Intel details GPU D-series for agentic AI at OCP Summit

Intel used the Open Compute Project Global Summit to flesh out its vision for the hardware layer of agentic AI. The company had previously signaled its intent to serve this market; now it is providing concrete technical specifications. The Intel GPU D-series, formerly code-named Crescent Island, is positioned as a purpose-built inference accelerator, with a memory capacity and power envelope that Intel says is tuned for sustained, memory-intensive workloads. Alongside it, Intel is pushing Xeon-based rack designs built on the Open Rack v3 standard, aiming to address the orchestration demands that agentic workflows place on CPUs. The software side is equally central to Intel's pitch: the company is emphasizing open frameworks, day-0 support for new models, and a common Xe software foundation across its GPUs, which together are meant to let enterprises deploy inference without overhauling their existing infrastructure. Intel also highlighted its ongoing contributions to the Open Compute Project, including a new standardized accelerator card form factor, as part of an effort to give customers flexibility and avoid vendor lock-in as AI workloads evolve.