Turning data into intelligence for the AI native era

The current "state of play"

Before highlighting exactly what Dell is doing, it's worth noting the current "state of play" as organizations globally look to build out their AI ambitions. Clearly every customer is different, but in general, here's how I see some of the key challenges shaping up.

• Data as a primary differentiator. In a world where almost every organization is essentially using the same AI models, differentiation needs to come from elsewhere. Increasingly, that will be through their data. That's simple to say, but for most, it's a much more difficult thing to do. They must learn how to harvest and assimilate the "right" data from their enormous, fragmented, distributed and immensely varied data sets. Our research already flags data-related issues, but for most, it's seen as a top challenge. 68% of IT leaders said that data management was the most challenging part of implementing AI in production. And this is before AI has been operationalized at scale within most organizations.

If building the compute/GPU environment (the AI Factories) was the first big step in turning AI's promise into reality, then the task of effectively, efficiently and securely connecting an organization's myriad data assets to agentic and other AI workflows is the next greatest challenge. Simplifying the entire process to make it repeatable, agile and work at near real-time speed to increase time to value, without increasing risk, is understandably daunting, but is where organizations must reach next.

• AI is a hybrid workload. It's almost a cliché to say that data has gravity, but that's because it's true. And while data for training LLMs is often concentrated in a single location (chiefly, the cloud), this may not be practical, preferable or even allowable for inference-oriented workloads -- for regulatory and sovereignty reasons.

Agentic AI needs to run wherever you have data. And that's everywhere; in the cloud, in the data center, at the edge and, increasingly, at the desk-level. AI began in the cloud, and of course that will remain a fundamental component; but 75% of IT leaders say AI is a hybrid workload that uses on- and off-premises data. Moreover, for many organizations, their most valuable data resides on-premises. Finding a way to effectively, but securely, use this data in AI workflows is the difference between merely "using" AI and having it drive real-world differentiation.

• Token anxiety is real. While most organizations are not deploying GPUs at scale, almost all are or soon will be consuming GPU-based agentic services. And though per-token costs are falling, overall token generation is exploding.

Just like shadow cloud became a significant stealth cost a decade ago, shadow AI is now a cause for anxiety for company accountants. Engineering teams can burn through thousands of dollars in token usage in just a few days. As a result, organizations are seeking effective ways to understand, predict and better manage costs to enable broader usage, rather than limit it.

• Simplified, flexible AI infrastructure deployment at scale. For the growing number of organizations that are deploying AI infrastructure at scale (such as large enterprises, public cloud service providers and education institutions), two challenges are rapidly emerging. First, how to ramp up new infrastructure as quickly as possible to satisfy burgeoning demand. And second, what to do if the requirements change?

AI is proving to be something of a mercurial technology -- largely due to its nascent, fast-evolving nature. However, it is vulnerable to its own hype. This makes anticipating future requirements extremely difficult. How can those building the AI factories of the future insulate themselves from these swings?

Evolving the AI Factory for the "useful" phase of AI

It's no overstatement that Dell is maniacally focused on addressing these and other challenges as part of its AI strategy. The sheer number and range of AI-related announcements the company made at Dell Technologies World across its AI Factory portfolio (which now includes major deployments with the likes of Eli Lilly, Samsung Electronics and Mistral AI) are testament to its obsession with leading the AI conversation. While there were far too many announcements overall to cover in detail (check out my colleague Scott Sinclair's blog that covers Dell's data center modernization announcements at the show), here are a few that stood out for me:

• Enhancements to the Dell AI Data Platform. Dell's offering for customers looking to fast-track an AI-ready data foundation. These improvements are designed to transform enterprise content into rich datasets ready for inferencing. It's a layered approach that spans data orchestration (with technology from its recent Dataloop acquisition), data transformation (analytics, search, etc.) and data storage (PowerScale, ObjectScale and the new Lightning parallel file system).

New enhancements boost orchestration and search, add GPU-based SQL analytics, and increase storage density using support for the new ObjectScale X770 ultradense appliance. These incremental additions highlight the degree to which Dell is placing the AI Data Platform at the center of its data story for enterprise-ready AI.

• Dell Deskside Agentic AI. A tight integration of Dell's high-performance workstations with Nvidia NemoClaw. This offers customers the option to securely build and run autonomous agents using local data that stays on the device. It's an innovative and creative offering that will likely appeal to specialized groups -- think software engineering teams and the like -- enabling them to scale token usage with more control versus variable cloud cost models.
• Dell PowerRack and ExaScale Storage. PowerRack is a new, fully-integrated system for compute, networking and storage. It features thermal design, power management and other software optimizations built in, enabling customers to accelerate AI and HPC rack-scale deployments without the overhead of component assembly.

Meanwhile, Dell's new ExaScale Storage enables customers to flexibly deploy Dell's storage software at extreme scale on Dell PowerEdge Servers. It will support PowerFlex block storage in addition to Dell's file and object storage software platforms.

• Broader Ecosystem support. Dell's AI portfolio is broad and deep, but partners are an essential aspect of its strategy. Both to enable customer choice while also simplifying deployment, security and operations through integrations wherever possible. Nvidia, of course, remains a foundational partner, but new and expanded partnerships were announced with Google (enabling Geminin3 Flash models to run on Google Distributed Cloud on Dell PowerEdge servers), Hugging Face, OpenAI, Palantir, SpaceXAI and ServiceNow, among others.

Of course, Dell operates in a highly competitive space -- and all its key rivals across the infrastructure spectrum are also investing heavily in AI-related software and innovation. However, there's one more aspect Dell brings to the conversation here that I think is significant for its customers, and that's the extent to which Dell is transforming itself into an "AI native" organization. Dell's investment in itself as "customer zero" signals to prospective customers that it practices what it preaches. And to a significant degree. The company is aggressively applying AI to simplify and streamline processes across its entire business, and in the process, extracting valuable learnings that it can feed into future product decisions and customer conversations.

"AI is no longer a feature -- it's the operating model of the modern enterprise," stated Michael Dell. As the AI market continues to evolve at breakneck speed, customers increasingly have access to products and tools that help them make this transition effectively.

Simon Robinson is principal analyst covering infrastructure at Enterprise Strategy Group, now part of Omdia.

Enterprise Strategy Group is part of Omdia. Its analysts have business relationships with technology vendors.