Intel Corporation (INTC) shares declined 3.20% to $101.35 at Monday's close before recovering 0.22% to $101.57 in after-hours trading. The drop came as the company revealed new GPU specifications and Xeon server designs at the Open Compute Project (OCP) Global Summit, underscoring Intel's push to expand its data center business through efficient artificial intelligence infrastructure. The OCP Global Summit is the annual industry event for open, standards-based data center hardware designs.

Intel Unveils GPU D-Series for Large AI Workloads

At the OCP Global Summit, Intel released additional technical details for its GPU D-series, previously known under the codename Crescent Island. Built on Intel's Xe3P architecture, the new processor targets memory-intensive AI inference across enterprise data centers, with the company aiming to improve processing efficiency while reducing infrastructure costs and power consumption. The announcements arrive in a data center market where AI accelerator hardware has become one of the most competitive segments of the semiconductor industry.

The GPU architecture supports up to 480GB of LPDDR5X memory and delivers 1.5TB/s of memory bandwidth. The design also provides 1.3 petaflops of FP8 computing performance within a 350-watt power limit. According to Intel, these specifications are suited to large language models, long-context processing, and multiple inference workloads running simultaneously. The memory figures are central to that goal: capacity determines how large a model can be served, while bandwidth shapes how quickly it responds.

Intel said a single GPU can support models containing up to 300 billion parameters, while an eight-GPU configuration can handle models exceeding one trillion parameters. Customer sampling has already begun, and the company continues to collaborate with hardware manufacturers and software providers on deployment solutions.

Intel Expands Xeon Server Capacity for Agentic AI

Intel also presented new Xeon-based rack designs intended to support the growing computing requirements of enterprise AI systems. The designs follow open infrastructure standards, including Open Rack v3, to simplify integration into existing data centers. As a result, organizations can expand processing capacity without completely replacing their current infrastructure, a consideration for operators weighing AI capacity growth against capital budgets.

Select Xeon 6+ configurations deliver more than 46,000 processor cores per rack. These systems operate at 133.45 kilowatts of total rack power while supporting high-density computing operations. Intel designed the configurations to handle data processing, workload coordination, and communication between specialized accelerators. Power densities at this level require data center operators to plan energy delivery and cooling around concentrated per-rack consumption.

Agentic AI systems require processors to manage several connected tasks rather than complete isolated requests. Intel therefore positions Xeon processors alongside GPUs to handle workflow management and data movement, an approach designed to support concurrent AI applications while improving resource allocation across enterprise computing environments.

Intel Strengthens Open Software and Infrastructure Development

Alongside the hardware announcements, Intel outlined additional software improvements that support applications across its CPU and GPU product lines. The company continues to integrate established AI frameworks, inference tools, and open development environments into its platforms, with the goal of simplifying application deployment and reducing software compatibility barriers.

Intel also highlighted its contributions to the Open Compute Project, including standardized accelerator card designs. Since 2011, the company has submitted 116 contributions supporting open data center infrastructure. Its latest work focuses on improving hardware compatibility and giving system manufacturers greater design flexibility.

The announcements extend Intel's strategy of combining general-purpose processors, specialized accelerators, and interoperable software. Meanwhile, the company plans further technical presentations during the OCP Summit on October 13 — the next checkpoint for how these specifications move from announcement toward enterprise deployment. Its expanding hardware portfolio targets enterprise demand for scalable AI processing and more efficient data center operations.

This article was originally reported by Blockonomi.