The Wrap | 10 – 16 Jan 2026
Asia Tech Lens’ weekly digest of the signals, shifts, and stories shaping Asia’s tech future
Editor’s Note: This week’s wrap reflects a region moving at different speeds. Some players are pushing to keep models training and products shipping despite tightening constraints, while others are focused on making AI agents useful in everyday routines, not just impressive in demos. Governments, meanwhile, are moving quickly when consumer-facing tools cross clear red lines. In the background, capital and infrastructure choices continue to shape who can scale, and which markets become the next hubs.
Podcast Spotlight | Kevin Choi, Mediwhale
How Mediwhale’s AI Sees Disease Before Doctors Can
After losing half his vision to glaucoma at 26, Kevin Choi, Co-Founder & CEO of Mediwhale, turned a personal crisis into purpose—building AI that can detect cardiovascular and metabolic diseases before doctors can, using a single retinal image.
From publishing in The Lancet Digital Health to raising US $22 million and convincing skeptical clinicians, Kevin’s story is one of resilience and clarity of mission. Hosted by Miro Lu, this episode explores the human side of AI in medicine—how science, empathy, and leadership intersect when tech is literally saving lives.
“Money is cold,” Kevin says. “But saving someone’s life—that’s real.”
Watch on YouTube | Listen on Spotify · Apple Podcasts
AI | China
China Blocks Nvidia H200 Imports
Chinese customs authorities have told customs agents that Nvidia’s H200 AI chips are not permitted to enter China, according to Reuters sources. The report also says Chinese officials summoned domestic tech firms to meetings and instructed them not to buy the chips unless necessary, with possible exceptions under discussion for research and university-linked projects. There are also reports from Nikkei Asia suggesting officials are drafting strict purchase rules to cap volumes and prioritize domestic alternatives.
The H200, Nvidia's second most powerful AI chip, is one of the biggest flashpoints in current U.S.-Sino relations. Beijing has not publicly explained the motive, and it is not clear whether this is a formal ban or a temporary measure.
Bottom Line: Analysts cited by Reuters suggest the move could serve multiple objectives: limiting dependence on US chips, buying space for domestic AI chipmakers, or creating leverage for US-China engagement.
AI | China
Zhipu Trains AI Model on Huawei Chips
Chinese AI startup Zhipu says it has released a new multimodal model that it claims is the first state-of-the-art system to be fully trained on domestic chips. The company says its open-source image generation model, GLM-Image, completed training using Huawei’s Ascend chips, running on Huawei’s Atlas 800T A2 server and MindSpore framework.
The announcement is being positioned as evidence that China can train high-performance generative models on a “full-stack” domestic platform, from chips to framework. That matters because a few have demonstrated full training pipelines on Huawei's Ascend stack at competitive performance.
Bottom Line: Zhipu demonstrates China can scale frontier AI despite U.S. chip curbs, using Huawei's full-stack with Ascend hardware optimizations—bolstering self-reliance claims and eroding Nvidia's pricing power even as export talks advance.
AI Agents | China
Alibaba Pushes Qwen into the Agentic Era
Alibaba unveiled major upgrades to its Qwen app that let users complete tasks inside the chat interface, such as ordering food delivery and booking travel. The features are now in public testing in China and plug directly into Alibaba’s ecosystem, including Taobao, Alipay, Fliggy, and Amap, so users can authorize and complete transactions without switching apps.
Alibaba also introduced a invite-only beta Task Assistant capable of phone calls to restaurants, processing document batches, and planning multi-stop itineraries. The update is part of Alibaba’s broader push into consumer-facing AI, an area where it has been playing catch-up with domestic rivals.
Bottom Line: Qwen now has a live sandbox to show whether agentic AI can move beyond demos; consistent, error-free execution on payments and bookings will determine if Alibaba can close the gap with domestic rivals.
AI | China
DeepSeek Pitches “Conditional Memory” to Power Larger AI Models
A new technical paper co-authored by DeepSeek researchers, including founder Liang Wenfeng, and Peking University researchers proposes a training technique aimed at scaling AI models while reducing dependence on GPU high-bandwidth memory (HBM). The approach, described as a “conditional memory” method called Engram, is designed to support more aggressive parameter expansion without being tightly constrained by GPU memory limits.
The paper lands as speculation builds around a major DeepSeek model release ahead of Lunar New Year, and reinforces the Hangzhou start-up’s positioning around cost-efficient progress despite China’s relative shortage of advanced compute compared with U.S. leaders.
Bottom Line: Engram could narrow the U.S. compute lead by enabling efficient scaling on memory-constrained hardware like China's available GPUs.
Quick Commerce | India
Blinkit Drops “10-Minute Delivery” Branding After Government Push
India’s quick commerce giant, Blinkit has removed “10-minute delivery” branding from its app and is expected to scrub fixed-time claims across ads and social media. The shift follows late-December gig worker strikes and follows government talks with major platforms over concerns that fixed delivery-time promises intensify rider pressure and pose safety risks.
The change does not necessarily mean deliveries will slow. It signals a move away from explicit time guarantees in public messaging as regulators sharpen focus on gig worker protections.
Bottom Line: India’s quick commerce players are keeping the speed model, but dropping the speed promise. The industry is being pushed to de-risk both rider safety and public optics.
AI | Asia
X Restricts Grok After ‘Undressing’ Deepfake Outcry
X says it has introduced new measures to stop Grok from generating sexualized “undressing” images of real people, including geo-blocking certain image prompts in jurisdictions where such content is illegal.
The move comes after severe backlash, with countries in Asia leaning on enforcement rather than rhetoric. Indonesia and Malaysia have temporarily blocked or restricted Grok over sexually explicit deepfakes, India has formally demanded tougher safeguards, and X’s Japan unit has warned users and pledged to remove illegal AI-generated images.
Bottom Line: Asia is emerging as the first real testbed for how hard governments are willing to push back on unsafe consumer AI.
AI & Semiconductors | UAE
UAE Joins U.S.-Led “Pax Silica” to Bolster Supply Chain
The United Arab Emirates has joined a US-led initiative called “Pax Silica” aimed at strengthening global AI and semiconductor supply chains. The program is framed as a “coalition of capabilities” and includes countries such as Australia, Britain, Israel, Japan, Qatar, Singapore, and South Korea, as Washington looks to deepen tech-aligned partnerships and reduce reliance on strategic rivals.
UAE’s role is being pitched around logistics, industrial capacity, and energy, alongside its ongoing buildout of AI infrastructure, including large-scale data center investment in Abu Dhabi using US technology.
Bottom Line: The Gulf is moving from AI buyer to AI enabler. By joining Pax Silica, the UAE and Qatar are monetizing what they uniquely have (power, logistics, capital) to become indispensable to the next phase of global compute buildout.
FinTech | APAC
KKR Closes $2.5B Asia Private Credit Fund
KKR, a global investment firm, has raised $2.5 billion for its second Asia-focused private credit strategy, combining $1.8 billion in its Asia Credit Opportunities Fund II and $700 million in separately managed accounts focused on similar investments, according to Reuters. The fund targets privately originated performing credit across Asia-Pacific.
The raise follows KKR’s first Asia Pacific private credit vehicle, which closed at $1.1 billion in 2022, and comes as investor interest in the region grows alongside improving exit conditions and rising attractiveness of markets like Japan and India. KKR also said it has completed more than 60 Asia-Pacific credit investments since 2019, representing $8.3 billion of invested capital and $27.5 billion in total transaction value.
Bottom Line: KKR’s $2.5B raise shows Asia private credit is becoming mainstream. Big investors are treating direct lending as a core asset class in the region.
Signals To Watch
China Blocks Nvidia H200 Imports
Any move from an informal border stop to a formal quota/licensing framework that defines who can import H200s and under what conditions.Zhipu Trains AI Model on Huawei Chips
Independent benchmarks showing whether Ascend-trained GLM-Image matches frontier performance at competitive cost.Alibaba Pushes Qwen into the Agentic Era
Whether Alibaba expands Qwen’s agent beyond its own services into third-party apps/sites, and if so, whether it triggers the same resistance ByteDance faced (blocks, throttling, or anti-bot/risk controls) once the agent operates across platforms.DeepSeek Pitches “Conditional Memory”
A new DeepSeek model release that demonstrates measurable training cost or memory-use reductions tied to Engram-like techniques.KKR Closes $2.5B Asia Performing Credit Fund
Where KKR deploys its first major cheques and at what terms, revealing which Asia markets are absorbing private credit fastest.
Takeaway
The leading questions are becoming operational: who can secure enough compute to keep iterating, who can translate models into reliable task completion, and who can ship features that survive platform risk controls and regulatory scrutiny. The result is a more execution-driven phase of competition, where advantage accrues to firms that can deliver consistent outcomes in messy real-world environments, not just benchmark wins.
Sources
Reuters: Exclusive: China’s customs agents told Nvidia’s H200 chips are not permitted, sources say
Bloomberg: China’s Zhipu Unveils New AI Model Trained on Huawei’s Chips
Yahoo Tech: Alibaba upgrades Qwen app to order food, book travel
SCMP: DeepSeek founder’s latest paper proposes new AI model training to bypass GPU limits
India Today: Blinkit drops 10-minute delivery branding as Centre steps in after gig strike
DigiTimes: Qatar, UAE join US-led Pax Silica supply chain pact
CNA: KKR secures $2.5 billion for second Asia private credit fund