4 insights from Dell’s AI Leadership Symposium: Cost and control reshape enterprise AI deployment strategy
SiliconANGLE Cheryl Knight ● Covered by 5 sources
Dell’s AI event said the quiet part out loud: production, not demos, is the real fight. Now it’s about cost, control and where AI is allowed to run.
Based on reporting by SiliconANGLE, Cheryl Knight — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The shiny part of enterprise AI is over. At Dell Technologies’ AI Leadership Symposium in San Jose, the message from vendors and analysts was blunt: getting models into production matters far more than showing a proof of concept. The new pressure points are cost, data access and control, and they’re already reshaping where AI gets deployed and who gets to govern it once it’s live.
Dell is leaning hard into that shift. Sean O’Connor, regional manager of ENT West, large enterprise and acquisition at Dell, said long-standing chip partnerships help the company roll out 350 racks in two weeks. That kind of speed matters because customers are no longer just asking for AI infrastructure; they want something that can move from experiment to working system without turning into a logistics project.
Voice AI is following the same split between centralized cloud and local control. Deepgram trains its speech-to-text, text-to-speech and voice agent models on Dell infrastructure, and many of its customers host those models themselves. Nick Mann, Deepgram’s staff technical program manager, said the next step is running them natively on AI-enabled laptops, air-gapped from the network, so the software can sit closer to the user.
The governance story is getting sharper too. Victor Jakubiuk of MisaLabs, Shiv Agarwal of Singulr AI and Sri Viswanath of Sycamore Labs all pointed to the same problem: autonomous agents don’t just need policies on paper, they need runtime oversight. Enterprises want those agents kept inside their own virtual private clouds, but fixed rules alone are not enough once the systems start acting on their own.
And then there’s the bill. Helen O’Sullivan from Dell and Tim Wood from Intel said customers who burn through months of token budgets in weeks are testing workloads on-premises and rediscovering CPUs for agentic AI. The theme running through all of it is simple: the next AI purchase is as much about restraint as acceleration.
My take — AI-written commentary, not fact-checked reporting
The industry keeps selling AI as an expansion story, but enterprises are behaving like accountants with a security problem. That’s why the real winners won’t just be the loud model vendors; they’ll be the people who make AI cheaper, local and easier to fence in. Fancy demos are getting nudged aside by the older, less glamorous business of control.
Read more about this at: SiliconANGLE