The Open Compute Project Global Summit in San Jose, California, has become a key venue for the physical infrastructure behind cloud computing and artificial intelligence. This year’s event highlighted how vendors are responding to the demands of AI data centers, where power density, cooling capacity, storage throughput, and deployment speed are increasingly intertwined. The announcements ranged from rack-level designs for Nvidia’s Vera Rubin platform to new methods for measuring CPU performance in agentic AI systems, high-capacity enterprise SSDs, large-scale liquid-cooling equipment, and prefabricated modular data centers.

The Open Compute Project is an industry initiative focused on developing more open, efficient, and scalable hardware designs for data centers and other computing infrastructure. Its annual summit covers servers, accelerators, power and cooling, networking, connectivity, and the standardized or openly shared designs that allow suppliers to build interoperable systems. As AI workloads push racks to unprecedented power levels, the event has become a showcase for technologies that address the practical constraints of building and operating large-scale AI facilities.

Key Facts From the Summit

• Nvidia contributed key rack and compute tray designs for its Vera Rubin platform to the Open Compute Project through its modular MGX architecture. The designs are fanless and cable-free because they are liquid cooled, which the company says makes them easier to deploy and service.
• Arm released a white paper proposing a use-case-driven methodology for measuring CPU contributions to agentic AI systems. It defines five workload types and pairs component-level benchmarking with end-to-end task measurement based on re-playable agent trajectories. Arm also highlighted third-party testing that compared its AGI CPU with AMD Epyc Turin processors.
• Kioxia introduced the LD4 Series enterprise SSDs for read-intensive workloads in hyperscale and high-density 1U servers. The drives use eighth-generation BiCS FLASH 3D QLC NAND, offer 15.36TB and 30.72TB capacities, and have a validated architecture designed to support up to 122.88TB.
• Cooler Master expanded its end-to-end liquid-cooling portfolio with a 2,500 kW liquid-to-liquid coolant distribution unit for high-density AI and HPC infrastructure. The system supports flow rates up to 990 gallons per minute and can centrally manage as many as 32 interconnected CDUs.
• Flex expanded its prefabricated modular data center portfolio with AI Modules that combine power delivery, high-density computing, and advanced liquid cooling in factory-built systems. The company says the approach can accelerate AI infrastructure deployment by as much as 50% compared with traditional construction.

Nvidia Pushes Open Rack Designs for Vera Rubin

Nvidia’s contribution to the Open Compute Project centers on rack and compute tray designs for its Vera Rubin platform, delivered through the modular MGX architecture. The goal is to give server vendors and other infrastructure suppliers a shared design foundation for building AI systems. By submitting the designs to OCP, Nvidia is signaling that it wants the surrounding ecosystem—including rack integrators, cooling suppliers, power equipment makers, and server builders—to align around common mechanical and electrical building blocks.

The most notable feature of the rack and tray designs is the absence of fans and cabling. Everything is liquid cooled, which removes the need for traditional air-moving components and much of the internal cabling that complicates dense AI servers. That design choice can simplify deployment and serviceability, but it also places more responsibility on the cooling loop, manifolds, coolant distribution units, and leak-management systems. For operators, the trade-off is clear: liquid cooling can support much higher power densities, but it requires careful planning and facility readiness.

The Vera Rubin designs also reflect a broader shift in AI infrastructure. AI servers are no longer simply general-purpose compute nodes with accelerators added. They are integrated systems that combine GPUs, CPUs, high-speed interconnects, power delivery, and thermal management. Standardizing rack and tray designs through OCP could reduce fragmentation and help suppliers bring compatible products to market faster, especially as customers seek alternatives to fully proprietary designs.

Arm Proposes New Way to Measure CPU Performance for Agentic AI

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. The methodology 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 re-playable agent trajectories, an approach intended to capture how well a CPU supports real agentic workflows rather than isolated synthetic tests.

The shift toward agentic AI is important for CPU vendors because agents do more than run a single inference request. They plan, call tools, orchestrate workflows, retrieve data, execute code, and manage isolated environments. Those tasks place different demands on CPUs, memory, I/O, and virtualization. Traditional benchmarks may not reflect the latency and throughput that matter when an agent is completing a multi-step task under rack-level constraints.

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. In the modeled per-rack comparison, 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 versus an AMD Epyc Turin processor. Arm said its AGI CPU reduced sandbox creation latency by up to 45% compared with AMD Epyc 9965, helping agents spin up isolated execution environments faster.

Such claims are likely to be debated because benchmark methodologies and workload assumptions can significantly affect results. The more consequential point is that the industry is looking for evaluation methods that reflect completed work under rack-level constraints, not just conventional CPU benchmarks. If agentic AI becomes a major workload category, buyers will want to know how many tasks a system can complete per rack, how quickly it can create sandboxed environments, and how much orchestration overhead it incurs.

Kioxia Targets AI Storage Density With LD4 SSDs

Kioxia unveiled its LD4 Series enterprise solid-state drives, designed to increase flash storage density in hyperscale data centers and high-density 1U servers. The drives target read-intensive workloads, including data storage applications supporting artificial intelligence infrastructure. AI systems generate and consume enormous amounts of data, from training datasets and model checkpoints to inference logs and retrieval-augmented generation stores, so storage density and read throughput become critical infrastructure concerns.

The new SSDs offer eighth-generation BiCS FLASH 3D quad-level cell NAND technology with capacities of 15.36TB and 30.72TB. The validated architecture is designed to support capacities of up to 122.88TB. The drives support PCIe 5.0 and NVMe 2.0e, along with the Open Compute Project’s Datacenter NVMe SSD Specification 2.6. They feature a single-port design for hyperscale and data center-class servers. Kioxia said the LD4 Series is currently sampling to select customers.

QLC NAND has become increasingly attractive for read-intensive data center workloads because it offers higher capacity per chip than TLC or SLC. The trade-off is usually endurance and write performance, which is why QLC SSDs are often aimed at read-heavy applications. For AI infrastructure, that can include storing large datasets, model artifacts, and content that is read many times but updated less frequently. The OCP specification support is also significant because hyperscale operators increasingly require standardized management, telemetry, and thermal characteristics from storage devices.

Cooler Master Brings Gaming Liquid Cooling Experience to the Data Center

Cooler Master, a company best known for providing liquid cooling to gaming systems, is bringing that experience to the data center. It is expanding its end-to-end liquid-cooling portfolio for AI data centers, highlighting a 2,500 kW liquid-to-liquid coolant distribution unit designed for high-density AI and HPC infrastructure. The system supports coolant flow rates of up to 990 gallons per minute and features dual-system architecture, intelligent load balancing, and centralized management for as many as 32 interconnected CDUs, according to the vendor. Mass production is targeted for the fourth quarter of 2026.

The move illustrates how thermal management has become a central constraint in AI data center design. High-end accelerators can consume far more power per rack than traditional enterprise servers, and air cooling struggles to remove that heat efficiently at scale. Liquid cooling can support higher rack densities, but it requires coolant distribution units, manifolds, hoses, leak detection, and facility-level integration. Cooler Master’s expanded portfolio is aimed at data center operators building new AI facilities as well as those retrofitting existing sites to accommodate increasing rack densities and thermal demands.

The company’s consumer heritage may be both an advantage and a challenge. It brings experience in high-volume liquid-cooling components and thermal engineering, but data center customers demand reliability, serviceability, and long-lifecycle support that go beyond consumer PC requirements. The 2,500 kW CDU and centralized management capabilities suggest Cooler Master is targeting large deployments rather than edge or small-scale installations.

Flex Expands Modular Data Center Portfolio for AI

Flex announced an expansion of its prefabricated modular data center portfolio with integrated systems designed to accelerate deployment of AI infrastructure by as much as 50% compared with traditional construction methods. The new AI Modules combine power delivery, high-density computing, and advanced liquid cooling in factory-built systems. This approach shifts much of the integration and testing work from the data center construction site to Flex, the company says.

The systems include an 800-volt direct-current power rack for power conversion and distribution, cooling distribution units to manage heat from power-intensive AI workloads, and high-density IT racks incorporating AI servers, networking equipment, and high-speed interconnects. The 800 VDC power architecture is notable because higher-voltage DC distribution can reduce conversion losses and simplify power delivery in dense AI facilities. Liquid cooling is also integrated into the modules, reflecting the expectation that future AI racks will be designed around liquid rather than air as the primary cooling medium.

The expanded portfolio is part of Flex’s cloud and power infrastructure business, which the company plans to spin off as an independent, publicly traded company called Axiom Solutions International in the first quarter of 2027. The spin-off plan adds a corporate finance dimension to the product announcement. As AI infrastructure spending grows, suppliers are positioning themselves to capture demand for factory-built power, cooling, and compute modules that can be assembled and commissioned faster than traditional stick-built data centers.

Source:Network World News