East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered to your inbox.
Free. Unsubscribe anytime.
3 Takeaways This Issue
- South Korea and Taiwan both surpassed Japan in total exports for the first half of 2026, driven by a surge in high-bandwidth memory and advanced foundry shipments that underscores Japan’s vulnerability in the hardware layer of the AI supply chain.
- Sakura Internet’s deployment of Preferred Networks’ “PLaMo 3.0 Prime” model on its domestic cloud infrastructure represents a METI-supported effort to build a sovereign AI alternative, though it faces a widening capability gap as OpenAI rolls out its GPT-5 “Sol” model to paid users.
- Japan’s Ministry of Justice finalized a report recommending legal protections for human voices against unauthorized AI cloning, establishing a distinct regulatory framework that will complicate localization efforts for Western voice-synthesis developers in the Japanese market.
Core Move
Global Chip Sales Up 35% in Q2 2026, Led by Americas, Asia Pacific, and China
📊 Featured Chart
Year-to-year sales increase
China’s big 112.8% year-over-year surge in semiconductor sales shows that Washington’s encirclement strategy has backfired. This growth accelerated the creation of a self-sustaining, decoupled East Asian supply chain. It did not choke off Chinese technological progress. Western commentators frequently interpret these spike numbers as evidence of desperate double-ordering or unsustainable state-subsidized inventory stuffing. However, domestic industry coverage in Shenzhen and Beijing paints a very different picture.
Inside the mainland, people celebrate this growth as a milestone of internal circulation maturity. Domestic automotive, consumer electronics, and industrial giants are actively swapping out foreign silicon. They are replacing foreign chips with mature-node and near-frontier domestic alternatives. This is not a temporary bubble. It is the structural solidification of a parallel semiconductor ecosystem.
At the same time, Japan’s quiet 39% year-over-year growth signals a highly coordinated, state-backed renaissance. This effort is fundamentally different from China’s defensive push. It is Japan’s version of post-war industrial reconstruction, but applied to high-bandwidth memory (HBM) and advanced packaging rather than steel and automotive assembly. The Ministry of Economy, Trade and Industry (METI) is not trying to build a completely independent stack. Instead, METI is systematically tying Japan’s domestic material giants, like Tokyo Ohka Kogyo and JSR, directly to global foundry giants. The domestic media focus in Tokyo is not on national self-sufficiency. It focus on securing Japan’s position as the indispensable gatekeeper of the global AI physical layer.
However, the prevailing market assumption that this broad-based Asian chip boom will lift all boats equally ignores a critical structural bottleneck. The primary risk to this dual-track expansion is the imminent domestic shortage of specialized chemical and wafer engineering talent within Japan. A shrinking working-age population cannot support the simultaneous fabrication demands of Rapidus, TSMC Kumamoto, and Micron Hiroshima. If these fabs cannot run at high yields due to personnel constraints, the billions invested by METI will yield expensive monument projects rather than commercial market share.
To gauge whether this regional momentum is sustainable, watch three concrete indicators over the next two quarters. First, monitor the yield rates and commercial wafer starts at TSMC’s Kumamoto Fab 1 to see if Japan can actually execute on mass-producing logic silicon. Second, track the quarterly capital expenditure reports of China’s SMIC. This will confirm if their legacy-node expansion continues to outpace Western tool restrictions. Finally, watch for any major procurement shift by Japanese automotive consortiums toward domestic Renesas chips over European legacy suppliers. That shift would signal a deliberate move toward regional supply chain hardening.
🗾 Japan Radar
What Japanese media is reporting that Western outlets miss
Japan is ceding the raw compute race to US hyperscalers, pivoting instead to domestic LLM integration and voice sovereignty.
🗾 AI & Machine Learning
ChatGPT Default Models Shift: ‘Sol’ for Plus/Pro, ‘Luna’ for Free Version with Unlimited Text Chat
OpenAI announced updates to its default ChatGPT models, with paid Plus and Pro plans now using GPT-5.6 Sol, and the free and low-cost ChatGPT Go plans switching to GPT-5.6 Luna. Sol is optimized for accuracy in specific domains, while Luna will offer unlimited text chat for free users and a ‘Think’ button for deeper reasoning. Both models show significant improvements in reducing factual inaccuracies compared to GPT-5.5 Instant.
Why it matters: The introduction of a ‘Think’ button for deeper reasoning in the free version, combined with unlimited text chat, is a clever way to hook users who might otherwise be drawn to competing models. OpenAI is improving accessibility while also setting expectations for paid services: if you want serious work done, you’ll still need the Plus or Pro version and its higher accuracy.
For Western readers: Western businesses building applications on OpenAI’s API or competing with ChatGPT should prepare for a wider and more sophisticated user base on the free tier, which will drive demand for more advanced, specialized models or custom solutions as users hit Luna’s limits.
🗾 AI & Machine Learning
PFN’s Domestic LLM “PLaMo 3.0 Prime” Now Available on Sakura Internet’s AI Inference Platform; Access by Application
Sakura Internet announced on August 4th the launch of Preferred Networks (PFN)’s “PLaMo 3.0 Prime” large language model (LLM) on its “Sakura AI Engine” inference API platform. Access to the model requires an application and is not available on free plans. PLaMo 3.0 Prime, PFN’s latest flagship domestic generative AI model developed from scratch, builds on a pre-trained model from a joint research effort with NICT and was officially released by PFN on June 22nd.
Why it matters: The availability of PFN’s PLaMo 3.0 Prime on Sakura Internet’s platform signals a maturing of Japan’s domestic AI ecosystem, particularly in offering enterprise-grade LLM access. This move not only enhances PFN’s reach but also gives Japanese companies a viable, locally hosted option that aligns with data sovereignty concerns, potentially reducing reliance on overseas cloud infrastructure and models for sensitive applications. The focus on improved Japanese language performance and contextual understanding targets specific needs of the domestic market.
For Western readers: Western cloud providers and AI model developers seeking to penetrate the Japanese enterprise market should expect increasing competition from domestic alternatives like PLaMo 3.0 Prime, especially where data residency, cost-efficiency, and nuanced Japanese language performance are key decision factors for local businesses.
AI & Machine Learning
4 US Tech Giants Amass $1.46 Trillion in Physical Assets for AI Build-Out
Four major US tech companies — Amazon, Alphabet, Microsoft, and Meta — have increased their property, plant, and equipment (PP&E) assets by 140% in three years, reaching $1.46 trillion, driven by AI investments. Amazon alone doubled its PP&E to $538.7 billion, making it the largest company globally by this measure. This shift transforms these historically ‘asset-light’ software businesses into infrastructure giants, mirroring the capital intensity of traditional energy majors, which has implications for data center development in East Asia.
Why it matters: The capital-intensive nature of AI infrastructure build-out fundamentally changes how we assess the global tech landscape. These US firms are becoming industrial players, not just software providers, which will force East Asian competitors and governments to reconsider their own investment strategies in foundational AI infrastructure. This isn’t just about software capabilities anymore; it’s about who owns the physical plant.
For Western readers: Western businesses supplying or competing in the AI infrastructure space should recognize that the capital demands for AI are now on par with traditional heavy industry, meaning slower ROIs and larger upfront investments are the new norm. If you’re looking to build data centers in Asia, expect land, power, and high-spec component costs to continue rising as these US giants outbid others.
Cross-Regional Analysis
South Korea, Taiwan Top Japan in Exports for First Time on AI Boom
For the first half of 2026, South Korea and Taiwan each surpassed Japan in total exports, marking a significant shift driven by booming global demand for AI-related semiconductors. This export surge highlights the increasing economic leverage of East Asia’s advanced chip manufacturers, particularly memory and logic foundries, benefiting more from the AI upswing than Japan’s more diversified industrial export base.
Why it matters: This isn’t just about export volumes; it’s about the evolving structure of global technology value chains. The AI boom is concentrating economic benefit in regions with leading-edge semiconductor manufacturing, which are increasingly South Korea and Taiwan. Japan, while strong in equipment and materials, is not capturing the same direct export value from the finished advanced chips.
For Western readers: Western businesses reliant on, or competing in, high-growth technology sectors should recognize that the centers of economic gravity for advanced hardware are shifting further toward South Korea and Taiwan. This implies tighter integration with their supply chains and potentially greater exposure to geopolitical risks impacting these two economies.
🗾 Policy & Regulation
Voice to Be Protected as a Right: Ministry of Justice Expert Panel Report Finalized
Japan’s Ministry of Justice expert panel has finalized a report recommending that human voice be legally protected as an intellectual property right. This moves to establish a new right akin to copyright for voice, separate from existing protections for performances or recordings, specifically addressing the challenges posed by AI voice synthesis.
Why it matters: Protecting voice as a distinct right allows individuals to control AI training data and commercial uses of their voice, which is a common monetization stream for content creators and public figures. For companies building voice AI products, this means needing explicit consent and potentially licensing agreements to use voices, even if sourced from publicly available content.
For Western readers: Western companies developing AI voice technologies or operating in Japan’s digital content market should prepare for a new compliance layer around voice data. This could set a precedent that influences similar legislative discussions in Europe and the US, where similar concerns about voice deepfakes and intellectual property are growing.
🇨🇳 China Watch
China’s technology moves, framed for Western readers
China is pivoting from raw compute chasing to hardware-software integration, optimizing domestic silicon to bypass Western chip bottlenecks.
AI & Machine Learning
China’s AI Infrastructure Race Has Quietly Moved to “Supernodes”
Chinese technology firms are shifting their focus from individual AI chips to integrated “supernodes,” which are large-scale, self-contained AI computing clusters, to circumvent U.S. sanctions and build domestic AI capabilities. This strategy, highlighted at the 2026 World Artificial Intelligence Conference (WAIC), involves companies like Biren Technology and Yixin AI developing full-stack solutions encompassing hardware, software, and large language models within these localized data centers. The goal is to provide comprehensive, sanction-resilient AI infrastructure for enterprise and government clients across China.
Why it matters: The shift to ‘supernodes’ in China is a pragmatic response to export controls, aiming to maximize the utility of available domestic hardware by tightly integrating it with software and application layers. This approach prioritizes deployable, secure domestic solutions over achieving benchmark parity with unrestricted Western systems, and it leverages China’s existing strengths in large-scale infrastructure deployment.
For Western readers: Western businesses operating in China or competing with Chinese AI providers should anticipate more sophisticated, vertically integrated Chinese AI offerings that are less dependent on Western components, and adjust their market strategy accordingly, particularly in enterprise AI and cloud services.
AI & Machine Learning
CASIA’s PhiZero Introduces ‘Physical Language’ for World Models, Reducing Token Usage by 175x
Researchers at the Chinese Academy of Sciences Institute of Automation (CASIA) have developed PhiZero, a new ‘world model’ that uses a ‘physical language‘ for representing environmental data. This approach significantly reduces the number of tokens required to represent physical states, achieving a 175x reduction compared to traditional methods.
Why it matters: The token reduction achieved by PhiZero is a practical step towards more efficient AI training and deployment in physically grounded applications like robotics and autonomous systems. Less data needed means faster iteration and lower compute costs, directly impacting the feasibility of deploying advanced AI in real-world scenarios, particularly relevant for China’s industrial automation goals.
For Western readers: Western researchers and developers should investigate PhiZero’s ‘physical language’ approach for its potential to streamline AI model development and reduce computational overhead in applications that require an understanding of physics, especially in robotics and simulation where data efficiency is a bottleneck.
AI & Machine Learning
DeepSeek Plans Significant API Price Increases for AI Services
Chinese AI model developer DeepSeek has announced impending, potentially substantial, price increases for its API services. The company has not yet disclosed new rates or an effective date, but users have been forewarned of the change.
Why it matters: DeepSeek’s decision to raise API prices, particularly without immediate disclosure of new rates, shows a growing confidence in its model’s value. This suggests that the intense price competition among Chinese foundational model providers may be moderating, allowing leading players to begin monetizing their offerings more aggressively.
For Western readers: Western businesses using or considering DeepSeek’s APIs should prepare for increased operational costs and assess the competitiveness of alternative providers. This could also indicate a broader trend towards price adjustments from other Chinese AI model developers as the market consolidates and matures.
Semiconductors & Hardware
Huawei Launches 798-Gram MateBook Pro S with Kirin XE90 Processor
📊 Featured Chart
Source: Huawei, 2026
Huawei has introduced the MateBook Pro S, a lightweight 798-gram laptop, featuring its domestically developed Kirin XE90 PC processor and running HarmonyOS. The device targets the premium segment of the Chinese consumer market, with prices starting at RMB7,999 and higher-end versions reaching RMB14,999.
Why it matters: Huawei’s ability to integrate its own Kirin XE90 processor into a competitive consumer laptop indicates a maturation of its domestic chip design capabilities, an outcome directly driven by US sanctions. This isn’t just a component swap; it reflects a systemic effort to build out a full-stack, end-to-end domestic hardware and software ecosystem.
For Western readers: Western PC manufacturers and chip designers should anticipate Huawei to capture an increasing share of the premium domestic Chinese laptop market, driven by nationalistic purchasing incentives and the growing integration of HarmonyOS and Huawei’s AI features.
Cross-Regional Analysis
China Rolls Out a Cluster of Heavyweight Technological Breakthroughs
China has announced several significant technological advancements, including a new high-speed steel production method capable of continuous casting, a substantial upgrade to its AI satellite capabilities, and progress in its ‘artificial sun’ nuclear fusion project. These breakthroughs underscore China’s concerted push for self-reliance and global leadership in key industrial and emerging technologies.
Why it matters: The high-speed steel production method targets efficiency and cost reduction in a foundational industry, signaling an effort to enhance China’s heavy industrial competitiveness. Meanwhile, the AI satellite advancements directly address surveillance and data processing needs, impacting both strategic military advantages and commercial remote sensing markets.
For Western readers: Western companies in heavy industries and satellite technology should assess the potential for increased competition from China, particularly in areas where these new capabilities can lead to lower costs or enhanced data services.
🔺 The Triangle
Where US, Japan, and China technology interests intersect
US technology curbs are accelerating China’s domestic hardware self-reliance while forcing Taiwan to speed up next-generation semiconductor manufacturing.
Semiconductors & Hardware
Counterpoint: Component Cost Dynamics Amid Memory Price Spikes
📊 Featured Chart
Source: Counterpoint Technology Market Research
Smartphone memory prices surged over 80% quarter-on-quarter in 2Q 2026, profoundly reshaping smartphone bill of materials (BOM) costs across all segments. This increase means memory is now the most expensive single component in premium smartphones, surpassing the SoC. Chinese and other Asian smartphone OEMs are adjusting product portfolios and pricing strategies to manage compressed margins and slowing technology adoption.
Why it matters: The dramatic increase in memory costs changes the underlying economics for smartphone manufacturers, particularly in East Asia, where high-volume, thin-margin segments are prevalent. OEMs will have to choose between absorbing costs, raising prices, or cutting features, which could slow innovation in camera and display technologies and shift consumer purchasing power towards lower-capacity devices.
For Western readers: If you are a Western-based component supplier to East Asian smartphone OEMs, expect increased pressure on your pricing and potential shifts in product mix from your customers. The cost of premium smartphone components, like the 1TB iPhone 18 Pro Max’s estimated $300 BOM increase, will inevitably translate to higher retail prices globally.
Semiconductors & Hardware
SEMICON Taiwan 2026 Spotlights Breakthroughs in Design, Memory, and Optical Interconnect
SEMICON Taiwan 2026 will focus on system-level integration of chips, memory, and optical interconnects to address AI power constraints. The event highlights growing demand for ASICs, with TrendForce projecting AI server shipments to increase over 28% in 2026, and ASIC-based systems reaching 27.8% of shipments.
Why it matters: The shift towards co-designing compute, memory, and interconnects as a single system is not just an engineering optimization; it’s a recalibration of where value accumulates in the AI hardware stack. Taiwan’s semiconductor ecosystem is positioning itself to be the critical enabler for this next phase of AI hardware development, beyond just chip fabrication.
For Western readers: Western AI and cloud service providers should recognize that future performance gains will increasingly depend on specialized, integrated system designs, not just raw compute power, pushing them deeper into co-design partnerships with East Asian suppliers for optimal efficiency.
Semiconductors & Hardware
2026 RF Market Slowdown: China’s Domestic Expansion Amid Global Shift to Defense, Auto, and 6G
📊 Featured Chart
Source: Yole, 2026
The RF device market faces a short-term slowdown in 2026 due to declining smartphone shipments, particularly impacting Japan and South Korea which are concentrated in affected segments. Meanwhile, Chinese suppliers Maxscend, Vanchip, Smarter Micro, and a re-emerging HiSilicon are rapidly expanding to build out a domestic RF supply chain, pushing for self-sufficiency in a consolidating global landscape.
Why it matters: The shift in RF market drivers from consumer smartphones to defense, automotive, and 6G infrastructure will reshape component supply chains. China’s focused investment in domestic RF suppliers like HiSilicon, even in its weakest consumer segment, illustrates a long-term strategic play for foundational technology control, mirroring its broader industrial policy objectives.
For Western readers: Western companies in defense and automotive RF should anticipate intensified competition and potentially divergent technical standards as China’s domestic ecosystem matures, especially for GaN-based power devices and 6G-related massive MIMO components.
Semiconductors & Hardware
TSMC 1.4nm Fab Ahead of Schedule; HP, Asus, Acer Adopting CXMT DRAM
TSMC is accelerating the construction of its 1.4nm fabrication facility in Taiwan, with trial production potentially beginning in Q3 2027. Concurrently, HP, Asus, and Acer have started integrating CXMT DRAM into their notebooks, although these products are not intended for the US market due to CXMT’s presence on the Pentagon’s Chinese Military Companies list.
Why it matters: TSMC’s ability to pull in its 1.4nm schedule reinforces its dominant position in advanced nodes, potentially widening the gap with competitors. The increasing use of CXMT DRAM by global PC players, even if restricted from the US market, indicates a growing acceptance of Chinese memory in other regions, driven by supply chain diversification or cost, chipping away at the market share of established players like Samsung and Micron.
For Western readers: Western hardware manufacturers should anticipate increased availability of Chinese-made DRAM for non-US markets, potentially offering cost advantages but also requiring careful navigation of evolving geopolitical restrictions and supply chain optics.
Policy & Regulation
US Considers Ban on Chinese Data Center Components Amid Broader Tech Curbs
The US is reportedly contemplating a ban on Chinese data center components, following recent FCC restrictions on other high-tech Chinese hardware. This move signals a tightening of American policy against Chinese technology imports.
Why it matters: Washington is tightening the screws on China’s access to the entire stack of enterprise computing, not just leading-edge semiconductors. This isn’t about data privacy alone; it’s a structural push to disentangle critical infrastructure from Chinese suppliers, forcing a bifurcated global tech supply chain.
For Western readers: Western data center operators and component suppliers should prepare for increased regulatory scrutiny on their supply chains and potential new compliance costs when procuring hardware for the US market, especially if any components originate from China.
🧩 Pattern This Issue
- Korea/Taiwan: AI boom pushes export values past Japan’s industrial shipping totals
- China: Tech firms shift focus to integrated supernode hardware infrastructure
- Japan: Sakura Internet deploys PFN’s domestic PLaMo 3.0 model on domestic cloud
East Asia’s AI race is shifting rapidly from software competition to hardware and infrastructure dominance, challenging the assumption that Western software leadership will guarantee economic supremacy in the region.
AsiaAI.FYI ·
Written by Dick Weisinger ·
Subscribe