AI Business

Dell Shifts AI Infrastructure Focus to Rack-Level Operations and Management

As AI clusters scale to supercomputing proportions, managing the operational complexity of running these systems has become more critical than acquiring the hardware itself. Dell is addressing this challenge with new tools designed to oversee entire racks rather than individual servers.

·3 min read
Scripts can’t run AI factories: Dell moves cluster management from the server to the rack
Scripts can’t run AI factories: Dell moves cluster management from the server to the rack

The challenge facing AI infrastructure providers has fundamentally changed. When artificial intelligence clusters reach supercomputing scale, the bottleneck is no longer procuring equipment. Instead, the real difficulty lies in maintaining the operational stability of racks, switches, GPUs and cooling systems after deployment.

This shift in operational demands is driving changes in what infrastructure vendors offer their customers. Dell Technologies Inc. has spent the past year expanding its AI Factory strategy beyond traditional servers and storage offerings. Customers increasingly seek guidance on deployment best practices and software tools to effectively manage the systems they purchase, according to Andrew Vaz, vice president of product management for the Telecom Systems Business at Dell.

What we are seeing is a fundamental shift from procuring supply to actually standing up and running these very complicated AI clusters and data centers. One of the interesting things that we've seen is it's actually now impossible to manage these things with just a few scripts, manually, in an old world manner. The toolsets are becoming more important.

Andrew Vaz, Dell Technologies

Rack-scale management for the supercomputing era

Dell unveiled two new management solutions at SC26: Dell OpenManage Rack Scale and Dell Fabric Manager. These platforms are engineered to handle AI infrastructure deployment and ongoing operations. OpenManage Rack Scale can manage 100,000 devices in its initial release, a substantial increase from the 8,000 servers supported by its predecessor, OpenManage Enterprise, according to Saurabh Kapoor, director of product management and strategy for AI compute and networking solutions at Dell. Modern systems like Nvidia Corp.'s NVL72 racks integrate compute, power shelves, switches and coolant distribution into unified units.

OpenManage Enterprise, the previous product we had, was perfect for servers. Server was the unit of compute. But you have seen in the last 18 months, the world has moved very quickly to larger ecosystems and infrastructures. Rack is the new unit of play.

Saurabh Kapoor, Dell Technologies

Observability across infrastructure layers represents the other critical component. Dell's iDRAC controller collects more than 320 metrics from each server, and an AI-powered chatbot can analyze this telemetry using locally deployed large language models, Kapoor explained. The stakes are high: a single cooling system failure can destroy millions of dollars in equipment, making comprehensive monitoring essential across every component.

It could be something as simple as, 'Hey, this transceiver's about to die that you can't diagnose with metrics you have today.' You need to use AI to go look at it. You need telemetry off the system to go collect all of that, correlate it, run your algorithms on them, train it and understand what's happening.

Andrew Vaz, Dell Technologies

Network infrastructure has become the critical constraint in AI deployments. Dell Fabric Manager converts validated architectural designs developed with Nvidia and Advanced Micro Devices Inc. into deployable configurations that work across different network operating systems and silicon platforms. The emergence of agentic AI is also transforming network traffic patterns, with actual bandwidth consumption far exceeding projections. Vaz noted that enterprise bandwidth growth is running at nine to 10 times the anticipated 3.5-fold increase.

https://www.youtube.com/embed/FkYg7ZywuPs?start=1&feature=oembed

We're getting blown away by it. It's more like nine to 10x. It's going everywhere: east, west, north, south.

Andrew Vaz, Dell Technologies