South Korea launches first fully sovereign AI service integrating domestic tech
Translated from Korean, summarized and contextualized by DistantNews.
At a glance
- Three South Korean tech companies have partnered to launch the country's first fully sovereign AI service, integrating domestic AI models, chips, and platforms.
- The new
South Korea's tech landscape is seeing a significant leap forward with the launch of "AI Summary" on the Daum portal, a service powered entirely by domestic technology. This collaboration between Upstage, FuriosaAI, and AXZ (operator of Daum) marks the nation's first commercial application of a "full-stack sovereign AI" system, encompassing AI chips, models, and services developed within the country.
The "AI Summary" feature allows users to input queries into the Daum search engine, which then retrieves relevant documents and uses an AI model to condense their core content. This eliminates the need for users to manually open and read through multiple links.
Small yet high-performing, and it's particularly good at summarizing Korean text.
Upstage's large language model, 'Solar,' is at the heart of this service, with FuriosaAI's neural processing unit, 'Renegade,' handling the model's computations. "Solar is small yet high-performing, and it's particularly good at summarizing Korean text," stated Kim Sung-hoon, CEO of Upstage. This domestic integration is a strategic move for Upstage, which acquired AXZ in May with the goal of transforming Daum into an AI-centric portal.
Representatives from all three companies emphasized the competitive edge of their homegrown technologies against global counterparts. Baek Jun-ho, CEO of FuriosaAI, highlighted that their domestic NPU can "drastically reduce token costs" and offers superior power and cost efficiency compared to NVIDIA's GPUs. FuriosaAI claims Renegade matches the performance of NVIDIA's H200 while reducing inference costs by over 50%. AXZ plans to expand the AI Summary service, currently exposed to about 20% of search queries, to over half, while also developing specialized services for various sectors like shopping and dining, and an "AI agent (secretary) for one person" to proactively provide personalized information.
Domestic neural processing units can drastically reduce token costs, and offer superior power and cost efficiency compared to NVIDIA's GPUs.
Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.