Intel SambaNova AI Inference with Xeon 6

Agentic AI is rapidly transitioning from research labs to real-world applications, prompting data center architects to reconsider how inference workloads are managed. In response to this shift, Intel and SambaNova Systems have unveiled a new heterogeneous architecture tailored to support the next wave of AI applications.

The collaboration between the two companies aims to combine GPUs, SambaNova RDUs, and Intel Xeon 6 processors to strike a balance between performance, efficiency, and compatibility. This development is particularly significant for engineers and system architects in the AI infrastructure space, signaling a potential departure from reliance solely on GPU-centric approaches.

Heterogeneous Approach for Agentic AI

The novel architecture is geared towards addressing the evolving demands of emerging "agentic AI" workloads, which involve systems capable of autonomous reasoning, planning, and execution of complex actions. These workloads necessitate varying computing resources at different stages of the inference process.

According to the blueprint, GPUs are leveraged for the prefill stage, where models process extensive prompts. Subsequently, SambaNova’s RDUs handle high-throughput decode operations, while Xeon 6 processors serve as host and action CPUs, managing tasks and executing application logic.

This approach aligns with a prevailing industry trend of matching specific AI inference phases with the most suitable hardware accelerators, moving beyond a reliance solely on GPUs. The design also ensures compatibility with the prevalent x86-based software stack that forms the backbone of many data center environments.

Leveraging the x86 Ecosystem

Ensuring compatibility with existing infrastructure is a pivotal aspect of the collaboration between Intel and SambaNova. Given that data centers and enterprise AI deployments heavily rely on x86-based software and management frameworks, seamless integration is a critical consideration for new architectures.

Highlighting the importance of this compatibility, Kevork Kechichian, Executive Vice President and General Manager of the Data Center Group (DCG) at Intel Corporation, emphasized, “The data center software ecosystem is built on x86, and it runs on Xeon — providing a mature, proven foundation that developers, enterprises, and cloud providers rely on at scale. Workloads of the future will require a heterogeneous mix of computing, and this collaboration with SambaNova delivers a cost-efficient, high-performance inference architecture designed to meet customer needs at scale — powered by Xeon 6.”

The jointly developed platform is slated to be accessible to enterprises, cloud providers, and sovereign AI deployments in the latter half of 2026, offering a promising solution for the evolving landscape of AI infrastructure.