East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered to your inbox.
Free. Unsubscribe anytime.
3 Takeaways This Issue
- CXMT’s transition to mass production of its 5th-generation DRAM platform places China’s leading memory maker within two years of global leaders Samsung and SK Hynix, severely weakening the long-term efficacy of Western semiconductor export controls.
- A survey of 143 AI project managers in South Korea revealed that 97.2% of enterprises are experiencing persistent model hallucinations and errors, exposing a critical gap between high-level AI adoption metrics and actual operational readiness in East Asia’s tech hub.
- While 40% of Japanese workers express a desire for AI upskilling, a survey by NEXER and LISKILLING shows that 90.7% have received no corporate training, creating a massive bottleneck for Japan’s domestic digitization push.
This Issue’s Analysis
The Signal
CXMT’s 5th-Gen DRAM Ramp: How China’s Yield Breakout Threatens the Global Memory Market
ChangXin Memory Technologies (CXMT) has reportedly started mass production of its 5th-generation DRAM platform, which it claims is approaching global top-tier standards. This devel
Semiconductors & Hardware
TSMC’s Production Moat: Why Scale Beats Tech Dominance for the AI Foundry Market
TSMC has established an impenetrable “steel moat” through its massive production capacity and technological barriers, dominating the advanced chip foundry market amid exploding glo
AI & Machine Learning
US-China AI Model Cycles Shrink to 44 Days as Autonomous R&D Accelerates
AI development in the US and China is accelerating at an unprecedented pace, with new models now being released every 44 days on average, a third of the previous development time.
Semiconductors & Hardware
KOSPI Rebounds Above 7,000 as Samsung and SK hynix Rallies Drive Korea’s Market Recovery
The KOSPI index closed above 7000 points for the first time in ten days, driven by a global rally in semiconductor stocks. Institutional investors were the primary buyers, while in
🧩 Pattern This Issue
- Taiwan/Korea: TSMC and CXMT expand production capacity to secure hardware dominance
- Korea: LG Electronics targets data center infrastructure with high-capacity turbo chillers
- Japan: Acceleration of US-China AI models intensifies hardware supply chain demands
While Western firms focus on software model benchmarks, East Asian giants are consolidating control over the physical layer of AI—from TSMC’s manufacturing moat to LG’s cooling systems—leaving the West increasingly dependent on Asian hardware to run its software revolution.
Also This Issue
🗾 Japan Radar
As reported in Japan — what the Japanese-language press is covering
🗾 AI & Machine Learning
AI-Powered Platform Launched to Support Elderly Without Relatives in Japan
Tsunagari AI, J-Lease, and Fine have launched the “Elderly Care AI Platform Business for Those Without Relatives” in Oita Prefecture. This initiative uses an AI chatbot, accessible via LINE, to provide daily companionship and proactively identify issues related to housing, health, isolation, and end-of-life planning for elderly individuals lacking family support. When human intervention is required, the AI connects them with a consortium of local medical, welfare, legal, financial, and administrative professionals.
Why it matters: This initiative matters because it leverages AI to address a critical social infrastructure gap in Japan, where the default assumption of family support for the elderly is increasingly untenable. By shifting from a reactive ‘wait-for-a-problem’ approach to proactive, AI-driven daily monitoring and early intervention, it aims to prevent crises like unmanaged finances or unattended deaths, which are becoming more common as the population ages and family structures change. This isn’t just about efficiency; it’s about re-engineering the social safety net.
For Western readers: Western businesses in elder care technology or social services should recognize that Japan is pioneering AI-driven, community-wide solutions to demographic challenges that will eventually face many developed nations. Watch for this platform’s scalability and adoption outside Oita as an indicator of effective models for integrating AI into social welfare infrastructure, especially for chronic issues where human resources are scarce.
🗾 Workforce & Culture
Less Than 10% of Japanese Workers Received AI Training, Despite Over 40% Wanting It
A survey by NEXER Group and LISKILLING found that 90.7% of Japanese workers have not received AI or DX (Digital Transformation) training, while 44.4% expressed a desire to acquire AI and digital skills. The survey also revealed that AI and DX are rarely discussed in the workplace, with 73.0% stating these topics are ‘never’ or ‘hardly ever’ discussed.
Why it matters: The survey results confirm what many have observed: while Japan’s Ministry of Economy, Trade and Industry (METI) pushes for DX and AI adoption to address labor shortages and boost productivity, the initiatives often fail to translate into tangible training or workplace integration at the company level. The desire for ‘practical, immediately usable’ skills suggests that current training offerings, or the lack thereof, are not meeting employees’ needs for immediate job relevance.
For Western readers: Western businesses operating in Japan, or those partnering with Japanese firms, should recognize that many Japanese employees lack fundamental AI/DX skills despite a willingness to learn, meaning partners may require more hands-on support in implementing new technologies. Do not assume your Japanese counterparts have internal AI training programs that will pick up the slack.
🇰🇷 Korea Signal
As reported in Korea — memory, chips and platform moves from Korean sources
🇰🇷 AI & Machine Learning
Korean AI Managers Report Widespread Issues with AI Hallucination and Errors
A survey of 143 AI managers in South Korea found that 97.2% experienced problems with AI providing false or distorted information. Specifically, 62.2% reported hallucination, with experts noting these errors are becoming harder to detect. As a result, 76.2% of managers trust expert opinions over AI outputs when discrepancies arise.
Why it matters: While hallucination is a known challenge in generative AI globally, this specific survey provides current, quantitative data from a significant East Asian market on how pervasive the problem is for managers. The finding that errors are becoming more sophisticated and harder to detect suggests that current mitigation strategies are insufficient, driving up operational costs as human review becomes essential.
For Western readers: Western businesses deploying AI solutions in East Asia, or relying on AI for analysis of East Asian markets, should assume a significant local skepticism about AI output and factor in the necessity of human verification, which adds cost and complexity. This also presents an opportunity for AI vendors who can credibly demonstrate superior accuracy or robust verification tools.
🇰🇷 Semiconductors & Hardware
LG Electronics Launches High-Capacity Turbo Chiller for AI Data Centers
LG Electronics has introduced a new line of high-capacity turbo chillers designed for cooling AI data centers. The new chillers aim to meet the increasing demand for efficient cooling solutions driven by the high heat generation of AI servers and high-bandwidth memory (HBM). This move positions LG to capitalize on the rapidly growing infrastructure needs of the AI industry.
Why it matters: The competition for data center cooling is heating up as companies face the thermal challenges of high-density AI racks. LG, primarily known for consumer electronics and appliances, is making a strategic play to leverage its HVAC expertise for industrial applications, aiming for a segment with higher margins and growth potential than its traditional markets.
For Western readers: Western data center operators and AI infrastructure providers should expect increasing competitive pressure and innovation in cooling solutions from East Asian manufacturers, potentially leading to more specialized and efficient options for managing extreme heat loads.
🇹🇼 Taiwan Silicon
As reported in Taiwan — foundry, hardware and enterprise IT from the Taiwanese press
🇹🇼 Policy & Regulation
South Korea Reforms ISMS-P Certification to Strengthen On-site Inspections Amidst Frequent Data Breaches
South Korea’s Personal Information Protection Commission (PIPC) announced reforms to its Information Security and Personal Information Protection Management System (ISMS-P) certification. This move follows a new Personal Information Protection Act, which can fine companies up to 10% of annual revenue for breaches affecting over 10 million people. The reforms aim to enhance the effectiveness of certifications, shifting from ‘snapshot’ audits to continuous operational reviews with strengthened on-site verification.
Why it matters: The reform of ISMS-P is a direct response to a series of high-profile data breaches at companies that already held certifications, which had eroded public trust in the system’s effectiveness. This move indicates a shift from a compliance-checkbox mentality to a more rigorous, continuous verification of actual security practices, which is a necessary step to bring the certification in line with its stated purpose.
For Western readers: Western companies with operations or data handling in South Korea must recognize that their existing ISMS-P certifications may no longer be sufficient; anticipate increased scrutiny and a need to demonstrate continuous, not just point-in-time, compliance with data protection standards.
🇹🇼 Policy & Regulation
North Korean Hacker Group WaterPlum Infects 30,000 Devices, Steals $10 Million in Crypto via Fake Job Offers
A joint alert from the US, Japan, Australia, and Germany revealed that the North Korean hacking group WaterPlum (aka Contagious Interview) infected over 30,000 devices in more than 100 countries between December 2025 and July 2026. The group targeted web designers, engineers, and crypto/Web3 professionals using fake job offers to distribute malware, subsequently stealing at least $10.71 million in cryptocurrency.
Why it matters: The detailed tracking and attribution of WaterPlum to North Korea’s Munitions Industry Department’s General Bureau 313 by the FBI and Japanese police confirms that state-sponsored cybercrime is a core component of Pyongyang’s economic strategy. The focus on stealing from individuals and using their compromised devices as stepping stones into corporate networks suggests a dual goal of direct financial gain and corporate espionage, underscoring the deep integration of these illicit activities into the regime’s operational toolkit.
For Western readers: Western companies and individuals should recognize that North Korean state-sponsored hacking groups like WaterPlum are actively targeting remote workers and individuals in the crypto space with sophisticated social engineering, so tighten HR and security protocols around remote hiring and crypto asset management to prevent becoming a vector for state-sponsored theft.
🇨🇳 China Watch
As reported in China — from Chinese-language technology media
🇨🇳 Semiconductors & Hardware
Xiaomi 18 Pro Series to Launch September 23rd with Price Increase; ChangXin Memory G5 DRAM in Mass Production
Xiaomi announced its 18 Pro series phones will launch on September 23rd, featuring a 2nm mobile platform and Leica dual 200-megapixel cameras, with President Lu Weibing confirming a price increase. Separately, ChangXin Memory (CXMT) has begun mass production of its G5 DRAM platform, reportedly boosting die per wafer by at least 50%.
Why it matters: Xiaomi’s move to raise prices for its flagship smartphones indicates a shift towards higher-margin products and brand premiumization, following similar strategies by other Chinese manufacturers. For CXMT, the G5 DRAM mass production suggests material progress in closing the technology gap with international memory leaders, critical for Beijing’s long-term semiconductor goals.
For Western readers: Western businesses tracking the smartphone market should note that Chinese brands like Xiaomi are pushing higher price points, indicating a maturing market and potentially shifting consumer expectations. Semiconductor firms should recognize CXMT’s G5 DRAM as a tangible sign of China’s sustained effort to build domestic memory capacity, which will eventually impact global supply and pricing dynamics.
Robotics & Automation
Unitree Releases Partial Weights for UnifoLM-WLA-1.0 After 64-Task Reveal
Chinese robotics company Unitree has released partial weights for its UnifoLM-WLA-1.0 model, a large language model designed for quadruped and humanoid robots. This release follows a demonstration where the model successfully completed 64 different robot control tasks across various scenarios.
Why it matters: Unitree’s move to open-source partial weights for a robotics-specific LLM aims to spur broader adoption and development within China’s robotics ecosystem. This is a common tactic to build a platform advantage, ensuring that future applications and research build upon their foundational work, rather than waiting for fully closed, proprietary solutions from abroad.
For Western readers: Western robotics firms and researchers should recognize that China is building a robust, domestically-driven robotics AI ecosystem, potentially leading to faster iteration and localized innovation that could become competitive in specific application areas, particularly where data sovereignty and local standards are critical.
🇨🇳 AI & Machine Learning
Hefei-based Anhui AI Safety Lab Releases Three AI Security Solutions
The Yangtze River Delta Security Artificial Intelligence Anhui Provincial Lab unveiled three AI security solutions — ‘Star Realm,’ ‘Star Steward,’ and ‘Star Witness’ — at the first China Cyberspace Security Conference (CCSC 2026) in Hefei. These solutions address large model content safety, intelligent agent application security, and verifiable propagation of AIGC-generated content. The lab’s director, Liu Junhua, highlighted the critical need for AI safety governance as AI adoption accelerates across industries.
Why it matters: China frames AI safety primarily as a matter of content control and managing risks from domestic large models and applications. This release reflects the government’s dual imperative: foster rapid AI development while ensuring content aligns with state values and preventing misuse, particularly for politically sensitive or misinformation-prone generative AI. The active involvement of iFLYTEK, a prominent Chinese AI company, suggests a strong public-private partnership in building these ‘national champion’ security capabilities.
For Western readers: Western companies operating or partnering in China’s AI ecosystem must understand that ‘AI safety’ in this context heavily emphasizes content moderation, data provenance, and the prevention of ‘harmful’ or ‘illegal’ content as defined by Chinese regulations, which differs from purely technical security concerns. Assume that any AI solution deployed in China will be subject to these domestic governance frameworks, potentially requiring integration with these types of tools.
🔺 The Prism
Where US and East Asian technology interests intersect
AI & Machine Learning2 STORIES
Securing and Enhancing the Next Generation of Autonomous AI Agents
Recent research from East Asian institutions introduces pivotal advancements in both the performance and security of autonomous AI agents. While the MOSCOPT algorithm optimizes multi-skill learning in LLM agents without parameter tuning, new security protocols establish a reliable method for revoking authorizations in complex, long-running agent workflows.
Why it matters: In East Asia’s rapidly automating industrial and service sectors, these innovations provide the dual assurance of high-performance multi-tasking and robust permission-revocation needed to safely integrate AI agents into critical business pipelines.
For Western readers: Discard the assumption that agent safety and agent capability are separate engineering tracks; Western developers must design security protocols like root-scoped revocation in lockstep with dynamic multi-skill optimization from day one.
Semiconductors & Hardware
MediaTek Moves Dimensity Flagship to 2nm with New AI Architecture
Taiwanese chip designer MediaTek has launched its Dimensity 9600 Pro flagship smartphone SoC, the company’s first chip built on TSMC’s 2nm process. The new processor features an all-big-core CPU, a dual-NPU architecture with an ‘Agentic AI Engine,’ and supports on-device AI models up to 30 billion parameters, while claiming significant performance gains and power efficiency improvements.
Why it matters: MediaTek’s move to TSMC’s 2nm process for its flagship Dimensity SoC is a direct challenge to Qualcomm’s Snapdragon series and further entrenches TSMC as the critical foundry partner for leading mobile chip designers. The emphasis on ‘Agentic AI’ and 30-billion-parameter on-device models shows MediaTek is serious about AI performance, which will be a key differentiator in the coming mobile generation.
For Western readers: Western smartphone OEMs should expect MediaTek-powered devices to offer comparable, if not superior, on-device AI capabilities and power efficiency to Qualcomm-equipped models, potentially increasing competition for premium Android device components.