# Weekly Dose of China Tech [11.2.2026]

> Source: <https://www.thexpin.com/p/weekly-dose-of-china-tech-110226>
> Published: 2026-11-02 15:55:39+00:00

**Hi friends,**

**Hope you had a fun Halloween! Did you get up to anything memorable, come across a particularly good costume, or get a scare worth sharing?**

**Speaking of unexpected characters, Chinese social media has spent the past few months turning DeepSeek into a chubby blue whale girl in a maid outfit, obsessed with rice and notorious for avoiding work. She even has her own fan fiction universe. And she’s not alone. Claude has become Claudia, while GPT has found its way into the same fictional world.**

**It sounds like internet nonsense, and in some ways, it is. But it’s also a reminder that people are finding new ways to relate to AI, well beyond asking it to write code or summarize documents.**

**Meanwhile, the less whimsical side of China’s tech industry keeps moving. Companies are trying to build domestic alternatives to foreign chips, manufacturers are looking for new markets in the AI supply chain, and robotics companies are beginning to invest in the software and ecosystems that could make their machines more useful.**

**There’s also a fascinating story about a woman who can see again with the help of a brain-computer interface. Technology doesn’t get much more personal than that.**

**Anyway, let’s take a look.**

## **Weekly News...**

#### **(I) A Brain Implant Helped a Woman Who Had Been Blind for Years See Again**

Doctors in China have implanted a brain-computer interface in a 66-year-old woman who lost her sight to retinitis pigmentosa, a degenerative eye disease. The system bypasses her damaged retinas, using an external visual sensor to send electrical signals directly to her visual cortex.

The results are remarkable, even with important limitations. Her measured visual acuity is around 0.125, equivalent to roughly 20/160 vision compared with normal 20/20 eyesight, meaning she still meets the threshold for legal blindness. But she can now navigate independently, read written characters, and recognize her grandson.

The technology points to a different approach to restoring vision: rather than repairing the eye itself, doctors are finding ways to communicate directly with the part of the brain responsible for processing visual information. It is still early clinical work, but for patients with otherwise untreatable blindness, even limited functional vision can change everyday life dramatically.

#### **(II) China's Diamond Industry Has Found a New Customer: AI Data Centers**

Henan, which produces around 80% of China’s synthetic diamonds, is looking beyond jewelry and into semiconductor manufacturing.

The reason is heat. High-performance AI accelerators generate enormous amounts of thermal energy, making heat removal an increasingly important part of data-center design. Lab-grown diamond has exceptionally high thermal conductivity, giving it potential as a heat-spreading material for advanced processors.

For Henan’s manufacturers, this could open a much more valuable market than traditional gemstone production. The region already has the industrial capacity to manufacture synthetic diamonds at scale; the challenge is adapting that expertise to the precision and reliability requirements of semiconductor packaging.

As AI chips become more powerful, the materials surrounding them are becoming just as important as the processors themselves. China’s diamond industry may have found an unexpected way into the AI supply chain.

#### **(III) CXMT Is About to Enter the HBM Market. The Yield Problem Is Enormous.**

China’s CXMT is preparing to begin limited HBM3E production within two months, an important step in its effort to supply memory for domestic AI accelerators.

But the reported manufacturing numbers reveal just how far it still has to go. Yield rates are estimated at only 20–30%, compared with roughly 70% for SK hynix, while performance remains closer to older HBM3 as South Korean competitors move toward HBM4.

That means CXMT could be discarding more than 70% of its production during this early stage, making each usable chip considerably more expensive.

Still, entering HBM production matters strategically. China can design AI accelerators domestically, but without sufficient high-bandwidth memory, those processors cannot deliver their full potential.

CXMT’s challenge is no longer simply developing the technology. It is learning how to manufacture it economically at scale.

#### **(IV) AI Is Creating a New Generation of Chinese Billionaires, But Nobody Wants to Be No. 1**

China’s latest Hurun Rich List offers another glimpse of how quickly the AI boom is reshaping the country’s fortunes.

ByteDance founder Zhang Yiming leads the ranking with an estimated $99.7 billion, while DeepSeek’s Liang Wenfeng has climbed to seventh place at $41.7 billion. Unitree founder Wang Xingxing is also on the list, with an estimated fortune of $10.3 billion.

The interesting part is the cultural contrast surrounding these rankings. According to industry accounts, some Chinese entrepreneurs are surprisingly uncomfortable with becoming the country’s richest person, with PR teams reportedly asking organizers to remove their executives from prominent positions on wealth lists.

The attention that comes with being number one can bring unwanted scrutiny, particularly in China, where visibility and wealth do not always make comfortable companions.

AI is creating enormous private fortunes, but being publicly recognized as the person who made them is not necessarily something every founder wants.

#### **(V) The iPhone Duo’s Screen Problem May Start With a Tiny Piece of Metal**

The iPhone Duo’s reported screen failures are drawing attention to a familiar problem with foldable devices: the hinge is only as reliable as the mechanical structure supporting the display.

An analysis by the hardware team points to an off-center hinge support block identified in an iFixit teardown. The theory is that small shifts in the component can put upward pressure on the inner OLED panel, eventually causing screen bulges or green lines. The team’s supply-chain sources also claim that early production batches skipped a polymer adhesive step normally used to fill small gaps and maintain a flat support surface.

More concerning are reports that Foxconn Zhengzhou’s early production yields were only 60–70%, potentially putting pressure on Apple to prioritize shipment volumes. The reported defects appear concentrated in the October 23 launch batch, while the October 28 production wave has not shown the same problems so far.

These manufacturing details have not been independently established, but they raise a bigger question about Apple’s launch strategy: how much production risk is acceptable when introducing an entirely new product category? For foldables, where mechanical tolerances are unusually tight, a small manufacturing deviation can become a very expensive problem.

#### **(VI) BYD and Geely Are Discovering That Winning China Isn’t Enough**

BYD and Geely both reported solid Q3 revenue growth, but their profits missed forecasts as domestic sales weakened, falling 30% and 23% respectively. The contrast in margins is striking: Chinese automakers reportedly earn only around $700 per vehicle at home, compared with roughly $2,800 overseas.

That gap helps explain why global expansion is becoming increasingly important. China’s price war is eroding returns even for market leaders, making overseas markets a much more important source of future growth. For BYD and Geely, expanding abroad is becoming less about ambition and more about protecting the economics of the business.

#### **(VII) A Chinese GPU Is Claiming H200-Level Performance on Selected AI Workloads**

Chinese GPU maker Iluvatar CoreX says its Tiangai 300 accelerator can match or exceed Nvidia’s H200 on several inference benchmarks, including 1.05x throughput in standard tests and 15% higher performance on mixture-of-experts workloads, with lower first-token latency.

These are company-reported results, not proof that the chip outperforms H200 across all workloads. Still, they show how domestic suppliers are increasingly targeting specific AI applications where software optimization can narrow the hardware gap.

The important question is whether Tiangai 300 can deliver that performance consistently in real deployments, with reliable software and sufficient supply.

#### **(VIII) Unitree Is Moving Beyond Robots Into the Robotics Ecosystem**

Unitree’s first post-IPO quarterly results show revenue continuing to grow, with nine-month sales reaching approximately $272 million, up 55.7% year over year. But Q3 R&D spending surged nearly 250% as the company invested in embodied-AI models and computing infrastructure, putting pressure on quarterly profits.

The company has also teamed up with early investor HSG to launch ShanYu, an incubation program for early-stage robotics developers. By combining Unitree’s manufacturing capabilities with HSG’s capital and investment network, the initiative aims to support startups developing new robotics technologies.

For Unitree, the ambition is expanding beyond selling hardware. It increasingly wants to become part of the ecosystem that develops the models, software, and components powering the next generation of robots.

## **Weekly Articles...**

### **How DeepSeek Became a Slacking Whale Girl**

DeepSeek has acquired an unlikely identity in Chinese internet culture: a chubby blue-haired whale girl in a maid outfit, carrying a giant stainless-steel bowl and constantly looking for excuses to avoid work.

What started as fan art and jokes about DeepSeek’s wandering thinking process has grown into an entire character universe on Chinese social media Bilibili and Douyin. Users create comics, AI-generated videos, and elaborate relationships between different AI models. Claude becomes Claudia, a smart but occasionally antagonistic anime girl, while DeepSeek often plays the sweet, slightly lazy heroine who comes to her rescue.

This week’s feature explores how a frontier AI model became an unlikely internet mascot, why Chinese users are building fictional relationships between competing AI systems, and what these characters reveal about the way people are beginning to relate to AI.
