
Intel unveils Xeon 6+ processors for agentic AI workloads
What Intel announced
At Computex 2026 in Taipei, Intel launched the Xeon 6+ family of data center processors and positioned them for "agentic AI", the class of software in which AI agents run many tasks in parallel, call tools and move data between services. The chips are built on Intel's 18A manufacturing process, which Intel says makes this their first use of that technology in a data center CPU.
Key specifications reported
| Detail | Reported figure |
|---|---|
| Manufacturing process | Intel 18A |
| Core design | Up to 288 efficient cores (E-cores) per processor |
| Target workloads | Orchestration, concurrency and data movement for AI agents |
| Companion hardware | Intel Ethernet E835 controllers and adapters |
Performance and density numbers, including claims about how many cores fit in a liquid-cooled rack, come from Intel's own messaging. Independent benchmarks had not been widely published at the time of the launch.
Why CPUs matter for AI agents
GPUs handle the heavy math of training and running large models, but agentic systems also need plenty of general-purpose computing: scheduling many small tasks, running code, handling network traffic and keeping databases responsive. A processor with a very high number of efficient cores is aimed at that surrounding work, where throughput per watt and rack density matter.
Partners and ecosystem
- Intel paired Xeon with SambaNova's SN50 accelerators for rack-scale inference systems, with shipments expected later in the year.
- Server makers such as Giga Computing (GIGABYTE) announced platforms that support the new processors.
- Ericsson was cited as an early adopter reporting performance gains.
The competitive picture
The launch arrives as Nvidia pushes its own CPUs and PC chips and as AMD and Arm-based designs compete for data center share. Intel is using 18A to show that its manufacturing roadmap is back on track. Whether buyers choose the new Xeons will depend on real-world performance, power efficiency, software support and price.
Related reading: Nvidia's RTX Spark superchip and TSMC's 2nm production ramp.
Advertisement
