPU-only inference systems by assigning distinct compute roles across the AI pipeline. GPUs will handle prefill, RDUs will manage high-throughput decoding, and Xeon 6 CPUs will act as host and execution processors for agentic workloads.

“Agentic AI is moving into production, and the winning pattern we’re seeing is GPUs to start the job, Intel Xeon 6 to run it, and SambaNova RDUs to finish it fast,” Rodrigo Liang, CEO of SambaNova, said in a statement. He added that the system can run in existing air-cooled data centres and supports widely used software tools.

Agentic AI has moved from testing to production, with coding agents now executing tasks such as compiling code, calling APIs, and coordinating workflows. The companies said this shift has exposed constraints in GPU-only systems, especially in real-world deployments that demand low latency and high efficiency.

The architecture assigns each inference stage to a specific component. GPUs process input prompts and store them in key-value caches. SambaNova RDUs generate outputs through high-throughput decoding. Xeon 6 CPUs coordinate tasks, execute code and manage system-level operations.

Kevork Kechichian, Executive Vice President at Intel, said the x86 ecosystem remains central to data centre software. “Workloads of the future will require a heterogeneous mix of computing, and this collaboration with SambaNova delivers a cost-efficient, high-performance inference architecture,” he said.

According to the statement, Xeon 6 improves compilation and database performance in coding workflows, helping developers move faster from development to deployment. The system also maintains compatibility with existing x86-based software environments used in enterprise data centres.

Ian Cutress, CEO of More Than Moore, said enterprises now prefer systems that balance performance and efficiency across different compute types rather than relying on a single chip architecture.