Dell and AMD Expand Enterprise AI Infrastructure with High-Density Hardware
Dell Technologies and AMD are teaming up to deliver modular, high-density AI infrastructure, bringing performance gains for agentic AI and model inference.
Dell Technologies and Advanced Micro Devices (AMD) are expanding their joint efforts to build enterprise-grade hardware tailored for demanding artificial intelligence workloads. As agentic AI and model inference drive up requirements for both general compute and specialized acceleration, data centers face physical constraints around power, space, and cooling. In response, Dell and AMD are focusing on rack-scale density, air-cooled hardware choices, and modular infrastructure designed to scale from small deployments up to massive supercomputers.
Server consolidation and density at the rack level
Agentic AI systems rely heavily on central processing units alongside graphics accelerators. However, integrating additional CPU nodes into data centers already filled with GPU racks presents a spatial challenge. According to Pradyumna Narendra, product planner and senior consultant at Dell, systems powered by new AMD processors achieve 70% generation-over-generation performance gains. This allows enterprise operators to condense up to 13 14th-generation Dell PowerEdge servers into a single new unit.
Additionally, Dell is adjusting physical form factors to optimize thermal performance. Form factor designs such as Dell's three-rack-unit (3U) server require less power for cooling compared to two-unit equivalents, enabling higher compute density per rack. Narendra noted that as inferencing and agentic workloads accelerate, the demand for underlying compute expands exponentially, making rack-level efficiency critical.
Modular software and PCIe hardware integration
To streamline enterprise adoption, the joint Dell AI Platform with AMD combines AMD EPYC CPUs and Instinct accelerators with Dell PowerEdge servers, PowerSwitch networking, and PowerScale storage. Dell has integrated AMD's ROCm open software stack into this environment, allowing organizations to train, fine-tune, and serve large language models on-premises. Varun Chhabra, senior vice president at Dell, emphasized that this modular approach allows enterprises to begin with smaller composable units and scale infrastructure without having to rearchitect their environment.
Hardware options have also expanded with PCIe form factors. Dell announced support for the AMD Instinct MI350P PCIe GPU across its PowerEdge XE7745 and R7725 servers. The MI350P features 144 gigabytes of HBM3E memory, offering significant bandwidth to improve time-to-first-token latency during inference. Mike Darby, senior manager at AMD, highlighted that memory bandwidth directly impacts user experience during model serving. Furthermore, Robert Hormuth, corporate vice president at AMD, pointed out that PCIe cards enable air-cooled deployments. This provides an immediate path for enterprises that are not yet ready to implement direct liquid cooling infrastructure.
Telecommunications and sovereign AI initiatives
The hardware partnership extends into targeted industry applications and research initiatives. In the telecommunications sector, Dell and AMD collaborated with AT&T on the Open Telco AI initiative. AT&T processed over one trillion tokens to train its OTel 2.0 models using a Dell PowerEdge XE9785 server equipped with eight AMD Instinct MI355X GPUs.
In Europe, Dell and AMD partnered with the University of Cambridge to launch the Sovereign AI Innovation Lab (SAIL). This initiative builds upon Cambridge’s Zenith AI supercomputer, which utilizes fifth-generation AMD EPYC processors and Instinct MI355X GPUs within Dell's infrastructure to advance scientific research. Dave Vellante of theCUBE Research noted that long-term leadership in AI infrastructure depends on engineering systems that continuously learn and improve the whole platform rather than relying solely on single-day benchmark figures.
What it means for developers
For engineering teams and developers, the focus on rack-scale hardware and optimized inference pipelines signals better performance predictability and lower latency for deployed models. High-bandwidth HBM3E memory on chips like the MI350P helps reduce time-to-first-token delays, which is crucial for real-time agentic workflows and interactive user applications. Modular deployment models also allow enterprise developers to prototype on smaller on-premises setups before scaling up to full production hardware.
While managing physical infrastructure remains a core focus for data center teams, software developers building applications often need straightforward access to these underlying models. For teams looking to test and build across multiple model architectures without managing on-premises hardware, developers can try top AI models cheaply through one API at https://apixoai.online. Ultimately, the hardware consolidation driven by Dell and AMD provides the physical compute foundation necessary to keep enterprise model training and high-throughput inference scalable.
Source: Rack-scale AI: How Dell and AMD are building infrastructure — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
