ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
News2 min read·2026-10-20DisconfirmAI Editorial

Cisco Korea Executive Director for Customer Experience Kim Jae-deok: AI Infrastructure Bottlenecks May Be More Organizational Than GPU-Related

At a Cisco Korea media tech session, Kim Jae-deok argued that even after deploying high-performance GPUs and networks, the delays caused by siloed security, server, application, and planning teams working alongside legacy systems may exceed technical bottlenecks. He cited the shift from spreadsheet-based asset tracking to real-time telemetry as an example of how poor visibility slows response.

ImportanceMaterialEvidenceE3 inspectableWrite-upStandard

Original publication time 2026-10-20 16:13 GMT+8

After deploying high-performance GPUs and networks, AI infrastructure still fails to deliver expected results — Kim Jae-deok believes the bottleneck may lie more in organizational coordination than in hardware itself. He is Executive Director overseeing Customer Experience at Cisco Korea, and made this judgment at a Cisco Korea media tech session held on October 20, 2026 in Samseong-dong, Seoul.

His core argument is that in enterprise environments where legacy systems and AI workloads coexist, the delays produced by security, server, application, and planning teams operating independently can exceed bottlenecks at the GPU or memory level. In other words, even if the infrastructure itself is fast enough, cross-departmental decision-making and incident response may still prevent actual services from going live.

Kim emphasized that most enterprises still carry a high proportion of legacy infrastructure, so how AI infrastructure coexists and coordinates with existing systems is the key to success or failure. Enterprise IT environments are simultaneously hosting traditional systems and AI agent workloads, connecting on-premises and multi-cloud, and continuously tracking which workloads run where and what their performance and security status is — that itself is an operational challenge.

Sign in / Sign up
ReadingResearchRadarInvestment framework

He used asset management as an example of how poor visibility slows response. Previously, device serial numbers, software versions, and end-of-support dates were often maintained separately in Excel spreadsheets; as device counts grew and physical and logical configurations kept changing, it was difficult for people to keep information current. Cisco IQ collects this data through telemetry in real time, helping operators identify unmanaged assets, devices already past end-of-support, and configuration errors. Kim said that in the past, this kind of work required significant operational manpower and time, and the environment itself would change again during the inspection cycle.

The boundary of this observation deserves attention: Kim was speaking at Cisco's own CX strategy presentation, and his diagnosis is connected to Cisco's full-lifecycle service positioning. The judgment that organizational silos create larger bottlenecks than hardware has independent merit, but the article provides no quantitative evidence, such as what proportion of total deployment cycles is consumed by organizational coordination delays.

The practical takeaway for readers is: when enterprise AI investment falls short of expectations, the first step is not to purchase more compute, but to examine the coordination processes between legacy systems and AI workloads — who approves changes, who responds to incidents, and whether asset status is visible in real time. Kim's judgment offers a concrete diagnostic starting point: shifting attention from the technology checklist to the organizational interface.

Sources

  • 原始出处zdnet.co.kr↗
Continue researching →