Dell Aims to Be the Go-To Source for Enterprise AI Infrastructure

Michael Dell is pitching a “decentralized” future for artificial intelligence that his company’s devices will make possible.   

“The future of AI will be decentralized, low-latency, and hyper-efficient,” predicted the Dell Technologies founder, chairman, and CEO in his Dell World keynote, which you can watch on YouTube. “AI will follow the data, not the other way around,” Dell said at Monday’s kickoff of the company’s four-day customer conference in Las Vegas.

Dell is betting that the complexity of deploying generative AI on-premise is driving companies to embrace a vendor with all of the parts, plus 24-hour-a-day service and support, including monitoring.

On day two of the show, Dell chief operating officer Jeffrey Clarke noted that Dell’s survey of enterprise customers shows 37% want an infrastructure vendor to “build their entire AI stack for them,” adding, “We think Dell is becoming an enterprise’s ‘one-stop shop’ for all AI infrastructure.”

Dell’s new offerings include products meant for so-called edge computing, that is, inside customers’ premises rather than in the cloud. For example, the Dell AI Factory is a managed service for AI on-premise, which Dell claims can be “up to 62% more cost-effective for inferencing LLMs on-premises than the public cloud.”

Dell brands one offering of its AI Factory with Nvidia to showcase the chip giant’s offerings. That includes, most prominently, revamped PowerEdge servers, running as many as 256 Nvidia Blackwell Ultra GPU chips, and some configurations that run the Grace-Blackwell combination of CPU and GPU.

Future versions of the PowerEdge servers will support the next versions of Nvidia CPU and GPU, Vera and Rubin, said Dell, without adding more detail. 

Dell also unveiled new networking switches running on either Nvidia’s Spectrum-X networking silicon or Nvidia’s InfiniBand technology. All of these parts, the PowerEdge servers and the network switches, conform to the standardized design that Nvidia has laid out as the Nvidia Enterprise AI factory.

A second batch of updated PowerEdge machines will support AMD’s competing GPU family, the Instinct MI350. Both PowerEdge flavors come in configurations with either air cooling or liquid cooling.

Complementing the Factory servers and switches are data storage enhancements, including updates to the company’s network-attached storage appliance, the PowerScale family, and the object-based storage system, ObjectScale. Dell introduced what it calls PowerScale Cybersecurity Suite, software designed to detect ransomware, and what Dell calls an “airgap vault” that keeps immutable backups separate from production data, to “ensure your critical data is isolated and safe.” 

The ObjectScale products gain support for remote data access (RDMA), for use with Amazon’s S3 object storage service. The technology more than triples the throughput of data transfers, said Dell, lowers the latency of transfers by 80%, and can reduce the load on CPUs by 98%.

“This is a game changer for faster AI deployments,” the company claimed. “We’ll leverage direct memory transfers to streamline data movement with minimal CPU involvement, making it ideal for scalable AI training and inference.”

Dell AI Factory also emphasizes the so-called AI PC, workstations tuned for running inference. That includes a new laptop running a Qualcomm circuit board, the AI 100 PC inference card. It is meant to make local predictions with Gen AI without having to go to a central server. 

The Dell Pro Max Plus laptop is “the world’s first mobile workstation with an enterprise-grade discrete NPU,” meaning a standalone chip for neural network processing, according to Dell’s analysis of workstation makers.

The Pro Max Plus is expected to be available later this year.

A number of Dell software offerings were put forward to aid the idea of the decentralized, “disaggregated” AI infrastructure. 

For example, the company made an extensive pitch for its file management software, Project Lightning, which it calls “the world’s fastest parallel file system per new testing,” and which it said can achieve “up to two times greater throughput than competing parallel file systems.” That’s important for inference operations that must rapidly intake large amounts of data, the company noted.

Also in the software bucket is what Dell calls its Dell Private Cloud software, which is meant to move customers between different software offerings for running servers and storage, including Broadcom’s VMware hypervisors, Nutanix’s hyper-converged offering, and IBM Red Hat’s competing offerings. 

The company claimed Dell Private Cloud’s automation capabilities can allow customers to “provision a private cloud stack in 90% fewer steps than manual processes, delivering a cluster in just two and a half hours with no manual effort.”