---

Agentic AI is quickly moving from research labs into real-world deployments, forcing data center architects to rethink how inference workloads are handled. In response, Intel and SambaNova Systems have introduced a new heterogeneous architecture designed to support the next generation of AI applications.

The companies say the design combines GPUs, SambaNova RDUs, and Intel Xeon 6 processors to balance performance, efficiency, and compatibility. For readers of eeNews Europe — particularly engineers and system architects working on AI infrastructure — the announcement highlights how future inference systems may move beyond GPU-only approaches.

Heterogeneous approach for agentic AI

The new architecture targets emerging “agentic AI” workloads, which involve systems capable of reasoning, planning, and executing complex actions autonomously. These workloads place different demands on computing resources at different stages of inference.

Under the blueprint, GPUs are used for the prefill stage, where models ingest large prompts. SambaNova’s RDUs (Reconfigurable Dataflow Units) then handle high-throughput decode operations, while Xeon 6 processors act as host and action CPUs, coordinating tasks and running application logic.

This division of labor reflects a growing industry trend: pairing specific phases of AI inference with the most suitable hardware accelerator rather than relying solely on GPUs. According to the companies, the design also maintains compatibility with the x86-based software stack that underpins most data center environments.

Leveraging the x86 ecosystem

Maintaining compatibility with existing infrastructure is a key part of the collaboration. Data centers and enterprise AI deployments often rely heavily on x86-based software and management frameworks, making integration a major concern for new architectures.

“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,” said Kevork Kechichian, Executive Vice President and General Manager of the Data Center Group (DCG) at Intel Corporation. “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 engineered platform is expected to become available to enterprises, cloud providers, and sovereign AI deployments in the second half of 2026.

---

---