Firms Ditch Costly AI, Switch to Cheaper Chinese Models

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

1. Sparing Fable5 and Using Chinese AI for Easy Tasks

- Key Summary: According to market research firm Silicon Data, the token spending index that tallies large language model (LLM) token prices fell 22% in just over a month to $1.64 per million tokens. The main backdrop is that companies, feeling the cost burden of AI transformation (AX), are moving from high-priced frontier models such as those from Anthropic and OpenAI to low-cost Chinese models such as DeepSeek and Zhipu AI. In fact, Chinese models' token share on OpenRouter reached 48%, ahead of U.S. models (20%), and Nexon temporarily suspended its use of Anthropic's 'Claude Fable5.' As a result, in the telecommunications industry, services that optimize companies' token spending are drawing attention as a new business area.

2. Hyundai Motor's U.S. Subsidiary Tests Tether Payments, Completes Remittance in 7 Minutes

- Key Summary: Hyundai Card announced on the 9th that it completed an experiment using the dollar stablecoin Tether (USDT) for an international remittance between Hyundai Motor's U.S. subsidiary (HMA) and Mexico subsidiary (HMM). The entire process, in which the U.S. subsidiary converted $20,000 into USDT, sent it to the Mexico subsidiary, and converted it back into dollars, took about seven minutes, drastically reducing processing time compared to bank transfers that typically take more than three to four hours. Tether, the blockchain platform Avalanche, and payment infrastructure company Axim participated in this experiment, and Hyundai Card plans to launch a second demonstration based on local currency targeting its European subsidiary from the end of this month. In the industry, assessments are emerging that Korean companies are taking the lead in using dollar stablecoins while the introduction of a won-based stablecoin is delayed.

3. Beyond Secondhand Trading to Local Communities: 10 Million Visit Daangn Meetup

- Key Summary: Daangn, the local life community operator, announced on the 7th that monthly active users (MAU) of its 'Daangn Meetup' surpassed 10 million in the first half of this year. The number of meetups rose 74% and members increased 128% year-on-year, while meetup schedule creation grew 106% and participation reviews rose 191%, showing that the entire process from opening meetups to sharing reviews has become active. By age group, users in their 40s were the most numerous at 27%, followed by those in their 30s at 26% and their 50s at 23%, while total time spent on the community tab also rose 103% from a year earlier. The number of communications between users, including posts, comments, and likes, reached 83.17 million, up 98% year-on-year.

4. Hyundai Motor Securities Accelerates AX, Becomes First Small-to-Mid-Sized Firm to Adopt Cloud AI Agent

- Key Summary: Hyundai Motor Securities officially distributed its cloud-based AI work support service 'HAI 1.0' to all employees on the 9th, becoming the first among small-to-mid-sized brokerages to do so. HAI 1.0 answers employee questions and summarizes documents based on internal regulations and work documents, and was developed in a cloud format using Anthropic's 'Claude Sonnet 4.5' as its base LLM. Hyundai Motor Securities plans to build a company-wide common AI platform 'HAI 2.0' within the year to develop customized AI agents for each employee and pension-specialized agents, and in 2027 plans to expand the scope of AI application to customer-facing services with 'HAI 3.0.' Bae Hyung-geun, CEO of Hyundai Motor Securities, presented AI innovation as a core management strategy at the regular shareholders' meeting in March this year.

5. Physical AI Brain Competition With Nvidia and Google: A Quantum Jump in Autonomous Manufacturing Capability

- Key Summary: LG Group has joined the physical AI competition with Nvidia, Google, and others by cooperating with PhysicsX to develop a 'world model' that serves as the brain of robots and machines. The world model is a highly difficult technology being developed by only a few companies, including Nvidia's 'Cosmos,' Google's 'Genie,' and 'JEPA' from AMI Labs, founded by New York University professor Yann LeCun. LG is taking a two-track strategy that runs it in parallel with the 'action model' it has previously focused on. Affiliates including LG Electronics (066570.KS), LG AI Research, and LG CNS are jointly developing an industrial robot foundation model (RFM) and conducting demonstrations (PoC), while LG Electronics is also building a robot data factory at its R&D campus in Yangjae, Seoul. Meanwhile, LG plans to invest 9.4 trillion won in the robot and physical AI fields in the Yeongnam region by 2030 through a mega project.

6. China's CXMT Kicks Off 6.5 Trillion Won Fundraising, Draws D-RAM Love Calls Even From Apple

- Key Summary: China's largest D-RAM maker CXMT (Changxin Memory Technologies) has submitted a prospectus to the Shanghai Stock Exchange and will conduct a public offering subscription starting on the 16th, launching an initial public offering (IPO) worth 6.53 trillion won (29.5 billion yuan). The raised funds will be invested in production capacity expansion, D-RAM technology development, and next-generation research and development, and CXMT, buoyed by the AI boom, raised its global D-RAM market share from around 3% last year to around 8% this year. Apple is reportedly testing CXMT products in devices for sale in China and lobbying the U.S. government to request approval for their use, though it is said there is a technology gap of about one to two years with Samsung Electronics (005930.KS) and SK hynix (000660.KS) in the high-bandwidth memory (HBM) field. China's memory chip self-sufficiency rate rose to 25% this year and is projected to reach 30% by 2028.