Industry Focus on AI Infrastructure at OCP Global Summit
The Open Compute Project (OCP) Global Summit in Silicon Valley drew record attendance, reportedly reaching 15,000 people, reflecting intense industry interest in the future of data center infrastructure. Major semiconductor companies, including Intel, AMD, and Nvidia, utilized the event to unveil substantial new technologies designed to power advanced artificial intelligence (AI) data centers. These disclosures heavily emphasized standardized server specifications championed by OCP.
Key announcements covered a range of advancements, from Intel’s upcoming air-cooled data center GPU and new server racks capable of housing over 46,000 cores of Intel Xeon CPUs, to AMD’s flexible server rack design and open-source updates from Nvidia, including information regarding Vera Rubin.
Addressing Enterprise AI Workloads
Industry experts noted that these announcements were tailored for different segments of the market. Some advancements specifically addressed partners and customers building data centers for agentic AI workloads, particularly the CPU-related details due to the processor’s critical role in enabling such applications. Other technologies aimed at supporting the broader expansion of AI data center capacity.
Bob Venero, CEO of Future Tech Enterprise, a solution provider ranked No. 90 on CRN’s 2026 Solution Provider 500, stated that “The channel needs AMD, Intel and Nvidia to continue to push the limits.” While acknowledging that hyperscalers and neocloud providers are currently building much of the capacity for agentic AI, Venero highlighted a significant opportunity for channel partners to assist enterprises with securing their own infrastructure due to concerns regarding cost and security.
“Customers are definitely figuring this out because they spent a ton of money and blew through their token budgets because they let [AI] agents run unregulated within their organizations,” he said. “So you’re seeing customers now more looking at, ‘Hey, how do we control costs? How do we drive these agents internally that will use our internal models and do things that are not necessarily cloud-based but on-prem?’”
Intel’s Hardware Advancements
GPU and CPU Solutions
Intel introduced its upcoming 480-GB data center GPU, designed for AI inference and codenamed “Crescent Island.” This GPU, branded as the Intel GPU D-series, is set to deliver up to 1.3 petaflops of 8-bit floating-point compute within a 350-watt power envelope. The chipmaker announced that the D-series GPU, which is currently being sampled with customers, will utilize a standard, air-cooled PCIe form factor, allowing up to eight cards per server.
Intel stated that the D-series GPU, built using the company’s Xe3P architecture, is suitable for large language models, long-context workloads, and concurrent services. A single card can support models up to 300 billion parameters, while an eight-GPU system can handle up to 1 trillion parameters. Intel is collaborating with a wide range of partners, including Asus, Cisco, Dell Technologies, HPE, Lenovo, Supermicro, NetApp, Nutanix, IBM-owned Red Hat, and Broadcom-owned VMware, to bring the D-series into production.
The company also presented new server rack designs capable of integrating over 46,000 cores from its Xeon 6+ CPUs, built on the Open Rack v3 standard. These high-density racks, developed with vendors like Asus, Gigabyte, and Supermicro, are designed to manage the high power demands of agentic AI workloads, fitting up to 46,000 cores within a 133.45 kilowatt power limit. The Xeon 6+ processors, internally known as “Clearwater Forest,” feature up to 288 efficiency cores and operate within a 450-watt power limit.
AMD’s Platform and Processor Updates
Flexible Rack Design and New CPU
AMD unveiled “Helion,” a flexible server rack reference design intended as an open blueprint. Helion is based on an OCP specification and can accommodate EPYC CPUs and Instinct GPUs, or CPUs alone. Supporting vendors, including Cisco, Dell Technologies, HPE, Lenovo, and Supermicro, plan to release Helion configurations beginning in November.
AMD intends for Helion to allow customers to customize their compute needs by matching the required core density, power, and GPU acceleration for their AI agents. For example, a CPU-only rack using 6th-Gen EPYC processors could house up to 32,768 cores for hyperscalers that require dedicated capacity for agent execution. Furthermore, the platform can enable an integrated rack combining “AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose compute” for enterprise users.
In a separate announcement, AMD CEO Lisa Su revealed “Verano,” a new EPYC CPU featuring up to 192 cores that incorporates an LPDDR memory interface for the first time in the family. Verano is part of the 6th-Gen EPYC “Venice” lineup and is optimized for low system power consumption. Production for Verano is scheduled for the first half of 2027.
Nvidia’s Software and Interoperability Efforts
Open-Sourcing AI Factory Components
Nvidia announced that it is open-sourcing various software components for managing AI data centers as part of its DSX AI Factory platform. This effort aims to help partners and customers maximize computational efficiency. The most significant open-source component is Run:ai, a Kubernetes-based workload management and orchestration software acquired by Nvidia in late 2024 for an estimated $700 million.
The open-source move makes the full Run:ai platform available, alongside individual components such as KAI Scheduler, GPU Fractioning Operator, KAI Resource Management, and Workload Map. Additionally, Nvidia released open-source components for network configuration, cluster validation, rack management, and bare metal provisioning. These components, originally designed for the DGX Cloud, enable cloud providers and software partners to integrate AI factory capabilities directly into their multi-tenant platforms.
Vera Rubin MGX Rack Design
Nvidia also presented the third generation of its MGX AI data center reference design, supporting its Vera Rubin NVL72 and Vera CPU server racks. This new specification, called MGX 1.2, is designed to facilitate the creation of interoperable platforms. Features include rack-scale blind-mate liquid cooling, an interoperable rack-scale interconnect infrastructure, and a liquid-cooled 5,000-amp busbar built for high-power AI systems. The architecture is designed to support deployments up to 250kW while simplifying serviceability and deployment.
