Intel's data center GPU push
Intel has revealed the first OAM designs for its D-Series GPU family, formerly known as Crescent Island, aimed at AI data centers. The images were shared by Intel itself at the OCP 2026 event and on social media.
A dual-chip, liquid-cooled card
One of the designs packs two Crescent Island GPUs onto a single liquid-cooled OAM card. It is the first concrete sign of how Intel plans to approach density and cooling at server scale.
Xe3P architecture and memory
The D-Series GPUs are based on the Xe3P architecture, and the family is said to offer up to 512 GB of memory. The first Xe3P die shots were also shown in single and dual packages, with the chip appearing smaller than its Battlemage predecessor.
Why it matters
Intel is looking to take share in an AI accelerator market dominated by NVIDIA. The OAM form factor and liquid cooling have become standard in modern data center racks, so Intel's alignment with that standard is critical for adoption by server makers.
