[Chip Talk] "Research Is Over, Now They're Making Real Money"... Korean-Made NPUs Win Over Lisa Su as Monetization Begins in Earnest
Korean-Made NPUs Emerge as GPU Alternatives with Superior Power Efficiency
Rebellions, FuriosaAI and DeepX Enter Phase of Generating Tangible Business Results
Support from Optimization and Back-End Partners Also Helps Drive Commercialization
The most eye-catching item on the schedule of AMD CEO Lisa Su, who visited South Korea on October 7, was a private meeting with executives from leading South Korean AI semiconductor startups. The list included prominent fabless companies such as DEEPX, Rebellions, and FuriosaAI. The reason a global tech giant seeking to challenge Nvidia's dominance reached out to South Korean fabless companies was clear: AMD aims to build a heterogeneous computing infrastructure that combines its CPUs and GPUs with South Korean-made NPUs, in a strategy to shake Nvidia's stronghold.
"Just one or two years ago, domestically developed AI semiconductors were still at the 'test chip' stage, supported by government project grants or public demonstration programs. Now they have entered a phase of generating meaningful business results," an industry official from a fabless company said in a phone interview with The Asia Business Daily on October 18. The chips have moved beyond the research stage to handle traffic from real commercial services, and companies have begun supplying mass-produced chips worth tens of billions of won to overseas data centers, with the profitability of their businesses starting to show up in the numbers.
Lisa Su, chair and chief executive officer of AMD, greets representatives of South Korean AI companies at the “AI Semiconductor Company Roundtable” held at the Four Seasons Hotel in Jongno-gu, Seoul, on October 7, 2026. Photo by Kang Jinhyung.
View original imageThe shift in the AI market paradigm has helped domestically developed NPUs prove their market potential and attract interest from global tech giants. In the early market, when companies were developing large language models (LLMs) such as ChatGPT, "training" - feeding in massive amounts of data - was the focus. More recently, "inference," or generating responses to actual user requests, has emerged as a key workload, accounting for more than 40% of data center power consumption and server capacity.
The problem is cost. GPUs, which also incorporate circuitry for general-purpose graphics processing, consume ever more power and drive hardware operating costs, or total cost of ownership (TCO), up exponentially when repeatedly performing inference tasks. By contrast, NPUs and application-specific integrated circuits (ASICs), designed specifically for AI computations such as matrix multiplication, are vastly more power-efficient than GPUs and can dramatically reduce chip procurement and maintenance costs.
According to market research firm TrendForce, shipments of ASIC-based servers, including those equipped with NPUs, are forecast to grow 44.6% this year, nearly three times the 16.1% growth projected for general-purpose GPU servers. The share of specialized semiconductors other than GPUs in the data center AI semiconductor market is also expected to expand from 18% last year to 41% by 2030. This is the structural reason that big tech companies and cloud service providers are beginning to spend on domestically developed NPUs.
The performance of South Korean AI semiconductor companies is showing up in the numbers. Rebellions, for example, has deployed its first-generation NPU chip, Atom, in key SK Telecom services such as A.Dot. It now processes an average of 14 million requests a day, equivalent to 4 billion computational tokens. This gives the company real-world operating data comparable to that of large-scale commercial deployments by big tech companies. Following its merger with Sapeon Korea, Rebellions is preparing its second-generation product, Rebel100, which combines Samsung Electronics' 4-nanometer process with 144GB of fifth-generation high-bandwidth memory (HBM3E). It has also finalized plans to supply the chips to Saudi Aramco and deliver more than 100 racks to a data center in Tokyo, Japan.
FuriosaAI has also scored its largest export deal since its founding, supplying an overseas data center. Its second-generation chip, Renegade (RNGD), featuring TSMC's 5-nanometer process and 48GB of SK hynix HBM3, has completed initial production of 4,000 units and begun full-scale shipments. It consumes just 180 watts while running high-performance models. A key advantage is that it can be plugged directly into a conventional air-cooled data center, without separate liquid-cooling equipment. The company has also secured a foothold in the European market by signing a KRW 123 billion contract to supply more than 8,800 RNGD chips in a single delivery to an AI data center in Stockholm, Sweden.
DEEPX has secured both strong yields and orders in the on-device and physical AI markets. The DX-M1, which powers robots, smart factories, and intelligent CCTV systems at ultra-low power consumption of around 1 to 5 watts, has achieved a yield of about 90% since entering mass production. On the back of this, the company has secured 77 purchase orders (POs) worth approximately $13 million (about KRW 1.75 billion) across nine countries, including South Korea, the United States, China, and countries in Europe. It has signed a three-year agreement to supply mass-production components to Taiwan's Aetina and has begun developing its next-generation product, the DX-M2, which will use Samsung Electronics' 2-nanometer process to run generative models in compact robots.
Partners that supported software optimization and back-end processes also played a decisive role in helping domestically developed NPUs cross the threshold into commercialization. For example, Nota, a company specializing in AI model optimization, has technology that reduces the size and computational requirements of large AI models while maintaining their performance. Nota has supplied this optimization software to FuriosaAI for its data center chips under a licensing agreement, improving the compatibility of domestically developed chips.
ITEK, a semiconductor testing specialist, has worked with DEEPX since the prototype development stage and is responsible for system-level testing (SLT), which precisely verifies semiconductor operation under real-world conditions.
"The lack of software support and global references, long identified as weaknesses of South Korean AI fabless companies, is beginning to be addressed this year through actual traffic handled for telecom operators and export contracts," an industry official said.
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
"For Rebellions and FuriosaAI, we need to see whether service validation and supply contracts lead to full-scale adoption and expansion. For DEEPX, the key question is whether design wins and initial orders lead to mass production of customer products and repeat orders," said Park Changyoon, a researcher at GL Research. "Whether order volumes and the sustainability of revenue grow together will be the basis for determining whether South Korean NPU companies are entering a phase of earnings growth."
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.