AI NewsWaveLablup Pursues KOSDAQ Listing, Touting GPU Operating Efficiency
Key points
- Lablup holds IPO briefing and pursues KOSDAQ listing
- Case of raising GPU utilization from 30% to 85%
- Expanding from GPU management to token cost management
Lablup, an AI infrastructure operations software company, is pursuing a listing on the KOSDAQ market. According to Techworld News and Platum, Lablup held an initial public offering (IPO) investor briefing on Oct. 19 at the Conrad Hotel in Yeouido, Seoul, and VentureSquare reported that it unveiled a global market expansion strategy ahead of the listing.
Byline Network reported that Lablup is targeting the market for AI infrastructure operating efficiency with GPU resource partitioning virtualization and automatic failure recovery technology. Performance figures were also presented. Byline Network reported that in an actual customer case, GPU utilization rose from 30% to 85% and the training time for a large-scale AI model fell from 120 days to 66 days. VentureSquare reported that the company said it improved GPU utilization from 30% to 85% in a specific operating case, and its headline expressed this as a 2.8-fold improvement in GPU operating efficiency.
According to VentureSquare, the flagship platform Backend.AI is operated at more than 120 sites in Korea and abroad. Platum pointed out that a common concern among companies running AI infrastructure is that a large share of the expensive GPUs they buy sits idle.
Techworld News reported on the direction of business expansion. Lablup plans to widen its business from managing AI accelerators such as GPUs to model inference and token usage and cost management, and to target the global neocloud market. It also plans to increase investment in overseas business and technology development on the occasion of the listing.