East Asian Technology Intelligence

Japan & China tech news — translated, contextualized, and delivered weekly.

Subscribe Free →

Weekly. Free. Unsubscribe anytime.

3 Takeaways This Week

• China’s aggressive domestic production push in the $73 billion semiconductor materials market is directly squeezing the market share of established Japanese suppliers like Showa Denko and Sumitomo Chemical.
• The emergence of a bustling grey market for ChangXin Memory Technologies (CXMT) DRAM chips demonstrates that US Entity List sanctions are failing to choke off the global distribution of Chinese memory silicon.
• Japan’s ambitious 370 trillion yen ($2.3 trillion) technology investment strategy will struggle to achieve its startup-led growth goals because METI’s top-down allocation model historical favors entrenched industrial conglomerates over high-risk, independent ventures.

🗾 Semiconductors & Hardware

Kioxia Begins Sample Shipments of New Memory, Utilizing Advanced Equipment at Iwate Factory

Kioxia Holdings announced on July 3rd that it has begun shipping samples of its new, high-capacity, low-power memory products, manufactured at its advanced second production facility in Kitakami, Iwate Prefecture. These next-generation NAND flash memories boast 33% faster data processing and lower power consumption, targeting the expanding data center demand driven by AI proliferation. The company aims for mass production following customer evaluation. Kioxia’s move to ramp up advanced NAND production, especially with new generations like this 10th-gen memory, is a direct response to the AI boom. While Western discourse often focuses on AI processors themselves, the memory fabric that feeds those processors is equally critical, and this is where Japanese firms still hold significant ground.

For Western readers: Western data center operators and AI infrastructure providers should expect continued tight supply in high-performance NAND memory, with Kioxia expanding its contribution but overall demand growing even faster. Diversify sourcing now or plan for longer lead times.

ITmedia AI+

🗾 AI & Machine Learning

AI is Smart Enough: Bottleneck Shifts from Model Performance to Evaluation, Says Databricks AI Scientist

Jonathan Frankle, Chief AI Scientist at Databricks and co-founder of MosaicML, states that current AI models are intelligent enough. He argues the primary bottlenecks for AI adoption are now ‘evaluation,’ ‘governance,’ and ‘cost efficiency,’ not further model performance improvements.
Frankle emphasizes the difficulty and importance of rigorously evaluating AI outputs, comparing it to autonomous driving where even a single error can have widespread consequences, requiring ‘orders of magnitude more rigorous evaluation’ than human tasks. The Japanese business press often emphasizes practical application and risk mitigation over headline-grabbing benchmarks. This article’s focus on ‘evaluation’ and ‘governance’ reflects a deeply engrained Japanese corporate ethos of quality control and long-term reliability in industrial systems, rather than simply chasing the most advanced algorithms. It suggests Japanese firms will prioritize robust integration and safety mechanisms for AI.

For Western readers: Western businesses should allocate more resources to AI evaluation and governance tools, rather than focusing solely on model selection, to ensure reliable and safe deployments that can meet stringent operational requirements.

ITmedia AI+

Policy & Regulation

Japan Targets $2.3 Trillion Investment in Tech Strategy, Prioritizing Startups

Japan is embarking on a substantial 370 trillion yen ($2.3 trillion) investment plan across 17 strategic sectors including AI-powered robots, semiconductors, and autonomous driving. This initiative uniquely centers on fostering startups, a notable shift from its historical corporate-centric approach to innovation. The focus on startups marks a pivot from Japan’s previous industrial policies, aiming to cultivate indigenous innovation rather than relying solely on established giants; Western media highlights the arrival of US venture capital as a sign of renewed interest.

For Western readers: Western businesses and investors, particularly in venture capital and deep tech, will find new opportunities for partnerships and market entry in Japan’s burgeoning startup ecosystem.

Nikkei Asia

Semiconductors & Hardware

China chip material makers battle Japan rivals for $73bn market

Chinese manufacturers are aggressively increasing production of advanced chip materials, directly challenging established Japanese market leaders like Nittobo, as Beijing prioritizes semiconductor self-sufficiency amidst a global push for AI chip packaging materials. The article underscores the increasing technological and economic rivalry between China and Japan, particularly in foundational semiconductor supply chain components, as China seeks to localize sophisticated manufacturing capabilities. Western media often frames this as a ‘tech war’ or ‘decoupling’ narrative.

For Western readers: Western semiconductor firms relying on Japanese materials or looking for alternative suppliers will find China’s growing capabilities both a potential opportunity and a geopolitical consideration.

Nikkei Asia

Startups & Funding

SAZO, a Japanese AI cross-border e-commerce service, secures funding from NAVER-affiliated fund

Japanese AI cross-border e-commerce service SAZO, which simplifies overseas product purchases, has received investment from NAVER D2SF, a venture capital arm of the South Korean internet giant NAVER. This funding follows a ¥710 million pre-Series A round in May 2025 and aims to accelerate global expansion, including to North America and Europe. This deal underscores the persistent drive in East Asia to smooth out international trade for individual consumers using AI, often framing it as consumer convenience rather than industrial policy. SAZO’s collaborations with Mercari and Rakuten Ichiba for the Korean market show how established platforms are already thinking about these AI tools for market access, rather than building similar functionality from scratch.

For Western readers: Western e-commerce and logistics providers should note that Asian players like SAZO, backed by giants like NAVER, are actively building out AI-driven cross-border infrastructure that could eventually compete for international parcel flow currently served by traditional players. If you are in global logistics, pay attention to the partnerships these platforms form in North America and Europe.

The Bridge (Japan Startups)

Robotics & Automation

74 Yuan for 3 Hours: Embodied AI Robots Begin Entering Chinese Homes for Household Chores

Embodied AI robots are starting to be deployed in Chinese homes for household chores, with companies like Huitianyu Technology offering services at a rate of 74 yuan for three hours. This marks a tangible step towards commercializing advanced robotics and AI in the vast Chinese consumer market, addressing an increasing demand for domestic assistance. This development signals China’s leading edge in rapidly integrating advanced AI robotics into daily life, which could create a significant domestic market advantage and accelerate R&D. Western media often focuses on industrial robotics, potentially overlooking China’s aggressive push into consumer and service robotics.

For Western readers: Western robotics companies and investors should monitor the scalability and acceptance of these services in China, as it could provide a blueprint for similar consumer-oriented robotics deployments globally.

Pandaily

AI & Machine Learning

CAS Institute of Software Launches Reasoning Lens, Making AI Model Thinking Processes Visible

The Chinese Academy of Sciences (CAS) Institute of Software has launched Reasoning Lens, a tool designed to enhance the interpretability of AI models by visualizing their reasoning processes. This development from a leading Chinese research institution aims to address the ‘black box’ problem in AI, a critical step for deploying reliable AI across various sectors in China. China’s focus on AI interpretability is a key component of its broader AI strategy, aiming to build trusted AI systems for both commercial and national security applications. This contrasts with some Western approaches that prioritize performance over transparency, though interpretability is an increasingly important global research area.

For Western readers: Western AI developers and policymakers should note China’s advancements in AI interpretability as a potential area for future collaboration or competition, particularly in standards development for explainable AI.

Pandaily

Policy & Regulation

ByteDance’s Doubao and Alibaba’s Qwen to Shut Down AI Agent Features

ByteDance’s Doubao and Alibaba’s Qwen platforms are discontinuing their AI agent creation features and user-generated agents on July 15, 2026. This move coincides with China’s new Interim Measures for the Administration of Anthropomorphic AI Interaction Services, which also take effect on that date. Existing agent data will be purged after October 15, 2026. China’s tech giants prioritize regulatory compliance above all else when new rules come out, even if it means shuttering features with user adoption. This isn’t just about privacy; it’s about controlling narratives and content generated by AI, mirroring how other online content has been managed. The speed of the response shows that platforms are not pushing back on these directives.

For Western readers: Western companies operating or planning to operate AI services in China should assume a high degree of regulatory scrutiny on user-generated AI content, and that open-ended AI creation tools will face significant hurdles or outright bans. Do not plan products that rely on an open, uncensored user-generated AI content model in China.

TechNode

Semiconductors & Hardware

Huawei Mate 90 Series to Feature Kirin 2026 Chip with Tao (τ) Law Logic Folding

Huawei’s upcoming Mate 90 series, expected this autumn, will reportedly integrate the new Kirin 2026 chipset, which implements the company’s recently introduced Tao (τ) Law for semiconductor development. This new framework, unveiled in May, focuses on reducing system time constants through techniques like Logic Folding to increase transistor density and performance. The Kirin 2026 is said to utilize dual-layer Logic Folding, achieving a 53.5% increase in transistor density over its predecessor. Huawei’s ‘Tao Law’ and ‘Logic Folding’ are a technical workaround for the fact that they cannot access leading-edge lithography tools. This is a design-centric strategy to get more out of older process nodes, similar to how China aims to improve existing fabs rather than chasing ASML’s latest. It’s about maximizing yield and efficiency from available rather than aspirational technology.

For Western readers: Western chip designers and foundries should track the actual performance gains and power efficiency of these Kirin chips, as Huawei is demonstrating how far architectural innovation can stretch older fab processes. This could influence R&D priorities if the gap between ‘leading-edge’ and ‘optimized older node’ narrows for specific applications.

TechNode

Policy & Regulation

Hong Kong anti-drug video fiasco raises concerns about government AI competence

Hong Kong’s Correctional Services Department released an AI-generated anti-drug video featuring K-pop characters that inadvertently glorified narcotics, forcing its removal twice. The incident overshadowed Chief Executive John Lee’s July 1st handover anniversary speech, sparking international ridicule and raising questions about the government’s ability to manage AI applications. The Hong Kong government’s misstep with this AI video indicates a significant gap between policy ambitions for AI adoption and the practical capabilities of its agencies to execute. It’s not just a public relations blunder; it points to a lack of understanding within government departments about AI’s potential pitfalls and the need for robust oversight mechanisms. Local commentary focuses on the embarrassment and incompetence rather than the broader implications for AI development, which misses the point that foundational AI infrastructure is worthless without competent deployment.

For Western readers: Western businesses considering partnerships or technology transfer with Hong Kong government entities for AI projects should prioritize thorough due diligence on project management, technical literacy, and internal review processes, not just the stated AI capabilities. Assume government agencies may lack the internal controls needed for effective AI deployment until proven otherwise.

South China Morning Post — Tech

Semiconductors & Hardware

Micron breaks ground on $9bn Hiroshima fab expansion with Japanese government support

Micron has commenced a $9.3 billion expansion of its Hiroshima fab, with the Japanese government providing $4.8 billion in subsidies. This facility is critical for Micron’s High Bandwidth Memory (HBM) production, with the first HBM production wafer having originated there. The Japanese government’s substantial subsidy, nearly 50% of the project cost, shows Tokyo is serious about onshoring critical chip production. This isn’t just about jobs; it’s about securing domestic access to advanced memory like HBM, which is foundational for AI, and ensuring Japan remains a key player in the global semiconductor ecosystem rather than just a consumer.

For Western readers: Western semiconductor companies considering expansions in advanced memory production should now factor in direct and significant Japanese government co-investment as a standard part of their site selection calculus, recognizing it as a strategic priority for Tokyo.

Electronics Weekly

Semiconductors & Hardware

Chinese Memory Chips Flow into Grey Market Amid US Sanctions

Amidst US Entity rules restricting legal purchases of Chinese chips, a significant grey market for Chinese DRAM and Flash memory from manufacturers like CXMT and YMTC has emerged. This illicit trade channels several billion dollars worth of chips, sourced from rejects and counterfeits in China, into Western markets via smuggling routes. The demand is driven by high prices from the ‘Big Four’ memory producers and the ongoing AI boom. The resilience of grey market channels for Chinese memory reveals how difficult it is to enforce broad-based technology sanctions when there’s an arbitrage opportunity. It means the “Big Four” face continued competition, even if indirect, that keeps prices from escalating unchecked, especially with the high demand for memory in AI applications. Western governments might believe their controls are effective at the top end, but the reality on the ground is more fluid and harder to contain.

For Western readers: Western companies needing memory for non-critical applications, particularly in Europe, can expect to find Chinese-made DRAM and Flash available below market rates through informal channels for the foreseeable future. However, assume these chips come with no guarantees on quality or origin.

Electronics Weekly

Semiconductors & Hardware

SiTime Completes Acquisition of Renesas’ Timing Business

SiTime Corp. has completed its acquisition of Renesas Electronics’ timing business, a move that will significantly expand SiTime’s clocking portfolio and presence in the high-growth AI datacenter market. The deal, first announced in February 2026, includes a strategic collaboration MOU for SiTime to integrate its MEMS resonators into Renesas’ embedded computing products. Renesas CEO Hidetoshi Shibata is also expected to join SiTime’s Board of Directors. Renesas is shedding a legacy business with 70% gross margins and 10,000 customers, rather than investing in it. This signals a sharp focus on its core embedded computing and power management strategy, rather than maintaining diversified component offerings. It also shows a willingness to partner with a non-Japanese specialist like SiTime, rather than keeping the business in-house or selling it to another Japanese entity, which is a pragmatic move to secure supply while divesting.

For Western readers: If you’re tracking consolidation in the semiconductor industry, recognize that Japanese firms like Renesas are increasingly willing to divest mature, profitable, but non-core units to foreign entities to sharpen their strategic focus, rather than retaining them for national champion narratives. This means more opportunities for Western specialized component makers.

EE Times Asia

Semiconductors & Hardware

Kioxia and Sandisk Start Production of 10th-gen 3D Flash Memory Devices

Japan-based Kioxia and its US joint venture partner Sandisk have initiated production of their 10th-generation 3D flash memory at the K2 facility in Kitakami, Iwate Prefecture. This facility, operational since September 2025, is scaling up advanced memory output to meet growing demand, particularly from the AI market. The partners highlighted continued investment and their extended joint venture framework through December 2034. Kioxia and Sandisk’s investment in the K2 fab reinforces the strategic importance of Japan in the global NAND flash supply chain, particularly for high-performance applications like AI. Unlike some of the newer national champion consortia, this is an established partnership with a track record of shipping product, indicating a genuine increase in capacity and technological advancement rather than just an announcement of intent.

For Western readers: If you are a major procurer of advanced NAND flash memory, factor this new production capacity in Japan into your supply chain resilience planning; it provides a stable source from a politically aligned region compared to potentially more volatile regions.

EE Times Asia

Cross-Regional Analysis

Thailand Secures $4.1B in EV Supply Chain Investment, Featuring Contributions from China, Korea, Japan

Thailand has secured over $4.1 billion in EV supply chain investment across 198 projects, with significant contributions from Chinese, Japanese, and Korean manufacturers. These investments are distributed across the entire EV ecosystem, including battery electric vehicles (BEVs), hybrids, battery production, critical components, and charging infrastructure. Chinese firms like BYD and SAIC are leading BEV production, while Japanese automakers are leveraging their hybrid technology expertise. Western coverage of EV manufacturing often focuses on the US-China competition, but this investment demonstrates how Southeast Asian nations like Thailand are actively shaping the regional supply chain. For Japan, using Thailand to bolster its hybrid tech means it is hedging its bets against pure BEV dominance and maintaining its existing strengths. Chinese firms are aggressively building capacity outside China, bypassing some tariff risks.

For Western readers: Western automotive companies should recognize that the Southeast Asian EV market is rapidly solidifying around Chinese and Japanese manufacturing bases, making direct entry increasingly difficult and underscoring the need for strategic partnerships or localized production to compete effectively in the region.

Technode Global

🧩 Pattern This Week

• China: local chip material suppliers aggressively expand capacity to challenge Japan
• The Triangle: grey market operations bypass US sanctions to distribute Chinese memory
• Japan: Kioxia counters supply chain threats by sampling advanced memory chips

Beijing’s aggressive localization of semiconductor materials and grey-market distribution of restricted silicon are neutralizing Western export controls, leaving Japanese tool and material giants highly exposed to a rapidly decoupling supply chain.