DEEP RESEARCH · ICRAFT

iCraft 4Q25 Review: AI Infrastructure Supercycle and the Re-rating Case

Testing whether the company is shifting from network integration to turnkey AI-infrastructure engineering

Published: 2026-02-09 · Earnings review / AI infrastructure · Original Naver Blog post

0. Bottom line first

Official fact: The source summarizes iCraft's February 9, 2026 disclosure of 4Q25 preliminary results as revenue of KRW 109.0 billion and operating profit of KRW 11.4 billion. The Kakao AI-infrastructure contract is presented at KRW 35.5 billion, and the Naver Cloud Infiniband contract at roughly KRW 12.3-12.36 billion.

Interpretation: The key is whether iCraft can move from a low-growth network-integration multiple to an AI-infrastructure solution multiple. But because the business is project-based, revenue recognition can be lumpy; annualizing the fourth quarter mechanically would be risky.

1. Business essence: engineering, not just distribution

The source starts from the idea that iCraft should be viewed not as a simple IT equipment distributor, but as an engineering company that designs and optimizes complex network environments. Its mid-sized enterprise status provides creditworthiness and financing capacity needed to execute large infrastructure projects.

2. Moat: Nvidia Elite Partner and Infiniband design capability

• Nvidia Elite Partner: The source states that iCraft holds the top-tier Elite Partner status in Nvidia's partner program. In an AI-chip shortage, this status has strategic value because it can support allocation priority and delivery reliability.
• Infiniband capability: AI data-center performance depends heavily on GPU-to-GPU communication speed, not just individual GPU compute. Based on more than 20 years of telecom network work, iCraft is described as one of the few Korean companies able to design and optimize complex Infiniband architectures.
• Value-chain position: Nvidia's dominance can limit purchase-price negotiation, but Elite Partner status helps secure volume, while customers prioritize technical reliability and delivery over price.

3. Direct drivers of the fourth-quarter surge

The 4Q earnings surprise is interpreted as the result of large contracts concentrated in the second half of 2025, with equipment delivery and revenue recognition falling into the fourth quarter.

Interpretation: Aswel AI matters strategically, not just for size. It extends iCraft from GPU hardware and Infiniband networks into data and solutions, creating a fuller AI-infrastructure value chain.

4. 2026 end market: Sovereign AI and Blackwell

Sovereign AI is the push to build AI infrastructure reflecting a country's language, culture, regulations, and data sovereignty. The source expects both the national AI computing center and private data-center expansion to increase GPU server and network equipment demand across public and private customers.

Blackwell systems require liquid cooling and more complex interconnect technology. That increases the value of system-integration partners that can design, install, optimize, and maintain systems, rather than merely reselling equipment.

5. Q/P/C view of earnings sensitivity

6. Financial health and shareholder returns

Debt ratios had increased as of the 3Q25 report, but the source views the quality of the debt as preparation for growth rather than a warning sign. Inventory of about KRW 18.0 billion at the end of 3Q was equipment secured for large 4Q revenue recognition, while payables and short-term borrowings were working capital for contracted revenue.

In 4Q, revenue recognition and large advance payments from customers such as Kakao likely improved cash flow. iCraft decided a KRW 100 per share cash dividend for 2025, with a 3.6% dividend yield and total dividend amount of roughly KRW 1.2 billion according to the source.

7. Valuation, risks, and my view

The source argues that iCraft should be valued not as a legacy NI company but as an AI-infrastructure solutions company. It cites domestic AI infrastructure and GPU-related peers such as XiiLab and MDS Tech at average PERs of 30-50x, and sees iCraft as undervalued at a market cap of roughly KRW 54.5 billion. Annualizing 4Q25 operating profit of KRW 11.4 billion implies nearly KRW 40 billion of operating-profit capacity, and even a conservative 2026 net-income assumption of KRW 15-20 billion would make the current share price difficult to explain, in the source's view.

• Revenue lumpiness: Large project-based orders can create major quarterly swings. The fourth quarter should not be assumed to repeat every quarter.
• Supply-chain risk: Blackwell yield problems or global supply delays could delay delivery and revenue recognition even when orders are secured.
• Overhang check: The source says no special issue is evident in the report, but CB, BW, or other potential selling supply should be monitored.

The source lands close to a Strong Buy conclusion. I am positive as well, but the sequence to verify is quarterly order disclosures, Blackwell supply schedule, Kakao and Naver Cloud remaining revenue recognition, Aswel AI consolidated margin, and inventory/payables turnover. The re-rating case depends on those five items continuing.