East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered weekly.
Weekly. Free. Unsubscribe anytime.
3 Takeaways This Week
- CXMT’s 470% debut surge on the Shanghai STAR Market demonstrates how Beijing’s state-guided capital is successfully decoupling China’s semiconductor funding from Western financial markets.
- SpaceX’s strict exclusion of Chinese nationals and components from its global supply chain forces aerospace and defense suppliers in Japan and South Korea to accelerate the costly partition of their dual-use technology pipelines.
- Kioxia’s 4,500% year-on-year net profit surge for the quarter ending June 30, 2026, secures the capital runway the Japanese flash memory giant needs to fund its next-generation fabrication lines without relying on a defensive Western merger.
Core Move
Thinking Machines Officially Releases Lightweight Model ‘Inkling-Small’: Quarter the Size, Equivalent Performance to ‘Inkling’
📊 Featured Chart
Inkling-Small scores (Inkling scores: 29.7, N/A, N/A)
Thinking Machines released its new Inkling-Small AI model. This launch shows a big shift in the business AI market. The new model focuses on easy setup and low resource use instead of massive size. This focus helps Asian firms that already own their own computer hardware.
The model cuts GPU memory use by 75% but keeps performance high. This change allows companies to run the model on their own local servers. Local setups are vital for fields that need fast speeds and strict data control. Japanese companies have long worried about using the public cloud because of data security rules. A lighter, local model solves this major problem for them.
Japanese firms have been slow to adopt large language models. Giant Western models cost too much and are too hard to run. Most Japanese firms lack the massive cloud budgets of Western rivals. Inkling-Small challenges the idea that bigger models are always better. Japan is now optimizing AI for its existing industrial hardware instead of buying new cloud systems.
The model also handles audio and images directly. This multi-mode feature makes it very useful for factories. Factory staff can use it for quality control, machine maintenance, and data analysis. These tasks rely on many data types beyond simple text.
Yet, Japanese firms still face challenges. Mixing these models with old IT systems will meet resistance from internal tech teams. Japan also has a major shortage of skilled AI engineers. Many local firms build simple trial projects. The true test is turning those trials into full, working systems.
Domestic tech giants like SoftBank and Fujitsu are building their own AI platforms. Inkling-Small gives them a strong, proven model that they can customize. They do not have to build new models from scratch. This choice helps them use top-tier tech while they keep control of their supply chains. In Japan, low compute costs and local rules drive AI adoption more than high test scores.
We must track how many pilot programs Inkling-Small wins in Japan and Korea over the next two quarters. Partnerships with firms like Hitachi or Toyota are especially important. We need to see if these trials turn into real, daily use.
We should also watch if firms change their hardware buying plans. They may choose systems built for this model’s 180GB+ memory needs instead of buying massive systems for larger models. Finally, we should follow statements from the Ministry of Economy, Trade and Industry, or METI, about local AI tools. Their policy goals align well with lighter, easier models.
🗾 Japan Radar
What Japanese media is reporting that Western outlets miss
Tokyo’s industrial base is pivoting to hardware-anchored AI and supply-chain decoupling to hedge against US-China technology friction.
🗾 Robotics & Automation
Google Announces Gemini Robotics 2 AI for Robots, Enabling Full Humanoid Control and Fine Motor Tasks
Google and Google DeepMind have announced Gemini Robotics 2, a suite of AI models designed for robotics. This new iteration allows for full-body control of humanoid robots, execution of fine motor tasks with specialized hands, and coordination among multiple robots, moving beyond previous models that only controlled upper-body tasks.
Why it matters: Google’s move into full-body humanoid control with Gemini Robotics 2 is a direct challenge to established robotics players in Japan and Korea, who often focus on hardware and specific industrial applications. By emphasizing multimodal AI that integrates vision, language, and action, Google aims to broaden the application scope of humanoid robots from factory floors to more general-purpose use, potentially shifting the competitive landscape toward software and AI model dominance.
For Western readers: Western AI companies, particularly those developing foundational models for robotics, should anticipate increased competition from Google’s integrated hardware-software approach. If you are developing robotic applications, focus on how these new VLA and embodied reasoning models might simplify complex task programming or enable more natural human-robot interaction.
🗾 Semiconductors & Hardware
Kioxia Q1 Net Profit Up 4500% Year-on-Year, Announces Stock Split and Share Buyback
📊 Featured Chart
Q1 FY2027 (Apr 1 – Jun 30, 2026)
Kioxia Holdings reported a significant increase in its Q1 FY2027 (April 1 to June 30, 2026) consolidated financial results, with net profit surging 4506.6% year-on-year to 842.17 billion yen. Revenue grew 415.5% to 1.767 trillion yen and operating profit rose 2728.6% to 1.27 trillion yen, driven by strong NAND demand for AI servers, increased shipments, and favorable exchange rates. The company also announced a 1-for-3 stock split effective September 30 and a share buyback program of up to 800 billion yen between August 30 and October 30.
Why it matters: Kioxia’s performance isn’t just a recovery story; it shows how the AI infrastructure buildout is creating demand ripple effects across the entire semiconductor stack, not just for HBM or logic. NAND, often seen as a commodity, is now directly benefiting from the need for vast storage in AI servers and data centers. The stock split and share buyback are moves to improve shareholder value and liquidity, potentially paving the way for a long-anticipated IPO if this upward trend continues.
For Western readers: Western tech companies, particularly those in cloud infrastructure and AI, should expect continued cost pressure on NAND flash memory as demand remains strong, impacting data center build-out and component sourcing decisions.
Policy & Regulation
SpaceX Moves to Exclude Chinese Nationals, Parts from Supply Chain
SpaceX is implementing strict measures to remove Chinese nationals and components from its global supply chain, aiming to protect US national security technologies. This directive affects partners worldwide, demanding that facilities producing for SpaceX not employ Chinese citizens or use Chinese-made parts.
Why it matters: SpaceX, being a spearhead of US space technology, sets a new benchmark for supply chain de-risking from China, beyond typical IT and telecom sector concerns. This is not merely about sourcing components; it’s about explicitly excluding human capital based on nationality within the supply chain, a significant escalation of decoupling that has broader implications for how companies in sensitive sectors operate internationally.
For Western readers: Western businesses in critical technology sectors should assume that US government pressure, explicit or implicit, will increasingly demand complete exclusion of Chinese nationals and components from their supply chains, even for non-military applications. Begin auditing your entire value chain, including sub-contractors, for Chinese involvement, especially human capital.
Semiconductors & Hardware
CXMT Listing Propels STAR Market, China’s Nasdaq, to Prominence
Chinese chipmaker CXMT’s blockbuster listing on the Shanghai STAR Market, which debuted on July 27, made it the most valuable company on mainland Chinese markets. This strong debut highlights the STAR Market’s increasing importance as a platform for China’s state-backed industrial push in its technology competition with the U.S.
Why it matters: The CXMT listing isn’t just about a single company; it’s a clear signal that Beijing is serious about using its financial infrastructure to insulate critical industries like memory chips from external pressure. This public market success, despite the broader global chip market slump, indicates a deep state commitment to indigenous chip development, even if the underlying economics wouldn’t support such valuations in a purely open market.
For Western readers: Western semiconductor firms should anticipate sustained, state-backed competition from Chinese players like CXMT, particularly in commodity memory, as China prioritizes domestic supply chains even at higher cost.
Semiconductors & Hardware
AI Boomtowns Worsen Divide in China’s Split-Screen Economy
CXMT, China’s leading memory chipmaker, saw its shares surge 470% on its Shanghai market debut on July 27, signaling strong investor confidence in domestic semiconductor firms. This growth is contributing to the rise of ‘AI boomtowns‘ in China, but also exacerbating economic disparities between these tech hubs and other regions.
Why it matters: The dramatic IPO performance of CXMT reflects China’s industrial policy success in channeling capital and talent into strategic technology sectors like memory chips. This indicates that despite Western export controls, Beijing is effectively using its domestic capital markets to build national champions and reduce reliance on foreign technology.
For Western readers: Western semiconductor firms should anticipate increased competition from well-funded, domestically-backed Chinese companies like CXMT, particularly in the memory sector, as China continues to prioritize indigenous production.
🇨🇳 China Watch
China’s technology moves, framed for Western readers
Sanctions are forcing Chinese AI firms to win on extreme compute efficiency, localized enterprise workflows, and overseas open-source distribution.
AI & Machine Learning
Moonshot AI Kimi K3 Open-Source Model Sees Massive Global Download Surge, Boosting Overseas Paid User Growth
Chinese AI firm Moonshot AI’s Kimi K3 open-source large language model, launched two days prior, captured 41% of global open-source model downloads and quadrupled its overseas paid user growth. This rapid adoption suggests significant international interest in China’s rapidly developing AI capabilities. The model’s success indicates Chinese developers are moving swiftly into the global open-source AI landscape.
Why it matters: The rapid global adoption of a Chinese-developed open-source model like Kimi K3 indicates that China is not just aiming for domestic AI leadership but actively seeding its technology internationally. This strategy enables Chinese AI firms to gain significant user data for model refinement and expand their influence, bypassing some of the political friction associated with direct commercial exports.
For Western readers: Western AI firms and policymakers should recognize that ‘open-source’ from China increasingly serves as a vehicle for industrial policy and international influence, not just academic collaboration; expect more such rapid-release models to challenge established Western platforms. If you’re a Western company building on open-source, you need to understand the implications of basing your work on a model that can gather usage telemetry back to its country of origin or be influenced by its developers’ national interests.
Robotics & Automation
Dark Horse Muka Robotics Charges WorldArena With Only 32 Zhenwu 810E GPUs: LJM World Model Achieves Global Second Place Defining the Breakthrough From Video Quality to Physical Logic
Chinese startup Muka Robotics, using just 32 Zhenwu 810E GPUs, achieved a second-place ranking in the WorldArena global competition with its LJM World Model. This performance underscores China’s progress in robot learning and world model development with comparatively constrained hardware resources. The model’s success in mastering complex physical tasks indicates a shift in focus from purely visual fidelity to robust physical logic for robotics.
Why it matters: The key here is efficient resource utilization. Muka Robotics isn’t throwing compute at the problem; they’re showing that clever architectural design and data efficiency can achieve top-tier results in complex robotics tasks, even when high-end GPUs are scarce. This matters for every company, not just those in China, looking to build robust robot control systems without an infinite budget.
For Western readers: Western researchers and robotics firms should closely examine the LJM World Model’s architecture and training methodologies; it suggests that scaling compute isn’t the only path to advanced robotic intelligence, potentially offering a playbook for developing capable systems under hardware constraints.
AI & Machine Learning · Enterprise & Cloud3 STORIES
Tencent WorkBuddy Dominates China’s AI Office Market, Sparking Fierce Tech Rivalry
Tencent’s WorkBuddy has surged past competitors like Alibaba’s DingTalk and ByteDance to become China’s top AI office platform, recently reaching 20.97 million monthly visits and launching a seamless ‘Human-Agent Dual Writing‘ feature. This rapid ascension has triggered an aggressive arms race in the domestic enterprise software market, forcing rivals to drastically accelerate their own AI integration cycles to match Tencent’s ecosystem advantages.
Why it matters: In the East Asian business landscape, capturing the enterprise collaboration suite is the ultimate gateway to monopolizing proprietary corporate data flows, making the AI office suite the primary battleground for B2B cloud dominance in the region.
For Western readers: Western executives must abandon the assumption that Microsoft and Google face no viable global rivals in AI-integrated productivity; Tencent’s rapid scale-up proves that Chinese tech giants have perfected the playbook for native AI-human collaboration at enterprise scale, positioning them to capture high-growth emerging markets.
🔺 The Triangle
Where US, Japan, and China technology interests intersect
Western chip giants are securing the supply chain’s edge, positioning hardware control as the ultimate moat against Chinese architectural advances.
Semiconductors & Hardware
Synopsys, NVIDIA Advance Autonomous AI Agents for Chip Design and Engineering
📊 Featured Chart
Synopsys claims at DAC 2026
Synopsys, in collaboration with NVIDIA, has introduced autonomous AI agents designed to automate complex chip verification, analog design, and thermal engineering workflows. Unveiled at the 2026 DAC Chips to Systems Conference, this technology integrates Synopsys’ EDA/CAE expertise with NVIDIA’s accelerated computing platform, Nemotron AI models, and OpenShell runtime. The goal is to transform labor-intensive engineering processes into automated ones, reducing verification closure times and improving design productivity. The introduction of autonomous AI agents in chip design is a step beyond existing AI-assisted tools; it aims to offload entire workflow segments. For East Asian fabs and design houses, this means faster turnaround on complex designs, but also a tighter integration with, and dependency on, the US-led EDA and AI infrastructure, potentially limiting indigenous innovation paths in these core tools.
Why it matters: The introduction of autonomous AI agents in chip design is a step beyond existing AI-assisted tools; it aims to offload entire workflow segments. For East Asian fabs and design houses, this means faster turnaround on complex designs, but also a tighter integration with, and dependency on, the US-led EDA and AI infrastructure, potentially limiting indigenous innovation paths in these core tools.
For Western readers: Western EDA and AI platform providers should anticipate increased demand from East Asian chip designers seeking to leverage these new autonomous agent capabilities to maintain competitive product development cycles.
Semiconductors & Hardware
DigiKey Expands Inventory with 27,000 New Products and 104 Suppliers in Q2 2026
📊 Featured Chart
Q2 2026
DigiKey, a global electronic components distributor, added over 27,000 in-stock products and 104 new suppliers in Q2 2026, improving immediate availability for engineers. This expansion, including suppliers like ABLIC from Japan, aims to shorten lead times and reduce minimum order quantities for prototype development worldwide.
Why it matters: DigiKey’s strategy of broadening its ‘Marketplace’ and ‘Fulfilled by DigiKey’ programs bypasses traditional direct sales channels, giving smaller East Asian component manufacturers a global distribution platform. This allows them to reach Western design engineers without significant overhead, potentially accelerating the adoption of niche or specialized components from the region.
For Western readers: Western design engineers should increasingly check major distributors like DigiKey for specialized components from lesser-known East Asian suppliers, as these platforms are becoming critical conduits for faster access to new technologies, avoiding the long lead times of direct procurement.
Embedded Systems
Moxa’s UC Series Arm-based Computers Achieve IEC 62443-4-2 SL2 Certification
Moxa Inc., a Taiwanese industrial connectivity and automation company, has secured IEC 62443-4-2 Security Level 2 (SL2) certification for its entire UC Series 64-bit Arm-based industrial computers. This expands on its 2023 certification of a single product, establishing a consistent security baseline across its data gateways and 5G edge-computing platforms for operational technology (OT) applications.
Why it matters: Securing the industrial edge against cyber threats is a growing concern, especially with increased adoption of 5G and IoT in critical infrastructure. Moxa’s certification sets a benchmark for integrated security in industrial embedded systems, which helps original equipment manufacturers and system integrators shorten their own qualification times and reduce project risk when deploying these devices. This isn’t just a compliance story; it reflects a tangible reduction in integration friction for major industrial deployments.
For Western readers: Western industrial solution providers and critical infrastructure operators should prioritize hardware vendors with established IEC 62443 certifications, as it simplifies regulatory compliance and reduces cybersecurity integration costs for large-scale deployments, especially as global standards converge on this framework.
Microchip Technology, a US-based embedded controller company, has acquired Israeli edge AI chip startup Hailo. Hailo is an established player with 100 industrial and automotive customers and a developer community of 10,000, 80% of whom use Raspberry Pi hardware. This acquisition marks Microchip’s first major deal since CEO Steve Sanghi returned to lead the company.
Why it matters: Microchip’s strategy of acquiring companies for cross-selling opportunities, rather than deep technology synergy, means Hailo’s edge AI capabilities will now be pushed into a much broader range of industrial and automotive customers globally. For East Asian industrial players, this expands the available ecosystem for integrating edge AI, potentially making it easier to adopt these solutions rather than developing in-house or relying on fragmented local providers.
For Western readers: Western industrial and automotive companies should expect Microchip to aggressively cross-sell Hailo’s edge AI solutions within its existing customer base, potentially increasing competitive pressure on smaller, specialized AI hardware providers. This also reinforces the trend of consolidation in the embedded AI space.