As artificial intelligence (AI) evolves into "agent AI" that carries out tasks on its own, corporations' concerns are shifting from whether to adopt AI to expense, security, and infrastructure operations. Dell Technologies moved to attack the enterprise AI market by promoting "hybrid AI," which distributes AI compute—once concentrated in data centers and the cloud—to employee PCs and workstations.
Dell Technologies held "Dell Technologies Forum (DTF) 2026" at Coex in Samseong-dong, Seoul, on the 25th. After a welcome address by Kim Kyung-jin, Korea Dell Technologies chair, President Yoo Sang-mo and Matt Dunphy, senior vice president at headquarters, delivered keynotes, and Ju Young-pyo, SK hynix vice president, and neuroscientist Jang Dong-seon also joined as guest speakers.
On the day, President Yoo and Senior Vice President Dunphy said, "It is now time to generate tangible business results through AI." Dell assessed that while AI adoption is spreading rapidly, corporations face new barriers—such as data center infrastructure, expense, security, and regulation—when applying it to real work.
According to the "2026 modern enterprise readiness survey" released by Dell, 87% of responding corporations said they are struggling to respond to a rapidly changing environment. Sixty-six percent said it is difficult to predict how their industry will change in three to five years.
In particular, 72% of domestic respondents said their current data center environments are not sufficient to support large-scale AI and analytics workloads. Eighty-four percent said they have data center modernization plans to consolidate and migrate multiple legacy devices to the latest servers and storage.
In response, Senior Vice President Dunphy pointed to three key tasks: ▲ data center modernization ▲ AI expansion based on safety and trust ▲ implementation of a modern work environment.
Security and regulatory issues are also obstacles to AI's spread. Seventy-five percent of domestic respondents said it is difficult to respond to government regulations on AI data, and 65% said they have experienced AI adoption being delayed or halted due to security and compliance issues. Sixty percent said it is hard to entrust critical company data to Generative AI tools that third parties may access.
President Yoo said, "Many corporations are actively embracing AI as a new growth opportunity, but they are encountering various challenges in applying AI to actual business, including security and regulation, data governance, and infrastructure readiness," adding, "Dell Technologies plans to fulfill its role as a strategic partner that presents a long-term technology roadmap for corporations through an end-to-end portfolio and an open ecosystem, and supports generating business outcomes through AI."
◇ "AI tokens have become a line item on the income statement"… expense burden grows for agent AI
The change Dell is eyeing is the rise in token expense as agent AI spreads. Unlike conventional chatbots that answer a user's question once, agent AI thinks, plans, and repeats tasks on its own, which significantly increases token usage.
Jeong Jae-uk, a Dell Technologies director, said, "As Generative AI spreads, the number of tokens consumed by corporations is growing exponentially," adding, "The problem is that even though the unit price per token is falling, the total volume is surging, so the actual bill is getting bigger."
He added, "From the perspective of corporations, AI tokens are no longer just a technical metric; they have become a line item on the income statement," and "the more you use AI, the more the expense rises, creating a dilemma."
To solve this, Dell proposed a hybrid AI strategy that combines cloud, data centers, and desk-side environments according to the nature of the work, rather than handling all AI compute in the cloud.
The cloud is used for sudden demand spikes or large-scale training, while internal corporate data and key workloads are handled in data centers. Repeated inference or tasks that must not send sensitive data outside are processed directly on employees' PCs and workstations.
Manager Jeong said, "There is no need to rent all tokens from the cloud," adding, "The key is not to choose just one, but to combine the three approaches to fit the workload characteristics."
Dell said, "What we are talking about now is how to handle security, how to handle expense, and how to handle operations," explaining that the issues corporations must solve in the actual operations phase after AI adoption have become more important.
◇ AI compute moves from data centers to 'on the desk'
Dell projected that, with the spread of agent AI, the "desk-side AI" market—where AI inference is performed directly on PCs and workstations used by employees—will grow.
Manager Jeong said, "Where AI is processed is expanding beyond the data center to each and every employee's desk," adding, "the paradigm of the work environment itself is changing."
From the perspective of corporations, the advantage is that reducing cloud usage makes expense easier to predict, and sensitive data can be processed internally without sending it outside.
AI PC penetration is still in its early stage. At a briefing the same day, Dell said the domestic AI PC adoption rate falls short of 40%. It also said corporations are reluctant to disclose actual AI use cases externally due to internal confidentiality and security issues.
In contrast, Dell's enterprise PC business is showing growth. According to Dell, total revenue for the Client Solutions Group (CSG) in the last fiscal year was $14.6 billion, up 17% from a year earlier. Of that, commercial client revenue—enterprise work devices—was $13.0 billion, up 18%.
Citing IDC data, Dell said it ranked fourth in the domestic enterprise PC market in the second quarter this year, and took the top spot in the enterprise desktop segment across both domestic and foreign vendors.
A recent rise in PC prices due to a memory supply crunch is a variable. In response to suggestions that higher memory prices could lead corporations to extend PC replacement cycles or prioritize price over high-performance AI PCs, Dell said, "Fluctuations in component costs can naturally affect product prices."
It added, "Rather than passing market volatility straight on to customers, we are continuously working to minimize the impact by optimizing the supply chain and improving operational efficiency."
◇ SK hynix: "In the era of agent AI, memory must also be stacked upward"
As AI use increases, the memory architecture inside data centers is also expected to change. In a keynote the same day, SK hynix Vice President Ju Young-pyo predicted that simply advancing existing high bandwidth memory (HBM) will not be enough in the long run to keep up with AI's data processing demand as agent AI spreads.
Vice President Ju said, "If you ask whether we can continue doing well with HBM as we are now and still supply sufficiently good performance, unfortunately I don't think so," adding, "because of agent AI."
He explained that, since the current approach of placing HBM flat around the GPU faces spatial constraints, memory will need to be stacked vertically to expand bandwidth in the future.
Vice President Ju said, "I think the near-only answer to increasing bandwidth in proportion to area in the future is stacking."
However, he cited heat as a challenge that must be solved. Given the high heat generation of GPUs themselves, stacking multiple layers of memory on top makes thermal management difficult, so cooperation is needed not only with memory corporations but also with system and infrastructure vendors such as server and cooling providers.
Vice President Ju said, "SK hynix aspires to be a full-stack memory creator, but we do not think we can do this alone," adding, "we will prepare with customers from the product research and development stage and build the ecosystem together."
On the day, Dell held 25 sessions across three tracks—AI, modern data centers, and modern workplaces. Forty-three domestic and foreign corporations and distributors/partners participated, including AMD, Intel, Microsoft, Nvidia, Samsung SDS, and SK hynix.