East Asian Technology Intelligence
Japan & China tech news — translated, contextualized, and delivered to your inbox.
Free. Unsubscribe anytime.
3 Takeaways This Issue
- By partnering with Japanese telecom giant Rakuten to broker its defense contracts, German drone unicorn Helsing is bypassing traditional *keiretsu* trading houses to fast-track its entry into Japan’s newly expanded military procurement budget.
- Google’s directive to shift 100% of Pixel hardware manufacturing out of China by 2027 establishes a blueprint for Western tech firms seeking to fully immunize their device supply chains against intensifying Beijing-Washington export controls.
- Unitree’s upcoming $905 million Shanghai IPO and the debut of its “Superman” R1 robot demonstrate that Chinese industrial policy is successfully shifting capital from consumer software toward high-yield, hardware-heavy humanoid robotics.
Core Move
Unitree Set for $905M Shanghai IPO as Humanoid Robotics Demand Surges
📊 Featured Chart
Source: Unitree product catalog pricing
Unitree has a projected Shanghai IPO value of $38 billion. This huge value does not just show off its new 28 mph “Superman” humanoid. Instead, it validates China’s state-guided plan to make embodied AI hardware cheap before the West can secure its own supply chains. Western analysts debate if Unitree’s computers match the brains of Boston Dynamics or Figure. Meanwhile, Chinese planners are running their classic consumer-electronics playbook. They want to scale manufacturing fast, drive unit costs to the floor, and lock in the global parts network. Unitree has already shipped 18,000 units. This volume turns humanoid robotics from a high-concept research expense into a repeatable factory floor discipline.
This domestic frenzy shows in an IPO oversubscribed by 8,000 times. It reveals a big split in how Beijing and New York view the future of robotics. Western venture capital treats humanoids as software-first challenges wrapped in silicon and carbon fiber. They are waiting for a “GPT-3 moment” for physical movement. In contrast, Chinese industrial policy treats them as advanced CNC machinery. This is a physical manufacturing challenge where winning depends on casting, servo motors, and local gear-reducer supply chains. China is applying its lithium-battery playbook to kinetic machinery. The goal is to control the physical platform. Whoever eventually writes the best AI model must still run it on Chinese-made chassis.
Yet, the huge speculative retail demand driving Unitree’s grey-market trading signals a shaky structural risk. Many assume this IPO capital will quickly turn into advanced manufacturing capacity. That view overlooks the tight bottleneck of high-precision component yields. Unitree risks burning through billions in capital on speculative stock values. This could happen long before Chinese Tier-1 suppliers can reliably mass-produce the high-torque cycloidal reducers needed for humanoid joints.
To see if this momentum will last or if it is a speculative bubble, track three specific signs over the next two quarters. First, watch if Unitree can push its shipping volume past the 25,000-unit mark. Second, track the post-IPO lockup performance of major backers Tencent and Alibaba. Third, see if domestic parts suppliers like Leaderdrive can secure exclusive, high-volume supply deals with Unitree to shut out foreign robotics rivals.
🗾 Japan Radar
What Japanese media is reporting that Western outlets miss
Japan is leveraging foreign partnerships to accelerate its transition from software-only AI to hardware-integrated defense, robotics, and energy systems.
🗾 Enterprise & Cloud
NVIDIA to Guarantee Up to $105 Billion in Debt for OpenAI’s Ohio AI Data Center, Partnering with SB Energy
📊 Featured Chart
Source: SB Energy & SEC filings
NVIDIA has announced it will serve as the exclusive AI computing provider for the massive 8GW PORTS-Pike Technology Campus in Ohio, where OpenAI will be the anchor tenant. Under a residual value guarantee agreement filed with the SEC, NVIDIA will assume up to $105 billion in debt liability if OpenAI defaults on its 20-year lease with builder and operator SB Energy. To support the project, SoftBank Group and SB Energy will construct over 10GW of new power generation and invest $4.2 billion in local grid infrastructure.
Why it matters: NVIDIA is moving beyond chip design to act as a colossal financial guarantor for the physical buildout of AI infrastructure, effectively absorbing the default risk of its largest customer. SoftBank Group is leveraging its US energy subsidiary, SB Energy, to secure a massive 20-year hardware and services lock-in for its key investments, demonstrating that the AI race has shifted entirely to securing gigawatt-scale power and physical land.
For Western readers: Do not assume OpenAI’s compute scaling will be constrained by its startup balance sheet; NVIDIA’s $105 billion safety net guarantees that capital markets will fund this buildout regardless of OpenAI’s near-term profitability.
🗾 Enterprise & Cloud
Cursor Launches Initial Beta of ‘Origin’ Code Hosting Service, Supporting Bidirectional Synchronization with GitHub
Anysphere, the developer of the AI-powered coding tool Cursor, has launched the initial beta of its new code hosting service called ‘Origin’ for all paid plan users. The service offers repositories, pull requests, code browsing, and real-time bidirectional synchronization with GitHub. This release follows the completion of SpaceX’s acquisition of Anysphere on August 14, 2026.
Why it matters: By hosting repositories directly on Origin, Anysphere is building the infrastructure required for autonomous AI agents to write, test, and deploy code without human intervention. This shifts the developer environment from human-first tools with AI assistants to an agent-native ecosystem where the AI is the primary operator.
For Western readers: Do not treat Cursor as just a popular editor extension; plan for it to aggressively decouple from GitHub’s infrastructure and establish its own end-to-end development pipeline backed by SpaceX’s capital.
Robotics & Automation
China’s Unitree Unveils ‘Superman’ Humanoid Robot Ahead of Shanghai Industry Exhibition
Chinese robotics manufacturer Unitree Robotics showcased its new R1 humanoid robot, nicknamed ‘Superman,’ during the Embodied Intelligent Robot Industry Exhibition in Shanghai. The debut highlights China’s aggressive push to commercialize bipedal robots by leveraging its domestic component supply chains and low-cost manufacturing capabilities. This launch adds momentum to Shanghai’s emerging role as a central hub for industrial and humanoid robotics development.
Why it matters: While US robotics firms focus heavily on advanced software and AI integration, Chinese players are ruthlessly optimizing the hardware bill of materials. By keeping manufacturing local and relying on established domestic supply chains for actuators and sensors, Unitree is positioning itself to flood the logistics and manufacturing sectors with affordable, functional humanoids long before Western competitors can scale down their production costs.
For Western readers: If you operate in warehouse logistics or light assembly, expect Chinese humanoid hardware to reach price parity with traditional automated guided vehicles within 24 months, forcing a choice between cheaper Chinese hardware platforms or high-cost, software-superior Western alternatives.
AI & Machine Learning
German Drone Maker Helsing Enlists Rakuten to Broker Japan Military Deal
German defense tech unicorn Helsing has partnered with Japanese e-commerce and telecommunications giant Rakuten to pitch its AI-enabled military drones to Japan’s Self-Defense Forces. The partnership leverages Rakuten’s local political connections and communication network infrastructure to position Helsing’s HX-2 autonomous strike drones for future procurement contracts. The move represents an unusual entry strategy into Japan’s highly protected defense market, which has traditionally been dominated by domestic heavy industry conglomerates.
Why it matters: By bypassing traditional defense contractors like Mitsubishi Heavy Industries, Helsing and Rakuten are challenging the entrenched keiretsu system in Japanese military procurement. If Rakuten successfully leverages its cellular network infrastructure and political capital to broker a major defense contract, it establishes a new blueprint for foreign dual-use technology firms to enter the Japanese market without giving up their core intellectual property to local industrial giants.
For Western readers: Western dual-use software and AI startups looking to sell to Tokyo should stop targeting traditional Japanese defense primes as their primary entry partners, and instead pitch consumer tech and telecom conglomerates eager to diversify into national security.
Semiconductors & Hardware
Google plans to move all Pixel production out of China by 2027
Google has instructed suppliers to transition the manufacturing of all Pixel smartphones, smartwatches, and wireless earbuds entirely outside of China by next year. The US tech giant is redirecting its hardware production footprint to expanding hubs in Vietnam and India to mitigate geopolitical risks.
Why it matters: The complete migration of Google’s hardware ecosystem proves that supply chain diversification is moving past simple hedging into full operational decoupling for high-profile US electronics. Vietnam and India stand to gain massive manufacturing capacity and technical expertise, while Chinese contract manufacturers lose premium assembly volumes.
For Western readers: If you design or retail consumer hardware, expect the cost structure of non-China assembly to dictate pricing and margin expectations as ecosystem players compete for limited high-quality capacity in India and Vietnam.
🇨🇳 China Watch
China’s technology moves, framed for Western readers
US sanctions are forcing China to abandon monolithic frontier models, pivoting instead to decentralized edge-compute and asymmetric military-tech.
Policy & Regulation
US Forces Nations to Choose Between Washington’s Pax Silica and China’s AI Bloc
The U.S. State Department is issuing an ultimatum to 35 nations that signed the U.S. AI Opportunity Statement, demanding they align with either the Washington-led tech ecosystem or China’s rival bloc. The directive specifically targets countries attempting to dual-track their alignment, such as those participating in both the U.S. ‘Pax Silica’ initiative and Beijing’s newly launched World Artificial Intelligence Cooperation Organization.
Why it matters: This ultimatum dismantles the middle-ground strategy many Asian and Eurasian countries used to hedge their technology bets, effectively slicing the global AI market into two mutually exclusive supply chains. For China, this accelerates their push to build a self-reliant domestic stack, while Japan and South Korea face immediate pressure to audit their joint research and development programs for any Chinese exposure.
For Western readers: Western tech firms must immediately audit their joint ventures and research partnerships in secondary markets like Central Asia, Southeast Asia, and the Middle East, as any co-development with entities linked to China’s AI organization will trigger exclusion from U.S.-aligned supply chains.
AI & Machine Learning2 STORIES
China’s AI Giants Pivot to Open-Source Agent Ecosystems and Local Compute
Chinese AI pioneers DeepSeek and Alibaba are shifting strategic focus away from monolithic proprietary models toward highly optimized, open-source agentic infrastructure. DeepSeek is integrating with the Cordis framework to build a modular plugin platform for autonomous agents, while Alibaba has released Qwen3.8-27B, a lightweight model designed to run complex coding and agent tasks locally on consumer-grade hardware.
Why it matters: In the East Asian business landscape, this shift enables local enterprises to bypass expensive cloud infrastructure and geopolitical chip restrictions, fostering a highly resilient domestic ecosystem of low-cost, decentralized AI applications.
For Western readers: Western developers and tech leaders must abandon the assumption that cutting-edge agentic workflows require massive, proprietary US cloud APIs; instead, they should actively evaluate China’s highly efficient, open-source local models to drastically reduce their own operational costs.
AI & Machine Learning
Chinese Researchers Develop AI System to Track F-22 and F-35 Heat Signatures
Chinese researchers have developed a lightweight AI system designed to identify and track the heat signatures of F-22 and F-35 stealth fighters. The system analyzes thermal data from aircraft engines and exhaust, achieving over 90 percent accuracy in simulated laboratory environments. Crucially, the model claims to successfully distinguish between actual aircraft heat signatures and defensive flare emissions.
Why it matters: If this AI model can be successfully compressed to run on edge hardware inside Chinese fighter jets or missile defense systems, it compromises the core tactical advantage of American stealth platforms. Shifting the detection battleground from radar cross-section reduction to thermal management forces a costly redesign of Western propulsion and cooling architectures.
For Western readers: Western defense contractors and military planners must assume that existing passive thermal suppression systems on the F-22 and F-35 are no longer sufficient to guarantee low observability against Chinese near-peer adversaries.
Enterprise & Cloud
Nvidia to provide up to US$105 billion guarantee for OpenAI’s Ohio data centre
📊 Featured Chart
Figures represent committed financing and capital investments as of August 2026.
Nvidia has committed to a massive US$105 billion guarantee to secure a 20-year lease for an OpenAI data center in Ohio. The facility is being developed by SB Energy, a renewable energy developer backed by Japan’s SoftBank Group. As part of the arrangement, Nvidia will also invest US$1.5 billion directly into SB Energy, building on previous infrastructure funding from SoftBank and OpenAI.
Why it matters: SoftBank is successfully leveraging its energy portfolio to secure a central role in the US AI infrastructure buildout, moving beyond mere venture investments into hard physical assets that lock in multi-decade revenue. By tying its energy assets directly to Nvidia’s hardware allocation and OpenAI’s computing needs, SoftBank guarantees long-term demand for its power generation while bypassing the typical volatility of public tech stocks.
For Western readers: Western tech buyers must recognize that access to leading-edge AI compute is no longer just about chip allocation, but about who controls the gigawatt-scale power generation assets backed by deep-pocketed Asian conglomerates.
🔺 The Triangle
Where US, Japan, and China technology interests intersect
Washington’s containment strategy is forcing China to bypass cutting-edge lithography by mastering domestic chiplets, shadow supply chains, and software workarounds.
Semiconductors & Hardware
China’s domestic AI stack gains momentum despite hardware gap
📊 Featured Chart
Source: Maybank Investment Bank report
Chinese tech giants are successfully training and running large-scale AI models on entirely domestic hardware clusters, bypassing severe Nvidia shortages and Beijing’s bans on foreign silicon. Notably, Meituan trained its trillion-parameter LongCat-2.0 model on a 50,000-chip domestic cluster, while Baidu’s Ernie-5.1 achieved a 97 percent effective training rate on domestic Kunlunxin chips.
Why it matters: The real bottleneck for China’s AI progression is shifting from hardware availability to the software abstraction layer. Companies bridging the software integration gap—like those deploying unified compilers to run workloads across disparate domestic GPU architectures—will capture the immediate domestic market value long before local foundries achieve physical transistor parity with Western fabricators.
For Western readers: Western enterprises assuming Chinese AI is structurally bottlenecked by hardware limits are underestimating the rapid viability of domestic workarounds and software-layer translation tools that bypass Nvidia’s CUDA ecosystem entirely.
Semiconductors & Hardware
Samsung Foundry Delays 1.4nm Process to 2029, Shuns High-NA EUV for Next Node
Samsung Electronics has postponed the introduction of its 1.4nm manufacturing process to 2029, electing to prolong the lifecycle of its 2nm process for three years. In a significant shift in lithography strategy, the company also confirmed it will not utilize ASML’s expensive High-NA EUV machines for the 1.4nm node due to cost and performance limitations.
Why it matters: TSMC secures an uncontested window of dominance in the sub-2nm arena for at least an extra two years, reducing the pricing leverage that chip designers like Nvidia, AMD, and Apple hoped to gain by playing the two dominant East Asian foundries against each other. Samsung’s public rejection of High-NA EUV for the 1.4nm generation validates concerns that the technology’s massive capital expenditure does not yield justifiable density or performance gains for current chip designs.
For Western readers: Western fabless designers must adjust their long-term product roadmaps to rely almost exclusively on TSMC for leading-edge silicon through the late 2020s, while assuming that early High-NA EUV adoption will remain a costly, low-yield niche rather than an industry-wide standard.
Semiconductors & Hardware
Silicon Motion SSD Reference Design Kit Targeted at AI Infra
Taiwanese chipmaker Silicon Motion has launched its MonTitan SSD Reference Design Kit featuring its PerformaShape technology to optimize enterprise SSDs for agentic AI workloads. Designed for both PCIe 5.0 and PCIe 6.0 architectures, the platform enables storage hardware to act as a persistent memory layer capable of handling heavy key-value cache offloading and multi-tenant AI operations.
Why it matters: The real battle in AI infrastructure is shifting from raw compute to memory bandwidth and storage endurance. Silicon Motion’s design kit allows second-tier SSD manufacturers to rapidly field enterprise-grade hardware capable of surviving the constant write cycles demanded by autonomous AI agents, bypassing the long R&D cycles usually dominated by Samsung or Solidigm.
For Western readers: Do not assume that Korean giants will maintain an exclusive monopoly on AI-optimized enterprise storage; Taiwan’s merchant silicon ecosystem is rapidly commoditizing the specialized controller technology needed for next-generation agentic AI data centers.
Semiconductors & Hardware
Chinese Chiplet Interconnect Specialist Kiwimoore Targets $2B Hong Kong IPO
📊 Featured Chart
Source: Electronics Weekly reporting
Five-year-old Chinese chiplet packaging and AI networking startup Kiwimoore is planning a Hong Kong stock exchange listing for 2027, targeting a $2 billion valuation. The company specializes in critical AI networking interconnect technology, including GPU links and network cards for high-performance computing clusters. CITIC Securities and ABC International are expected to lead the offering, which follows foundry Nexchip’s $880 million Hong Kong listing.
Why it matters: Beijing is actively funding and fast-tracking domestic chiplet packaging startups to bypass physical lithography limits imposed by Western sanctions. By listing in Hong Kong, Kiwimoore secures international liquidity while serving as a critical hardware supplier for domestic computing clusters built on mainland-designed GPUs.
For Western readers: Western hardware architects should assume that China’s AI datacenter builds will increasingly bypass monolithic Nvidia accelerators in favor of domestic multi-chiplet modules tied together by Kiwimoore’s proprietary interconnects, making Chinese AI infrastructure highly specialized and decoupled from global standards.
Policy & Regulation
White House Report Names Nine ASEAN Nations in Global ‘Shadow Transshipment Network’ Evading China Tariffs
A White House transshipment report identifies nine of the ten ASEAN member states as key conduits in a global network helping Chinese-origin goods bypass U.S. tariffs. While none of these Southeast Asian nations are classified under the highest-risk Tier 1 category, Vietnam, Thailand, Malaysia, and Indonesia are labeled as Tier 2 risks due to their deep economic integration with Chinese manufacturing pipelines.
Why it matters: Southeast Asian nations are acting as functional trade shields for Chinese industrial capacity rather than truly independent manufacturing alternatives. U.S. tariff enforcement is increasingly targeting these intermediary corridors—specifically the Vietnam-Mexico-India hubs which collectively leaked $28 billion in tariff revenue in 2025—meaning Chinese tech firms using ASEAN assembly plants to bypass trade barriers will face escalating customs scrutiny and potential retroactive penalties.
For Western readers: If you rely on Southeast Asian assembly for electronics, machinery, or components, audit your suppliers’ bill of materials immediately to identify Chinese-origin inputs before impending U.S. customs enforcement actions disrupt your logistics.
🧩 Pattern This Issue
- Japan: SB Energy partners with Nvidia to secure $105B in debt for OpenAI’s Ohio data center
- China: Unitree aims for a $905M Shanghai IPO amid surging demand for humanoid robotics
- China: DeepSeek and Alibaba pivot to open-source agent ecosystems using local compute
East Asian capital and technology are shifting from general software development to heavy physical and localized infrastructure, meaning Western firms relying purely on cloud-based AI models will lose ground to competitors who control the physical hardware and localized computing layer.
AsiaAI.FYI ·
Written by Dick Weisinger ·
Subscribe