Santa Clara, CA – Intel Expands Gaudi 3 AI Accelerator Deployment Options for Enterprise Workloads

New PCIe Cards and Rack-Scale Systems Offer Flexible, High-Performance Alternatives for AI Infrastructure

Intel has announced expanded deployment options for its Gaudi 3 AI accelerators, designed to make advanced AI hardware more accessible and versatile for enterprise users. The accelerators will now be available as standard PCIe add-in cards, allowing for easier integration into existing server environments without the need for specialized systems.

In addition to the PCIe option, Intel is introducing a new rack-scale configuration, capable of housing dozens of Gaudi 3 chips in a single chassis. This setup targets organizations requiring high-density AI computing for tasks like large-scale training, inference, and model experimentation.

Built with open-source compatibility at their core, Gaudi 3 accelerators are optimized to run a broad spectrum of AI models—including LLaMA 3 8B, LLaMA 4 Scout, and Maverick—while delivering high throughput and cost-efficiency. This move reflects Intel’s broader mission to offer enterprises competitive, non-proprietary alternatives to incumbent AI hardware platforms.

“Enterprises are demanding flexibility and value as they scale AI workloads,” said an Intel spokesperson. “With these expanded deployment options, Gaudi 3 makes it easier than ever to adopt open, efficient, and powerful AI infrastructure.”

Intel’s announcement underscores the increasing momentum behind scalable, open-access AI compute solutions, and positions Gaudi 3 as a strong contender in the rapidly growing enterprise AI accelerator market.