AI Infrastructure Enters a Third Act as Deskside Gains Ground
IT DAILY ·
✦ AI Summary
Enterprise AI infrastructure is rapidly shifting from a two-tier cloud-and-on-premises model to a three-tier structure that includes edge and deskside.
The change is being driven by the rise in AI inference workloads and by generative AI use moving from training to everyday work and on-site inference.
Cloud handles large-scale training and general-purpose computing, on-premises systems handle core data management and governance, and edge and deskside handle everyday inference and real-time tasks.
The enterprise AI infrastructure market has long been organized around cloud and on-premises servers. The existing model was a two-tier structure of cloud and on-premises systems. But the market is moving toward a three-tier structure that includes edge and deskside, and the pace of that shift is fast.
This change stems from a growing recognition that relying on only the two-tier model is inefficient for responding to the surge in AI inference workloads in the field. Public cloud handles large-scale pretraining of AI models, while on-premises data centers are responsible for managing enterprise-wide core data and governance.
At the core of the structural shift is a move centered on generative AI use. The focus of generative AI use is shifting from model training to everyday work and on-site inference. As a result, the structure of enterprise AI infrastructure is also changing in a direction that makes it difficult to cope with centralized processing alone.
The approach of sending all data in a single stream to a central data center or remote cloud for processing is showing its limits. This method increases network traffic costs and also creates Latency. For these reasons, the enterprise AI infrastructure market is rapidly shifting from a two-tier model centered on cloud and on-premises systems to a three-tier structure that includes edge and deskside.
Security requirements are being tightened, especially in sensitive industries such as finance, healthcare, and manufacturing, to protect internal confidential information and IP and to block data leakage outside the company at the source. In this trend, adoption of Agentic AI is also expanding.
Agentic AI is being applied to user work environments and factory equipment sites, and it has the characteristic of carrying out autonomous tasks on site. As a result, the spread of Agentic AI is increasing demand for real-time immediate response that bypasses the central network, and the need for local computing hubs is also rising sharply. At the same time, edge and deskside infrastructure is drawing new attention as a way to fill functional gaps in existing infrastructure.
Deskside infrastructure, based on high-performance AI workstations and industrial edge appliances, ensures strong real-time performance and service continuity through on-site processing without central network Latency and millisecond-level data processing. It also keeps sensitive raw data from leaving the terminal, allowing strong security comparable to an Air-gapped environment. In addition, distributed processing for sLLM fine-tuning and inference can reduce cloud API call token costs, offering clear advantages in terms of corporate ROI as well.
Industry observers expect AI infrastructure to settle not into a single environment, but into a hybrid three-tier model consisting of cloud, on-premises data centers, edge, and deskside. The defining feature of this hybrid three-tier model is the clear division of labor among each tier.
Within this structure, cloud handles the foundational training of large foundation models and general-purpose computing. On-premises data centers are responsible for managing enterprise-wide core data and corporate governance. Edge and deskside are in charge of everyday inference and agent-based real-time tasks. This three-tier structure has the character of a three-way collaborative system.
As the AI adoption stage moves beyond PoC into enterprise-wide operational implementation, enterprise infrastructure is expected to be reorganized away from centralized investment and toward distributing computing resources to physical hubs closest to users. Accordingly, the pace at which companies restructure their infrastructure is expected to accelerate.
Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=241822
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.