East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered weekly.
Weekly. Free. Unsubscribe anytime.
3 Takeaways This Week
- CXMT’s 465% stock surge on its Shanghai STAR Market debut pushed the Chinese memory maker’s market cap past Intel, demonstrating how state-backed capital inside China is aggressively revaluing domestic semiconductor champions as Western sanctions squeeze the supply chain.
- By shipping five domestically produced immersion DUV lithography machines this year through Shanghai Aishengna Technology, China is establishing a parallel semiconductor equipment pipeline that reduces its long-term reliance on Dutch giant ASML.
- Japanese university startups like Telexistence and FingerVision are deploying AI-driven tactile sensors on robot arms to automate delicate tasks in convenience stores and food factories, leveraging Japan’s physical engineering expertise to solve its acute labor shortage.
Core Move
Shanghai Yuliansheng to Ship Five Immersion DUV Machines This Year #
Shanghai Aishengna Technology plans to ship five domestic immersion DUV lithography machines this year. This move shows that China is speeding up its timeline to make its own chips. It is moving past talk and into real manufacturing. US-led export controls aimed to stop China at the 7nm node. Yet, this progress shows how Chinese foundries can keep and grow their 28nm production. These firms include SMIC, Hua Hong, and CXMT. They can use multi-patterning techniques to push their chips to 7nm or 5nm.
This move shows Beijing’s deep commitment to industrial policy. It focuses on areas that affect national security. Shanghai Yuliansheng Technology is backed by the state, and its team formed Aishengna. This setup shows a clear pattern. When China faces pressure from abroad, it puts national resources behind a single champion effort. Chinese state media, including the Global Times, calls these wins a sign of national strength. They frame them as a direct reply to foreign containment, not just a business upgrade.
Western observers often focus too much on complex tools like ASML’s High-NA EUV. They miss the industrial value of strong immersion DUV. This approach is like Japan’s post-war industrial plan. In that plan, mastering high-volume work came before making major breakthroughs. The current goal is not to beat TSMC’s best nodes. Instead, China wants to detach its key supply chains from foreign control at the most profitable nodes.
A major risk lies in the yield and throughput of these machines. Shipping five units is a vital step, but getting good volume and consistency is harder. History shows that even top firms struggle to scale up lithography tools. Simply owning machines does not guarantee low costs or good performance.
To see if this works, we must see if Aishengna can ship 20 units next year. We should also watch capacity use and growth plans at SMIC and Hua Hong for 28nm and 14nm chips over the next 12 to 18 months. Lastly, we must track what Chinese foundries say about the performance and total cost of these tools compared to older foreign gear.
🗾 Japan Radar #
What Japanese media is reporting that Western outlets miss
Japan anchors its AI strategy in physical robotics while quietly adopting pragmatically deployed Chinese and American frontier software.
🗾 AI & Machine Learning
Chinese AI company Moonshot AI released the model weights and a technical report for its latest model, Kimi K3, on July 27. Japanese software development firm Fixstars reported successful deployment and operation of Kimi K3 on a single node with eight NVIDIA B300 GPUs, noting an 88.8-minute startup time, with model taking 81.4 minutes.
Why it matters: The rapid deployment and performance testing of Kimi K3 in Japan, immediately following its release, highlights how quickly advanced Chinese AI models are being adopted and integrated into global tech infrastructure. While Western regulators debate the risks, technical teams are already putting these models to work.
For Western readers: Western businesses and policymakers should recognize that Chinese open-weight models are now immediately available globally and directly competitive with leading Western models, making attempts to contain their spread through policy alone increasingly difficult. 🗾 AI & Machine Learning
Google’s Gemini 3.6 Flash, released on July 21st, initially faced user complaints on X (formerly Twitter) regarding its accuracy, including basic numerical errors and fabricating details about non-existent subjects. One week later, ITmedia’s verification found that some simple errors, like numerical comparisons, were no longer reproducible, but reports of hallucinations and coding mistakes persist.
Why it matters: The speed at which Google appears to be patching critical flaws in Gemini 3.6 Flash, even if anecdotal, points to the pressure to deploy and iterate quickly in the AI race. The ‘Flash’ models are meant to be fast and cost-effective, so accuracy issues undermine their core value proposition for enterprise users. The continued reports of hallucination, especially in coding, indicate that fundamental reliability remains a challenge even with quick fixes.
For Western readers: Western businesses evaluating Gemini 3.6 Flash for production use should assume that initial accuracy reports from July will quickly become outdated. However, they should also recognize that persistent issues like coding hallucinations mean models in this class still require significant human oversight for mission-critical applications. The ‘Flash’ model, while faster, isn’t yet a reliable hands-off tool. Policy & Regulation
China Accuses U.S. of ‘AI Hegemonism,’ Threatens Countermeasures Over Potential Probes China has accused the United States of ‘AI hegemonism‘ and threatening Chinese companies with sanctions for allegedly copying advanced U.S. AI models via ‘** distillation**,’ claims China disputes as lacking factual or legal basis. This comes as the U.S. reportedly considers new probes into Chinese AI development. The dispute underscores the deepening technological rivalry between the two nations.
Why it matters: Beijing views the U.S. pursuit of ‘distillation’ allegations as a pretext for further restricting its AI industry, much like earlier moves on semiconductor exports. This isn’t just about IP; it’s about control over foundational technologies. Expect China to frame any countermeasures not as retaliation, but as necessary steps to ensure its own digital sovereignty.
For Western readers: Western businesses operating in China’s AI sector or relying on its supply chain should prepare for increased regulatory scrutiny from both sides, as China’s ‘countermeasures’ will likely target U.S. companies or their allies operating within its borders. Robotics & Automation
Japanese University Startups Develop AI Robot Arms for Delicate Tasks
Japanese university-backed startups are advancing physical AI technologies to enable robot arms to handle fragile objects with human-like dexterity, targeting applications from food preparation to logistics. Companies like Toyota Motor are showing interest, recognizing the potential for labor savings in an aging society.
Why it matters: Japan’s strength in precision engineering and factory automation is converging with AI research to create genuinely capable physical AI systems. This isn’t just an announcement; the engagement from companies like Toyota suggests real-world deployment is closer than some might assume from typical startup news.
For Western readers: Western robotics and automation firms should recognize that Japan is not solely playing defense in AI-driven industrial tech; these developments in delicate handling represent a specific area where Japanese innovation is leading, not just catching up. Semiconductors & Hardware
China’s CXMT Jumps 465% on Debut, Tops Intel’s Market Cap Chinese memory chip maker CXMT debuted on the Shanghai STAR Market with a 465% stock jump, raising 57.9 billion yuan ($8.6 billion) in Asia’s largest IPO this year. The state-backed DRAM supplier‘s market capitalization surpassed Intel’s, buoyed by optimism around the AI boom. The speed and scale of CXMT’s IPO, despite earlier valuation concerns, indicate a strong domestic appetite for Chinese-made semiconductors, reflecting a deeper trend than just an ‘AI boom.’ It’s a statement about where capital is flowing and what kinds of companies are being prioritized for investment inside China.
Why it matters: The speed and scale of CXMT’s IPO, despite earlier valuation concerns, indicate a strong domestic appetite for Chinese-made semiconductors, reflecting a deeper trend than just an ‘AI boom.’ It’s a statement about where capital is flowing and what kinds of companies are being prioritized for investment inside China.
For Western readers: Western semiconductor investors should adjust their models to account for China’s expanding capital markets and domestic funding mechanisms, which can propel local champions to significant valuations even without direct access to Western markets.
🇨🇳 China Watch #
China’s technology moves, framed for Western readers
China is aggressively exporting its hardware manufacturing dominance into robotics while localizing AI for enterprise and social control.
Robotics & Automation
China’s Cleaning Robots Dominate Global Market with 70% Share, Driven by Innovation, Not Just Price Chinese cleaning robot manufacturers, led by Roborock, Ecovacs, and Dreame, now command 70% of the global market share. This dominance is attributed to technological advancements like stair-climbing and robotic arm integration, challenging the perception that their success relies solely on low prices. The narrative in China around these companies emphasizes their engineering prowess and unique features, directly refuting the common Western assumption that their market share gains are purely a function of cost arbitrage. This focus on innovation signals a mature approach to global competition, rather than just volume play.
Why it matters: The narrative in China around these companies emphasizes their engineering prowess and unique features, directly refuting the common Western assumption that their market share gains are purely a function of cost arbitrage. This focus on innovation signals a mature approach to global competition, rather than just volume play.
For Western readers: Western appliance and robotics companies should recognize that Chinese competitors are now setting feature and innovation benchmarks, not just undercutting on price, meaning competitive strategy must shift from cost-cutting to accelerated R&D. Robotics & Automation
Chinese Humanoid Robot Maker AgiBot Initiates Hong Kong IPO Process Chinese humanoid robot developer AgiBot, founded in Shanghai in 2023, has begun the process for a Hong Kong IPO. The company specializes in humanoid robots and embodied AI systems for industrial and commercial use, and has also developed the AgiBot World dataset for embodied AI training. China’s strategy here isn’t just about AI models; it’s about integrating AI with physical hardware to gain control over manufacturing and logistics automation. An IPO for a company like AgiBot indicates that private capital, alongside state initiatives, is flowing into creating the physical infrastructure for AI deployment, not just the software behind it. This is a practical step towards autonomous factories and distribution systems.
Why it matters: China’s strategy here isn’t just about AI models; it’s about integrating AI with physical hardware to gain control over manufacturing and logistics automation. An IPO for a company like AgiBot indicates that private capital, alongside state initiatives, is flowing into creating the physical infrastructure for AI deployment, not just the software behind it. This is a practical step towards autonomous factories and distribution systems.
For Western readers: Western robotics firms should anticipate increased competition from well-funded Chinese players in industrial automation and logistics, especially in regions receptive to China’s ‘Belt and Road’ initiatives. AI & Machine Learning
Moonshot AI to make Kimi K3 model weights available for public download Chinese AI firm Moonshot AI is releasing the model weights for its Kimi K3 large language model, making it available for developers to download, modify, and host independently. The Beijing-based company aims to position Kimi K3, which features 2.8 trillion parameters and a one-million-token context window, as an open-weight model for advanced coding and reasoning tasks. Releasing model weights provides a concrete foundation for local developers to build upon, which is more impactful than an API-only approach for fostering a domestic AI ecosystem. This signals Moonshot AI’s intent to become a foundational layer in China’s AI infrastructure, similar to how early open-source models enabled rapid development in the West.
Why it matters: Releasing model weights provides a concrete foundation for local developers to build upon, which is more impactful than an API-only approach for fostering a domestic AI ecosystem. This signals Moonshot AI’s intent to become a foundational layer in China’s AI infrastructure, similar to how early open-source models enabled rapid development in the West.
For Western readers: Western developers and businesses should assess Kimi K3’s performance in long-context coding and reasoning tasks; if it proves competitive, it provides a strong, Chinese-controlled alternative that could accelerate development cycles within China’s tech sphere, potentially challenging reliance on Western models for sensitive applications. AI & Machine Learning
SenseTime Partners with Bank of Ningbo for AI-Generated Digital Human Live Show
Chinese AI firm SenseTime collaborated with Bank of Ningbo to launch an AI-generated content live show featuring a digital human. This initiative demonstrates the application of AI-powered digital humans in financial services for customer engagement and content creation.
Why it matters: This move shows how Chinese financial institutions are quickly adopting AI to streamline customer interaction and content generation, moving beyond traditional human-led models. It’s a clear signal that companies here are comfortable deploying these tools in public-facing roles now.
For Western readers: Western financial institutions should note the speed at which Chinese banks are implementing digital human technology for customer engagement, as this could set a benchmark for efficiency and scalability in AI-driven services globally. AI & Machine Learning
Chinese Academy of Sciences Unveils ZhiJing Social Intelligence Model for Advanced Social Context Understanding The Chinese Academy of Sciences (CAS) has developed the ZhiJing Social Intelligence Model, an AI system designed to understand complex social contexts, emotions, and unspoken signals. This initiative aims to address a critical weakness in current AI: the lack of nuanced social comprehension, which is particularly relevant for applications interacting with people in culturally specific ways. While Western AI development often focuses on raw computational power and benchmark scores for general intelligence, China’s emphasis on “social intelligence” reflects an understanding that AI’s utility in real-world applications, especially in densely populated, interaction-heavy environments, relies on cultural and emotional nuance. This is not about bigger models, but smarter interaction tailored to specific environments.
Why it matters: While Western AI development often focuses on raw computational power and benchmark scores for general intelligence, China’s emphasis on “social intelligence” reflects an understanding that AI’s utility in real-world applications, especially in densely populated, interaction-heavy environments, relies on cultural and emotional nuance. This is not about bigger models, but smarter interaction tailored to specific environments.
For Western readers: Western AI developers should recognize that “social intelligence” in Chinese AI research may prioritize different interaction patterns and contextual cues than those derived from Western cultural datasets, influencing user acceptance and functionality in diverse global markets.
🔺 The Triangle #
Where US, Japan, and China technology interests intersect
US-China tech decoupling is accelerating a quiet, desperate race to control the physical hardware and supply chains behind AI.
Semiconductors & Hardware
Shanghai Yuliansheng to Ship Five Immersion DUV Machines This Year Shanghai Aishengna Technology, leveraging the team from state-backed Shanghai Yuliansheng Technology, plans to ship five domestic immersion DUV lithography machines to Chinese IC manufacturers, including SMIC, Hua Hong, and CXMT, this year. These machines are designed for 28nm processes with single exposure and multi-patterning for 7nm or 5nm. Aishengna aims to ship 20 units next year, signaling China’s accelerated efforts in self-sufficiency for critical semiconductor manufacturing equipment.
Why it matters: China achieving domestic DUV lithography machine shipments, even if at older nodes or lower throughput, means US export controls are getting circumvented in practice. This isn’t just an announcement; it’s about physical equipment being delivered to major fabs like SMIC. It shows how China is approaching self-sufficiency not with a single breakthrough, but by systematically replacing foreign components in critical supply chains.
For Western readers: Western semiconductor equipment manufacturers, particularly those in lithography, should anticipate a sustained, aggressive erosion of their market share in China, first at mature nodes, then at increasingly advanced ones, as these domestic machines scale in production and capability. Semiconductors & Hardware
TrendForce: NVIDIA and Broadcom Begin Volume Ramp of CPO Switches NVIDIA and Broadcom are beginning volume production of Co-packaged optics (CPO) switches, marking a shift from early adoption to commercial deployment for this power-efficient networking technology. However, the market’s expansion faces significant supply constraints in optical engines, silicon photonics (SiPh) chips, and advanced packaging capacity, which are largely concentrated in East Asian manufacturing hubs.
Why it matters: This isn’t just about faster switches; it’s about control over the underlying supply chain that enables AI infrastructure. The power efficiency CPO offers is critical for hyperscale operators, but the real leverage lies with those who can secure manufacturing capacity for optical engines, silicon photonics, and advanced packaging, much of which is in Taiwan.
For Western readers: If you are a hyperscale cloud provider or AI developer, understand that the availability and cost of your next-gen network infrastructure will be dictated by the tight market for advanced packaging and SiPh components from East Asia. Actively secure supply chain commitments, or expect slower deployments and higher costs. Semiconductors & Hardware
Power Electronics Market Entering New Growth Phase, Driven by EVs and AI The global power electronics industry, particularly SiC and GaN technologies, is entering a new growth phase, propelled by electric vehicles (EVs) and AI data centers. China is identified as the dominant market for SiC-equipped BEVs, with companies like BYD demonstrating strong vertical integration across the SiC supply chain.
Why it matters: The fact that China continues to widen its lead in SiC-equipped BEVs, particularly through integrated players like BYD, means they are not just consuming the technology but shaping its development and supply chain. This is not just a market share story; it’s about control over a foundational technology for electrification.
For Western readers: Western automotive and power electronics firms need to assess how China’s vertical integration in SiC for EVs affects their own long-term market access and supply chain stability, especially given the rising efficiency demands from AI data centers which also use these technologies. Engineers from Samsung’s semiconductor divisions are applying en masse to SK Hynix, driven by significantly higher bonuses at SK Hynix stemming from its profitability in high-bandwidth memory (HBM) chips for AI accelerators. Samsung’s internal bonus structure, which ties payouts to individual division performance, has left foundry workers feeling shortchanged compared to their memory division counterparts and SK Hynix employees.
Why it matters: SK Hynix’s ability to attract top Samsung talent through superior HBM profitability signals a real shift in who controls the critical HBM engineering bench. This isn’t just about money; it’s about where the best minds in next-gen memory are concentrating, which directly influences production roadmaps and IP development for the entire AI hardware stack. Western firms relying on these components need to understand the underlying drivers of that talent shift.
For Western readers: Western AI accelerator developers and cloud providers sourcing HBM from Samsung or SK Hynix should factor this talent drain into their risk assessments; a stronger SK Hynix could lead to more stable supply and faster innovation, but a weakened Samsung foundry talent pool might introduce future supply constraints for logic chips.