Dell’s Top 10 AI and a Learning Approach to Scaling Enterprise AI
As businesses move from AI experiments to operationalization, Dell Technologies positions themselves as both vendors and case studies. COO Jeff Clarke used his Dell Technologies World Keynote to outline internal AI transformations that provided a window into the company’s AI products as well as the company’s size adoption mechanism.
Clark set the stage for the entire AI landscape from Dell with a list of the top 10, heading towards the top of his lecture. “You know I love the good top 10 list,” he said. So here’s the list and clearly edited.
“Changes, changes, changes, roadmap, roadmap, roadmap, acceleration, acceleration.” This includes the commercialization of the Nvidia B200, GB200, B300, and GB300 platforms on Dell hardware. “A truly capable multimodal model and large-scale context LLM are improving the results significantly.” “The technique is moving forward at a rate close to exponential. Quantization, distillation allows for more capabilities, smaller models, and faster training of new models.” “One of my personal favorites – inference and inference models is much more computing-intensive than we thought a year ago. At least 100 times more computing concentrated than we thought at least a year ago and it could grow from there.” “Tokens, tokens, tokens… We’re building a token factory. In 2024, 25 trillion tokens were generated. By 2028, the number would be 35,000 trillion tokens. “A very capable NPU is included in our PCs. It will be destructive. It will enable Rag at the very edge of the network.” “Even though the cost of training the next new generation model has increased by 10 times, there are more base models today than a year ago.” “The cost per token has actually decreased by four orders over the past four years. Therefore, continuous training for new models is 10 times, and the percentage of token costs has decreased by four orders over the past four years.” “Agents are the answer to everything… By 2028, a third of all interactions with Gen AI will use autonomous agents to complete the task.” We’ll go into more detail later. “The world’s largest companies have moved beyond proof of concept.”
The key here is that AI scales fast, rebuilds infrastructure demands and pushes calculations to every edge of the network. We see exponential growth in model complexity, inference strength, and token amounts. This is to force research into IT systems from PCs to hyperscale data centers.
Dell laid out its own enterprise AI journey
Dell is recognized as a complex and huge operation, and Clarke detailed the company’s own internal AI journey and also provided useful insights to its customers about AI benefits. He cited the votes of 3,800 corporate decision makers to understand their ideas about AI. He pulled out two data points. The top two technical challenges are the fundamental tasks regarding the scale of implementation and the data required to operate AI. He also noted that 39% of all data center power supplies are unused. As we progress, put those ideas in your mind.
“In the spirit of Misery of Misery Love Company,” Clark said. He said Dell had over 900 “AI projects” within the company, addressing the general lack of next data governance and clarity and purpose.
So, Clark used something like a list to lay out the foundation structure that guided (and did) Dell’s internal AI ambitions.
First, Dell had to define an AI data architecture and build an enterprise data mesh that connects all the relevant data. “The process had to be simplified, standardized and automated. When I applied AI to crappy processes, it became very clear, with faster and especially bad things getting crappy answers.”
Second, the AI strategy and accompanying use cases should match what is most important to Dell. Without the alignment, we wouldn’t have done it, Clark said. And finally, a meaningful ROI had to be committed. “We weren’t going to fund it unless we were willing to register for real dollars, true efficiency and productivity.”
Before diving into the use case, let’s go back to Clark’s mention of what’s most important to Dell. He details the list: end-to-end product strategies, moving engines to markets, global supply chains, and global service capabilities.
In that last point, Clark explained how Dell leveraged data and AI to promote efficiency, productivity and quality for a global service organization consisting of tens of thousands of staff from over 170 countries supporting more than 250 million on-site assets. He said that the investigation of the underlying data revealed five important data sets that held important insights, including case history, knowledge base, self-reported information from the product, dispatch logs and repair information, when properly connected.
“All the data that exists in us…it was everywhere,” Clark said. And the company had previously tried to connect and utilize these datasets. “It wasn’t until advances in generator AI that the reality of extracting value from all five of these datasets simultaneously came to fruition.”
The result was a digital “shoulder genius” that could be used by members of the Global Services team. “We have created service assistants for all service members within our organization. It’s very extraordinary for us,” Clark said the agents will close more cases faster and increase resolution rates. Dispatch rates will drop, and repetitive shipping fees will drop even further. “And most importantly, the most important thing is that customer satisfaction is rising. So we can solve more cases, which can be resolved faster, whether it’s the first time or the second time we’ve sent some. These rates are happier customers and we didn’t need the latest models.
Enterprise AI Playbook – Thinking about cases and things
Based on the big picture and Dell’s own experience, Clark identified it (in list format). Six common enterprise use cases are poised to disrupt AI, and surprisingly, we’ve listed five things companies need to think about when increasing their investment. With more ADO:
Here are six common enterprise use cases where AI is poised to be confused:
Support Support Content Creation and Management Natural Language Search Design and Data Creation Code Generation Document Automation
And five things companies need to think about:
“It’s time to keep busy…the threat is existential…if you haven’t started, you’re behind.” “There’s a one-size-fits-all approach.” “An existing data center has a lot of power, cooling and space.” “You don’t need the latest models. You don’t need the latest GPU to get started.” “The right use cases within your organization have a compelling ROI.”
Clark’s point is that when reading between lines, technology is not the barrier to enterprise AI. It is organizational discipline, building preparation, and ruthless ROI pursuit. His keynotes reflect Dell’s strategy of experience as well as infrastructure. The message to the CIOS is clear. Don’t wait for it to be perfect. Optimize what you already have, start with an aligned use case and iterate through the results.
Find out more about our Dell Technologies World research. And here’s the video from Clark’s keynote if you’re so leaning.
