Intel Bets Xeon 6 Can Run 205 AI Agents At Once, No Nvidia GPUs Needed

Intel is betting its Xeon 6 processors and open designs can capture a share of agentic AI infrastructure as companies build and deploy AI agents.

Key Takeaways

• Intel announced open hardware and software designs for agentic AI workloads at the Open Compute Project summit
• Dell’s PowerEdge XE9780LAP server supported up to 205 concurrent AI agents in internal tests using Intel Xeon 6 processors
• Intel argues CPUs can efficiently handle agentic AI tasks such as coordinating work, retrieving data and running business logic
• Enterprise adoption and independent benchmarks will determine whether Xeon 6 infrastructure performs beyond Dell’s controlled internal testing

At the Open Compute Project summit Monday, the company announced a set of open hardware and software designs for agentic AI workloads, alongside partner hardware from Dell and QCT shown at the same event.

Intel’s announcement said its software stack is built for “real-world AI applications and agents,” targeting workloads that span simple chatbot responses through to more complex, multistep coding agents that require sustained compute over longer task sequences.

A separate QCT showcase at the same summit detailed Intel Xeon 6-based data center platforms spanning enterprise applications, virtualization and AI analytics.

Dell separately disclosed that its PowerEdge XE9780LAP server, built on Xeon 6, supported up to 205 AI agents running concurrently in internal tests, using Intel’s Advanced Matrix Extensions, a set of processor instructions that speed up the matrix math underlying most AI model computation.

Why Intel Is Betting On CPUs For Agentic Workloads

Nearly every major AI model today trains and runs on GPUs, chips originally built for graphics rendering that happen to excel at the parallel math AI requires.

Intel’s pitch is different, it argues a growing share of agentic AI work, coordinating tasks, retrieving data, running business logic, can run efficiently on CPUs rather than requiring GPU clusters for every step.

Also Read: Satya Nadella Bets Big: Maia Chip To Run Microsoft’s AI At Gigawatt Scale By 2027

Dell’s 205-agent test result gives that argument a concrete number, suggesting CPU-only infrastructure can support meaningfully more simultaneous agents than enterprises might assume, at potentially lower cost than GPU-heavy deployments.

How Intel’s Open Playbook Differs From Past Cycles

Intel has spent much of the past three years losing ground to Nvidia in AI data center chips, a gap a Reuters-cited IDC report on China’s market illustrates in a different geography.

Rather than compete head-on for the same GPU-heavy workloads, Intel’s OCP showing leans into open, standardized rack and software designs that multiple hardware partners can adopt, a strategy closer to how the x86 processor architecture spread decades earlier.

Intel is wagering that agentic AI, which spreads compute across many smaller, cheaper operations rather than a few training runs, favors this more distributed, CPU-inclusive approach.

What Enterprise Buyers Should Track Next

The near-term signal to watch is adoption, whether cloud providers and enterprises deploy Xeon 6-based agentic AI infrastructure at scale beyond Dell’s internal test, and whether independent benchmarks confirm the 205-agent figure under production conditions rather than controlled testing.

Intel’s broader open infrastructure strategy also depends on partner uptake at OCP continuing beyond this single summit cycle.

Read Next: HCLTech Becomes Anthropic’s Top Claude Partner, Builds New AI Business Unit