[Edaily Reporter Kwon Oh Seok ] On the 21st, the Corporate Research Center of the Korea IR Association analyzed Acryl Inc., a leading AI infrastructure company with the technological capability to resolve GPU network bottlenecks through software, noting that its operating loss is expected to narrow in 2026 due to increased revenue. However, the center did not provide an investment recommendation or a target stock price.

Acryl Inc. is a company that supports the entire process from AI service development to infrastructure operation. It provides “JONATHAN,” an integrated AI platform that supports the training, deployment, and management of large language models (LLMs) for enterprises, and “NADIA,” which optimizes AI models for the healthcare industry.
Kim Seon-ho, a researcher at the Corporate Research Center of the Korea IR Association, identified “resolving GPU network bottlenecks through software” as Acryl Inc.’s key investment point. He analyzed, “As AI models grow larger and distributed training—where multiple GPUs share and process data—becomes more widespread, network efficiency—which handles data transfer between GPUs—is emerging as a key factor determining AI infrastructure performance and investment efficiency, alongside individual GPU performance.”
He continued, “In particular, while the use of RoCEv2—which implements RDMA (Remote Direct Memory Access) in a general-purpose Ethernet environment—is expanding, GPU latency can occur due to interference between training, inference, and data transfer traffic, as well as traffic congestion on specific paths.” He explained, “In such cases, even if high-end GPUs are procured, network bottlenecks can reduce utilization rates.”
He also noted, “Acryl Inc.’s GPUBASE controls data transmission paths and traffic priorities using server-side software without replacing expensive network equipment,” adding, “It alleviates network bottlenecks by distributing large volumes of data across multiple paths and differentiating training and inference traffic.”
Researcher Kim assessed, “The center of gravity in the AI infrastructure market is shifting from InfiniBand—a dedicated network—to Ethernet-based RoCEv2, which offers lower deployment costs and the flexibility to utilize a variety of equipment,” adding, “Given that it reduces dependence on specific overseas equipment vendors and can be applied to existing infrastructure, GPUBASE’s business opportunities are likely to expand.”
Revenue in 2026 is projected to reach 20.8 billion won, a 57% increase from the previous year. The operating loss is expected to narrow to 1.7 billion won, and the operating margin is projected to improve from -16.9% in 2025 to -8.0% in 2026.
Analyst Kim stated, “Revenue growth is expected to be driven by the medical solution Nadia and government subsidy revenue,” adding, “Nadia’s revenue is estimated at 5.1 billion won—a 501% increase year-over-year—based on Central Asian projects such as the construction of health insurance and hospital information systems in Uzbekistan and the digital transformation of medical data in Tajikistan. Government subsidy revenue is also expected to reach 6.2 billion won, a 65% increase year-over-year.”
He went on to emphasize, “In terms of valuation, the company needs to prove its worth through earnings growth,” adding, “We need to see that Nadia’s follow-up projects in Central Asia and the GPUBASE-based AI infrastructure business are actually translating into orders and revenue.”