NPU Startups Face Revenue Test as Customer Acquisition Race Heats Up

*Editor's Note: 'AI PRISM' (Personalized Report & Insight Summarizing Media) is an 'AI-based customized news recommendation and summarization service' developed with support from the Korea Press Foundation. It selects and provides six tailored news stories for each reader type.

1. From Subscription Products to Full Commercialization — NPU Startups Pick Up Speed

- Key Summary: FuriosaAI is accelerating the commercialization of its second-generation neural processing unit (NPU) "Renegade" through collaboration with Samsung SDS and other partners. Renegade can handle up to 7.4 times more simultaneous users than Nvidia's RTX Pro 6000 at the same power consumption, and Samsung SDS plans to launch a Renegade-based NPU-as-a-Service (NPUaaS) in July. Meanwhile, Mobilint received 3 billion won in investment from POSCO DX (022100.KQ) to pursue NPU integration in industrial control systems, and DeepX is deploying its NPU in Lotte Innovate's (286940.KQ) intelligent transportation system (ITS), as each company races to secure customers. Industry observers say NPU startups that grew on the back of government support and investment have now reached a point where they must prove their results through actual revenue in industrial settings.

2. "High-Quality Real-World Data Is Key… Targeting 10,000 Humanoid Shipments This Year"

- Key Summary: China's Agibot, after becoming the first company in the world to surpass 5,000 humanoid shipments last year, is aggressively expanding its data infrastructure with a target of 10,000 units this year. A single robot collects 2.5 to 3 hours of high-quality data per day, meaning that with more than 1,000 robots deployed daily, annual data accumulation on the scale of one million hours becomes possible. Accordingly, Agibot completed its second data collection center — a 12-story facility — in January this year and has broken ground on a third center. Building on the "Agibot World" AI training dataset released in 2024, the company is also expected to pursue full-scale "data industrialization," selling real-world data for use in training other companies' models.

3. "Tariffs, Shipping, Environmental Costs — Agents Have the Answers"

- Key Summary: At the "Cello Square Conference 2026," Samsung SDS presented the outlook that agentic AI — AI that autonomously sets goals and takes action — will become a key solution for reducing supply chain crises and the burden of tariffs, shipping costs and environmental expenses. At the event, Samsung SDS presented three core insights: △automated decision-making control towers, △the rise of digital twins (simulation technology that virtually replicates the physical world), and △the expansion of total-cost-based decision-making. Total-cost-based decision-making is a concept that goes beyond traditional costs such as shipping and tariffs to include carbon emissions, supply chain risk and the opportunity cost of cargo. Oh Gu-il, executive vice president of Samsung SDS's logistics business division, said the company plans to combine field data with AI technology to support customers' rapid decision-making and resilient supply chain construction.

4. Code Leak Hits Anthropic Ahead of Listing — Company Scrambles to Contain Damage

- Key Summary: Anthropic, which is preparing for an IPO within the year, suffered a leak of more than 512,000 lines of source code from its core development tool "Claude Code" through the NPM package registry. Anthropic explained the incident was a product release packaging issue caused by human error rather than hacking or a security breach, and requested takedowns from 8,100 repositories on GitHub. The incident is the second to occur within a single week, following the brief exposure of specifications related to its next-generation AI model "Claude Mythos" on the 25th of last month. The market estimates Anthropic's enterprise value at $380 billion (approximately 567.46 trillion won) and rival OpenAI's at $852 billion (approximately 1,292.65 trillion won), drawing attention to how this incident may affect the IPO competition landscape.

5. From Elder Care to Warehouses — Collecting Data for Physical AI Applications

- Key Summary: Korean private companies are moving to secure physical AI (AI that interacts with physical environments such as robots) behavioral data at sites with clear demand, including care, logistics and manufacturing. ConfigIntelligence Korea has collected more than 100,000 hours of data since the second half of last year at its human data factory in Hanoi, Vietnam, while RealWorld is directly acquiring behavioral data from actual field workers through partnerships with CJ (001040.KS), Lotte Hotel, SK and Japan's KDDI. The government has designated the next three years as the golden window for determining physical AI supremacy and has released a "Strategy for Securing Core Physical AI Competitiveness" aimed at developing general-purpose robot foundation models and generating large-scale training data. Bae Kyung-hoon, Deputy Prime Minister and Minister of Science and ICT, emphasized that "we must mobilize the combined capabilities of industry, academia and research to activate a full-cycle support system from independent physical AI technology development to field deployment."

6. New Drugs and Exports Drive Growth — K-Pharma and Bio Revenue Tops 30 Trillion Won Last Year

- Key Summary: The combined revenue of Korea's top 22 pharmaceutical and bio companies reached 30.26 trillion won last year, growing 12.8% year-on-year. Samsung Biologics (207940.KS) posted revenue of 4.557 trillion won and operating profit of 2.069 trillion won, becoming the first Korean pharma-bio company to surpass 2 trillion won in operating profit. Celltrion (068270.KS) recorded revenue of 4.163 trillion won on the back of an expanded biosimilar (biopharmaceutical generic drug) lineup. SK Biopharmaceuticals (326030.KS) achieved revenue of 706.7 billion won — a 99.1% year-on-year increase — driven by full-scale U.S. sales of its epilepsy drug Cenobamate (brand name: Xcopri). HK inno.N (195940.KQ) surpassed 1 trillion won in revenue for the first time, powered by the growth of its gastroesophageal reflux disease drug K-CAB. Meanwhile, leading companies are also accelerating efforts to secure next-generation growth engines including ADCs (antibody-drug conjugates), obesity therapeutics and AI drug discovery platforms, leveraging their accumulated cash reserves.