AI Infrastructure · San Jose, California
Intel Details Inference GPU and Xeon Rack Designs at OCP
Intel’s October 12 announcement details GPU D-series inference hardware and Xeon rack designs; customer sampling is underway.
Intel used the OCP summit to detail its GPU D-series inference accelerator and new Xeon-based rack designs, according to its October 12 announcement.
The GPU platform is designed around a 350-watt envelope and a conventional air-cooled PCIe card. Intel describes partner options up to 480 GB of memory, while its own branded card has 160 GB. Those capacities should not be conflated.
Intel says customer sampling is underway. Separately, select Xeon 6+ rack configurations exceed 46,000 cores at 133.45 kW of total rack power, underscoring the need to assess the complete rack rather than processor power alone.
These are Intel’s technical claims. The release does not establish universal commercial availability or independently measured customer performance; procurement teams should verify the exact platform and delivery schedule.
Original publisher (October 12, 2026). Source checked October 19, 2026. · AI Infrastructure.
Key details
- GPU power
- 350-watt power envelope
- GPU performance
- 1.3 petaflops FP8 and 1.5 TB/s memory bandwidth
Why it matters
The GPU's PCIe and air-cooled design is intended to fit existing data centers, while high-density Xeon rack designs raise power-delivery and cooling requirements. The announcement is a product and architecture update, not an open procurement.
Status and next steps
Intel said customer sampling of the GPU D-series was underway; it did not announce a general-availability date in the reviewed release.
Review the official next step →Project contacts and next steps
A direct project contact has not been verified for this article. Consult the linked source organization for the appropriate office.
Primary sources
- Intel investor release: open infrastructure for agentic AI · October 12, 2026