Intel introduced its D-series GPU, formerly known as Crescent Island, at the Open Compute Project Global Summit. The hardware is based on the Xe3P architecture, offering 1.3 petaflops of FP8 compute and 1.5 terabytes per second of memory bandwidth. The card uses LPDDR5X memory, supporting up to 480 gigabytes per card, which allows for running models with up to 300 billion parameters.
The company also showcased new Xeon rack designs, including configurations with Xeon 6+ processors that scale beyond 46,000 cores per rack at 133.45 kilowatts. These systems are designed to handle agentic AI workloads, which require significant CPU resources for orchestration, data processing, and tool execution. Intel aims to shift focus from raw core counts to concurrent agent capacity.
Intel reported second-quarter revenue of $16.1 billion, representing a 25% year-over-year increase. Data Center and AI revenue grew by 59% to $6.3 billion. Despite this growth, the company noted that it could meet only about half of the current customer demand for CPUs, even after increasing its 2026 capital expenditure budget to over $20 billion.
The strategy focuses on inference rather than training, targeting enterprise needs for cost-effective, power-efficient infrastructure. Intel is promoting an orchestration layer that matches tasks to appropriate models, citing Vercel AI Gateway data that shows open-weight model traffic share grew from roughly 31% in early July to about 78% by late September.
