OCP Global Summit Showcases AI Infrastructure Momentum
The OCP Global Summit is underway in San Jose, California, bringing together hardware vendors, cloud operators, chip designers, and data center engineers for a week of announcements focused on the physical layer of artificial intelligence. The annual event is hosted by the Open Compute Project, an industry initiative that develops open, efficient, and scalable hardware designs for data centers and other computing infrastructure. This year’s agenda spans AI servers and accelerators, standardized hardware designs, power and cooling systems, networking, and connectivity. The common thread is the pressure that AI workloads are placing on every part of the data center, from the rack to the coolant loop to the storage tier.
Key Facts at a Glance
- Nvidia contributed key rack and compute tray designs for its Vera Rubin platform to the Open Compute Project through its modular MGX architecture.
- Arm released a white paper proposing a use-case-driven methodology for measuring CPU contributions to agentic AI systems, with five representative workload types.
- Independent testing highlighted by Arm found its AGI CPU could deliver up to twice the rack-level performance of AMD’s Epyc Turin generation across select agentic AI workloads.
- Kioxia introduced its LD4 Series enterprise SSDs for AI and hyperscale data centers, using eighth-generation BiCS FLASH 3D QLC NAND.
- Cooler Master expanded its enterprise liquid-cooling portfolio with a 2,500 kW liquid-to-liquid coolant distribution unit.
- Flex announced prefabricated modular data center systems designed to accelerate AI infrastructure deployment by as much as 50% compared with traditional construction.
Open Compute Project and the Push for Scalable AI Infrastructure
The Open Compute Project began as an effort to rethink data center hardware for scale and efficiency, and it has evolved into a broad collaboration that touches servers, storage, networking, and facility design. The organization’s work matters more than ever as AI models grow larger and inference moves into production. AI training and inference demand dense compute, high-bandwidth memory, fast interconnects, and enormous amounts of power. Traditional data center designs, which were built around lower-density racks and air cooling, are increasingly strained. The OCP Global Summit has become a venue where suppliers can align around shared specifications that reduce duplication and speed adoption.
The announcements this week reflect that shift. They cover AI-optimized hardware, liquid cooling equipment, enterprise storage, and modular construction. Each piece addresses a different bottleneck in the AI infrastructure stack. Nvidia is trying to make rack-scale AI systems easier to build. Arm is trying to define how CPU performance should be measured in agentic workflows. Kioxia is targeting storage density. Cooler Master is attacking heat removal. Flex is attacking deployment time. Together, they show how the AI boom is reshaping the data center supply chain.
Nvidia Contributes Vera Rubin Designs to OCP
Nvidia submitted key rack and compute tray designs for its Vera Rubin platform to the Open Compute Project through its modular MGX architecture. The goal is to give server vendors and other infrastructure suppliers a shared design foundation for building AI systems. By contributing these designs, Nvidia is signaling that openness around form factors and integration can help accelerate adoption of its platform. The MGX architecture is modular, which means suppliers can mix and match components while staying within a common mechanical and electrical framework.
One feature that distinguishes Nvidia’s rack and tray designs is that there are no fans or cabling because everything is liquid cooled. That is a significant departure from conventional server racks, where airflow and cable management dominate the layout. A fanless, cable-free design can make systems easier to deploy and service. It can also improve reliability by removing moving parts and reducing the number of physical connections that can fail. Liquid cooling is becoming essential for AI accelerators because air alone cannot efficiently remove the heat generated by high-density compute. By standardizing liquid-cooled rack and tray designs through OCP, Nvidia is helping to create an ecosystem of compatible components, from cold plates to manifolds to facility-side coolant distribution.
Arm Announces Methodology for Evaluating CPU Performance
Arm released a white paper, developed with partners, that proposes a use-case-driven methodology for measuring the CPU’s contribution to agentic AI systems. Agentic AI refers to systems that can plan, reason, use tools, and execute multi-step tasks with some degree of autonomy. Those workloads place different demands on CPUs than traditional web serving or batch processing. They involve orchestration, sandboxing, tool calls, memory management, and coordination with accelerators. Arm’s paper defines five representative workload types: software engineering, enterprise operations, financial and scientific research, personal assistants, and robotics. It pairs component-level benchmarking with end-to-end task measurement built on replayable agent trajectories.
Arm also highlighted independent testing by Signal65 that found its AGI CPU could deliver up to twice the rack-level performance of AMD’s Epyc Turin generation server processors across select agentic AI workloads. The comparison emphasizes completed work under rack-level constraints rather than conventional CPU benchmarks alone. In the testing, an Arm AGI CPU averaged 1.7x better performance for tooling, 2.0x for agentic workflows and orchestration, and 2.2x for reasoning and planning in modeled per-rack performance versus an AMD Epyc Turin processor. Arm’s AGI CPU reduces sandbox creation latency by up to 45% versus AMD Epyc 9965, helping agents spin up isolated execution environments faster, according to Arm.
The methodology matters because agentic AI is difficult to benchmark. A single metric such as instructions per cycle or a generic integer score does not capture how quickly a system can complete a realistic task. By focusing on replayable agent trajectories and end-to-end task completion, Arm is trying to shift the conversation toward practical outcomes. The claim of up to twice the rack-level performance will likely spark debate, especially because rack-level constraints include power, cooling, and density. But it reflects a broader industry trend: vendors are racing to define benchmarks for AI infrastructure that go beyond peak chip specifications.