East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered to your inbox.
Free. Unsubscribe anytime.
3 Takeaways This Issue
- By doubling its global fab construction pace with projects like Kumamoto Fab 2 and Dresden, TSMC is locking in its dominance over the physical bottlenecks of the AI boom while shifting the execution risk of localized supply chains onto host governments.
- Japan’s Ministry of Defense is integrating AI into Self-Defense Forces command-and-control systems alongside a record ¥8.89 trillion budget request, cementing a shift toward sovereign, high-readiness military hardware over experimental software pilots.
- Through its new “Security Defense Alliance” with Anthropic, NEC is using the Claude Mythos Preview model to pitch highly secure, sovereign LLM deployments to Japan’s public sector and critical infrastructure clients who reject public-cloud AI.
This Issue’s Analysis
The Signal
TSMC’s 20-Fab Global Expansion: Why a Double-Sized Equipment Forecast Strains AI Supply Chains
TSMC Senior VP Jeffrey Hou revealed at SEMICON Taiwan 2026 that AI demand led to a near-doubling of equipment procurement forecasts within six months, forcing TSMC to simultaneousl
Semiconductors & Hardware
$200B Samsung-Broadcom Alliance: Securing HBM and 2nm Foundry Capacity Through 2030
Samsung Electronics has announced a significant expansion of its strategic collaboration with Broadcom in AI memory and foundry technologies. The partnership, valued at over $200 b
Policy & Regulation
Japan’s ¥8.5T Defense Budget: AI Command Systems and Rapid Drone Acquisition on the Front Line
Japan’s Ministry of Defense has submitted a record-high budget request of ¥8.89 trillion (approximately $57 billion USD) for fiscal year 2027, focusing on modernizing its defense c
AI & Machine Learning
Google’s Taipei Expansion: Why Shilin Is Central to custom TPU and Infrastructure Cycles
Amin Vahdat, Google’s Senior VP for AI & Infrastructure, spoke at SEMICON Taiwan 2026, outlining Google’s integrated AI infrastructure from custom chips to data centers. He announc
🧩 Pattern This Issue
- Japan: NEC deploys Anthropic Claude Mythos to secure domestic defense supply chains
- Korea/Taiwan: Google expands Shilin R&D center as TSMC doubles global fab projects
- Policy: Ministry of Defense requests record ¥8.89T for AI-driven SDF command systems
East Asian nations are rapidly integrating Western frontier AI models directly into their sovereign defense and physical infrastructure, creating a highly locked-in security alliance that permanently shuts out Chinese hardware and software alternatives from the region’s supply chain.
Also This Issue
🗾 Japan Radar
As reported in Japan — what the Japanese-language press is covering
🗾 AI & Machine Learning
NEC Finally Adopts Advanced AI ‘Claude Mythos,’ What’s Possible with Anthropic-Led ‘Security Defense Alliance’?
NEC announced its adoption of Anthropic’s advanced AI model, Claude Mythos Preview, to enhance internal IT system security. This model will be used for early detection of vulnerabilities and the automation of security measures. NEC’s participation in Anthropic’s ‘Project Glasswing‘—a ‘security defense alliance’ of over 200 companies including Google and Microsoft—grants it access to this restricted AI.
Why it matters: NEC’s deployment of Claude Mythos is more than just an internal upgrade; it’s a direct response to the increasing sophistication of cyber threats and a validation of AI’s role in defensive security. For Japan, securing its digital infrastructure with state-of-the-art AI is a strategic imperative, particularly given the nation’s push towards digitalization across industries and the ongoing geopolitical tensions that heighten cyber risks.
For Western readers: Western security firms and large enterprises should recognize that advanced AI models are quickly becoming a core component of national cybersecurity strategies, moving beyond academic benchmarks into practical deployment for vulnerability detection at scale.
🇰🇷 Korea Signal
As reported in Korea — memory, chips and platform moves from Korean sources
🇰🇷 AI & Machine Learning
Anthropic Secures 48 Trillion Won in AI Computing Backed by NVIDIA and Others
Anthropic has reportedly secured approximately 48 trillion Korean won (about $35 billion USD) worth of AI computing resources, leveraging support from key partners including NVIDIA. This massive investment aims to significantly expand Anthropic’s computational capabilities for developing advanced AI models.
Why it matters: This figure, if accurate, makes Anthropic one of the largest purchasers of AI computing resources outside of a few hyperscalers, implying significant future demand for NVIDIA’s high-end GPUs like the H100 and upcoming B200. It also signals that major AI developers are moving to lock in long-term compute capacity, which will further tighten the supply for smaller players.
For Western readers: Western AI startups and enterprises not already tied into long-term compute contracts should assume a continuing seller’s market for high-performance GPUs and cloud AI infrastructure over the next 18-24 months, with prices remaining elevated and availability constrained.
🇰🇷 Startups & Funding
Wrtn AI, Valued at 1 Trillion Won, Selects IPO Underwriters for Post-2027 Listing Target
Wrtn AI, a South Korean generative AI startup, has initiated the process of selecting lead underwriters for its initial public offering (IPO), targeting a market valuation of 1 trillion won (approximately $720 million USD). The company aims to list on the KOSDAQ market after 2027, making it one of the first domestic generative AI startups to pursue an IPO.
Why it matters: Wrtn AI’s planned IPO shows Korea’s domestic capital markets are starting to validate the country’s generative AI startups with significant valuations. This provides a direct path for local AI companies to scale and retain talent without immediately seeking funding from global investors, supporting Korea’s push for self-sufficiency in AI.
For Western readers: Western AI investors and companies should recognize that South Korea is fostering significant domestic AI development with local funding mechanisms, potentially creating strong regional competitors that may not appear on their radar through traditional Silicon Valley-centric investment channels.
🇰🇷 Policy & Regulation
Ministry of Interior and Safety to Invest 1.47 Trillion Won in ‘AI Democratic Government’ Next Year, Expanding Cloud Transition
South Korea’s Ministry of Interior and Safety (MOIS) plans to allocate 1.47 trillion won (approximately $1.1 billion USD) for its ‘AI Democratic Government‘ initiative in 2027, marking a 16.1% increase from the current year. This investment will accelerate the application of AI-based administrative services, bolster public information system stability, and expand the transition to cloud infrastructure. Key projects include an 11-fold budget increase for an AI integrated civil complaint platform and significant growth in AI administrative task application and a government-wide AI support service called ‘OnAI’.
Why it matters: The ‘AI Democratic Government’ initiative, coupled with a push for cloud adoption, creates a significant demand pipeline for AI technologies and cloud services within the Korean public sector. This isn’t just about efficiency; it’s a strategic move to build data sovereignty and potentially reduce reliance on foreign systems over time, even as it opens immediate opportunities for technology providers.
For Western readers: Western cloud providers and AI solution developers should recognize the scale of this Korean government investment as a clear signal of market opportunity for public sector contracts, but also understand that ‘domestic AI’ framing suggests an eventual preference for local solutions where feasible.
🇹🇼 Taiwan Silicon
As reported in Taiwan — foundry, hardware and enterprise IT from the Taiwanese press
🇹🇼 Policy & Regulation
NSA Releases ASIC Security Best Practices to Enhance Chip Supply Chain Defense
The U.S. National Security Agency (NSA) issued two technical reports on August 25 focused on security best practices for Application-Specific Integrated Circuits (ASICs). These reports identify 18 categories of threats to ASIC design and manufacturing processes and propose risk mitigation measures, extending the U.S. government’s supply chain security focus to custom chip levels. This builds on earlier NSA guidance for FPGAs in defense systems, now applying the same assurance framework to ASICs.
Why it matters: The NSA’s detailed threat catalog and mitigation strategies for ASICs represent a clear effort to formalize and enforce security standards throughout the custom chip supply chain. This is not just a defensive measure; it is a long-term play to dictate design and manufacturing protocols that will implicitly favor suppliers who can meet these increasingly stringent U.S. government-driven security requirements, potentially reshaping supply chain relationships for critical military and dual-use technologies. It aims to reduce the attack surface for adversaries in custom silicon.
For Western readers: Western businesses involved in ASIC design, manufacturing, or IP provision, particularly those with U.S. government contracts or seeking to enter that supply chain, should review the NSA’s LoA1 best practices immediately and prepare to integrate these security protocols into their operations and vendor management. Non-compliance could lead to exclusion from lucrative defense-related projects and, over time, potentially impact commercial designs if these standards become de facto industry norms.
🇹🇼 Enterprise & Cloud
Palo Alto Networks Acquires Console to Expand AI Agent Cybersecurity Capabilities
Palo Alto Networks has acquired Console, an AI-native IT Service Management (ITSM) platform, to integrate AI agent capabilities into its Cortex security operations platform. This acquisition aims to enhance security teams’ ability to use AI agents for investigating security signals, prioritizing tasks, and taking action in cybersecurity operations.
Why it matters: The acquisition underscores Palo Alto Networks’ aggressive strategy to embed AI deeper into its product suite, moving beyond basic threat intelligence to active AI-driven response within enterprise security operations. This shifts the focus from purely reactive human-driven security to proactive, automated, AI-agent-led security actions.
For Western readers: Western security leaders should expect a greater push from their vendors to adopt AI-agent-driven tools for security operations, which promises to shift staffing models from pure manual analysis to oversight of autonomous AI agents.
🇨🇳 China Watch
As reported in China — from Chinese-language technology media
🇨🇳 AI & Machine Learning
Fei-Fei Li’s World Labs Unveils Atlas, the First Multimodal World Model
Fei-Fei Li’s World Labs has released Atlas, described as the “global first” multimodal world model capable of 3D reconstruction from a single image, camera-controlled image/video generation, and spatiotemporal simulation. Atlas aims to provide realistic training environments for robotics by generating RGB and depth data from a few photos, which is a critical step for Real-to-Sim applications in embodied AI.
Why it matters: Li’s emphasis on simulators and the acquisition of robotics simulation company SceniX earlier this year underscores a deliberate strategy to tackle the fundamental bottleneck in robotics: the lack of scalable, affordable training data. Atlas addresses this by creating realistic synthetic environments, suggesting a future where robots learn primarily in virtual spaces, reducing the cost and time of physical prototyping.
For Western readers: Western robotics and AI firms should evaluate Atlas’s capabilities for synthetic data generation and simulated training environments, as advancements here directly impact robot deployment timelines and costs. A robust Real-to-Sim pipeline, if proven effective, could accelerate the development of embodied AI applications faster than current hardware-centric approaches.
🇨🇳 AI & Machine Learning
Ant Group Wins VLDB Industrial Best Paper for OmniTable System, Managing 35 PB of LLM Data with 5.6x Efficiency Gain
Ant Group announced its OmniTable unified wide-table system won the VLDB 2026 Industrial Best Paper award. OmniTable manages over 35 petabytes (PB) and 305 billion records of large language model (LLM) training data, streamlining the data preparation process. This system has reduced end-to-end data preparation time for a real SFT task from 14 days to 2.5 days and manual steps from 45 to 12.
Why it matters: Ant Group’s OmniTable system demonstrates a practical approach to tackling the ‘messy middle’ of AI development: efficient data curation. While Western coverage often focuses on model architecture and benchmark scores, the sheer engineering effort required to clean and manage training data at scale remains a hidden cost. Ant’s solution, recognized by a top-tier database conference, offers a blueprint for reducing this operational burden, which is critical for continuous model improvement and deployment.
For Western readers: Western enterprises developing or deploying large AI models should recognize that data preparation and management are substantial, often underestimated, cost centers; solutions like OmniTable could become a competitive advantage for those who master data pipelines efficiently rather than just focusing on raw compute and model scale.
Startups & Funding
Kuaishou’s Kling AI Unit Secures RMB1.4 Billion from China AI Fund
Kuaishou’s generative-video AI unit, Beijing Kling Technology, has attracted RMB1.4 billion (approximately $195 million USD) from the state-backed China Artificial Intelligence Industry Investment Fund for a 1.14% stake. Thai conglomerate Charoen Pokphand Robot also invested RMB131.45 million ($19.29 million USD) for 0.11% of the company. These investments follow Kling’s spinout from Kuaishou, providing it with additional capital and strategic shareholders.
Why it matters: The China Artificial Intelligence Industry Investment Fund’s participation is a clear signal that Kuaishou’s generative video capabilities are seen as strategically important for national AI development. This isn’t just a funding round; it’s a state endorsement that aligns Kling’s growth with broader industrial policy goals, likely including data sovereignty and domestic content control.
For Western readers: Western investors and companies should recognize that the Chinese state is actively shaping its AI landscape, favoring domestic champions like Kling with capital and strategic backing, which will create a more formidable, state-supported competitive environment for any foreign AI services or platforms seeking to operate in China.
🔺 The Prism
Where US and East Asian technology interests intersect
AI & Machine Learning
UI-Venus-2 Technical Report
Chinese researchers from the Venus Team have released a technical report on UI-Venus-2, a multimodal GUI agent designed for cross-environment digital task automation. The agent covers over 170 multilingual mobile apps and native desktop operating systems, employing advanced task generation and verification methods to enhance reliability and practicality.
Why it matters: The development of a robust, general-purpose GUI agent like UI-Venus-2 demonstrates China’s ongoing push to build foundational AI capabilities that can operate across various digital environments. This is not just an academic exercise; it’s about establishing practical tools for automation and control over the digital interface layer, which is a critical piece of the ‘domestic AI’ puzzle for Beijing.
For Western readers: Western businesses operating in East Asia, particularly those relying on local digital ecosystems, should expect increasingly sophisticated AI automation tools to emerge from Chinese developers, potentially altering local competitive dynamics in enterprise software and consumer applications.
Cross-Regional Analysis
Sir Robin Saxby Reflects on AI’s Impact and Geopolitics, Implications for East Asia
ARM founder Sir Robin Saxby highlighted the importance of balancing ‘local champions‘ with global collaboration in the semiconductor industry, a critical consideration for East Asian nations seeking technological self-sufficiency. He noted that despite the push for regionalization, the supply chain remains deeply intertwined, necessitating continued international partnerships. Saxby emphasized that AI’s development is inherently global and cannot be constrained by national borders, an observation with significant implications for China’s AI ambitions and Taiwan’s chip dominance.
Why it matters: Saxby’s observation that AI’s development transcends national borders directly challenges the ‘data sovereignty’ and ‘domestic AI’ narratives pushed by countries like Japan and China. While those narratives are often industrial policy in disguise, Saxby points out that the fundamental nature of AI as a global endeavor will continue to force reliance on international collaboration, particularly for hardware and foundational research.
For Western readers: Western businesses should understand that despite government-backed efforts in East Asia to create self-sufficient ‘local champions,’ the underlying global dependencies in semiconductors and AI development remain strong. Do not over-interpret nationalistic rhetoric as an immediate severing of supply chain ties; watch for concrete policy changes and investment in dual-use technologies instead.
Semiconductors & Hardware
Nvidia Supplier Unimicron Probed for Relabeling China-Made Parts
Taiwanese prosecutors are investigating Unimicron, a major supplier of chip substrates and PCBs to Nvidia and Intel, over allegations it falsely relabeled parts made in China as originating from Taiwan. This potentially violates laws designed to prevent origin laundering and circumvent trade restrictions.
Why it matters: This investigation exposes the continued pressure on Taiwanese manufacturers to manage their China operations and supply chains under increasing scrutiny, especially when serving critical Western technology clients. While Unimicron is a key player, the larger implication is the potential for disruptions to high-end AI component supplies if major Taiwanese suppliers are found to be actively obscuring origins.
For Western readers: If you are a Western AI hardware company relying on advanced PCB or substrate suppliers from Taiwan, you should immediately review your supply chain integrity policies and verify the true origin of your components, especially those with high demand. Non-compliance could lead to severe disruptions or legal issues.