Production Challenges for Enterprise AI Workloads Highlighted at Dell Symposium
As enterprises seek to move artificial intelligence workloads into production, chief information officers are grappling with significant challenges. Key among these challenges are managing costs, scaling AI infrastructure, and establishing governance models that ensure agents operate within secure parameters.
Discussions focused on cost and control took center stage at Dell Technologies Inc.’s AI Leadership Symposium. Enlarging AI workloads can lead to soaring infrastructure and model costs, and the deployment of autonomous agents raises critical questions about security, oversight, and data management.
In parallel, companies are exploring varied methods to operate AI solutions, from private cloud setups to models that execute directly on devices. This evolution is increasing demand for enterprise AI infrastructure capable of meeting a wider array of deployment needs.
“No one is asking whether to adopt AI; that’s pretty much a done deal,” commented John Furrier, executive analyst at theCUBE Research. “They’re asking who controls it, where it runs, and whose law it operates.”
During the symposium, Furrier engaged with industry leaders from Dell, Deepgram Inc., Singulr AI Inc., MisaLabs Inc., and Intel Corp., among others. These conversations delved into the infrastructure, partnerships, and computational power necessary to scale AI effectively. (* Disclosure below.)
See theCUBE’s complete keynote analysis from John Furrier:
Insight #1: Rising Costs Prompt Rethinking of AI Budgets
Cloud token expenses continue to present a major burden on enterprise AI budgets. This situation has led many organizations to seek strategies for alleviating this financial strain, as noted by Helen O’Sullivan, an AI business development manager and solutions specialist at Dell.
“Many customers I support in the Bay Area are facing the immediate challenge of exhausting their token budget within weeks that were intended for several months,” she stated in a conversation with theCUBE. “They continue to use advanced models, but shifting some workloads to open-source solutions and on-premise infrastructure can reduce token pressure significantly.”
O’Sullivan and Tim Wood, sales director for Intel’s Northwest region, emphasized the importance of optimizing existing data center capacity when making investment decisions regarding enterprise AI infrastructure.
“Many enterprises already have central processing units in their data centers,” Wood explained. “There may be underutilized resources that could be leveraged for next-generation applications and workloads.”
Watch the complete video interview with Helen O’Sullivan and Tim Wood:
Insight #2: Increased Speed Heightens AI Governance and Security Concerns
As enterprises increasingly deploy AI agents, they must develop robust mechanisms to monitor and ensure that these agents adhere to company policies. This necessitates the introduction of runtime governance that evaluates AI agent behavior during execution, according to Shiv Agarwal, CEO and co-founder of Singulr AI Inc.
“Even a simple action, such as an agent accessing the internet for data or sending an email to a private account, can be hidden within an agent’s operations,” Agarwal noted during the symposium, alongside Sri Viswanath, CEO of Sycamore Labs Inc., and Victor Jakubiuk, co-founder of MisaLabs Inc.
Effective AI governance also entails establishing clear boundaries regarding data access and processing protocols. According to Viswanath, CIOs often favor virtual private clouds to maintain sensitive data within regulated environments.
“They strive to architect systems that ensure all of the data remains in-house,” he added. “There is a lack of trust in sending data outside, as there are uncertainties about where it may end up.”
The rapid pace of AI innovation complicates these governance challenges. Organizations must ensure their systems behave as expected even as AI capabilities progress rapidly, Jakubiuk pointed out.
“AI development is accelerating at an unprecedented rate,” he said. “Enterprise leaders must remain updated with the latest advancements.”
View the complete video interview featuring Victor Jakubiuk, Shiv Agarwal, and Sri Viswanath:
Insight #3: Silicon Valley’s DIY Culture Shapes Enterprise AI Infrastructure
Deepgram is collaborating with Dell and Fortanix Inc. to extend its voice AI models from the cloud to consumers’ devices. This initiative encompasses potential applications like interactive voice response systems for small businesses and AI-powered laptops designed for educational purposes, as stated by Nick Mann, Deepgram’s staff technical program manager.
“We aim to run our models natively on devices without internet access,” Mann remarked, emphasizing that AI must evolve to meet users where they are and how they intend to utilize the technology.
See the complete interview with Nick Mann:
Deepgram’s strategy is indicative of a larger trend toward creating AI solutions that align with customer preferences for building and deploying technology. Silicon Valley’s technically adept buyers often lean toward custom builds rather than off-the-shelf solutions. In response, Dell is expanding its partner ecosystem, according to Sean O’Connor, regional manager of large enterprises, who spoke at the symposium.
“Our ecosystem is evolving,” O’Connor said. “Today we partner with Intel Corp., SHI International Corp., and Microsoft Corp. Additionally, we are increasingly engaging with private equity and venture capital partners.”
Watch the complete interview with Sean O’Connor:
Catch up on theCUBE’s video coverage of Dell’s AI Leadership Symposium:
https://www.youtube.com/watch?v=videoseries
(* Disclosure: TheCUBE is a paid media partner for Dell’s “AI Leadership Symposium.” Neither Dell, the event’s sponsor, nor other sponsors exert editorial control over content produced by theCUBE or SiliconANGLE.)

