Intel has begun customer sampling of its D-Series data-center GPUs as it seeks a larger share of the enterprise AI infrastructure market. The D270, formerly associated with the Crescent Island codename, uses Intel's Xe3P architecture and targets memory-intensive inference, long-context processing and agentic workloads rather than the largest training systems. It delivers approximately 1.3 petaflops of peak FP8 performance at a 350-watt power limit, with 32 Xe3P cores, 256 XMX engines and a unified 32 MB L2 cache. Intel's reference configuration uses 160GB of LPDDR5X memory, while partner designs can reach 480GB with estimated bandwidth of 1.5 terabytes per second. Intel said one GPU can support models with up to 300 billion parameters and eight GPUs can handle models exceeding one trillion parameters. The company also presented open-standard Xeon 6+ rack designs, including selected configurations with more than 46,000 processor cores and 133.45 kilowatts of total rack power, positioning Xeon processors alongside GPUs to coordinate data, workflows and accelerator communication. Intel is working with OEMs including ASUS, Dell, HPE and Supermicro, as well as software providers, and plans further technical presentations at the Open Compute Project Summit on October 13.