AI · SEMICONDUCTOR · FACT CHECK
SKD WIREMostly True
South Korea's government-backed GPU infrastructure is currently operating at utilization rates in the 50-percent range, and measures are being prepared to reclaim and reallocate GPUs that remain underused for two months, according to remarks by Baek Byung-hun, who oversees AI computing policy.
Baek said GPUs provided through government support programs are showing utilization in the 50-percent range, leaving significant computing capacity idle. To address this, he indicated that GPUs found to be underutilized for a period of two months would be subject to recall and redistribution to other users or projects with greater demand.
The policy aims to raise the efficiency of the national GPU pool by shifting computing resources away from low-demand holders toward AI developers and researchers who need them, rather than leaving capacity tied up.
The remarks come amid broader efforts to maximize the return on public investment in AI computing infrastructure. With GPU demand among Korean AI developers far exceeding supply, low utilization of government-supported hardware has been a recurring point of criticism. The two-month threshold introduces a concrete mechanism for reallocating capacity that would otherwise sit idle.
The government has not released detailed figures breaking down utilization by program or provider, and the scope of the recall-and-redistribution measures remains subject to further implementation details.
The claim that government-supported GPUs are running at roughly 50-percent utilization and that underused units for two months will be recalled and redistributed, as stated by Baek Byung-hun, is Mostly True.
