China’s AI push advances on all fronts - Asian Tech Roundup

Plus: Trouble at the top at TSMC

Image:
China’s AI push advances on all fronts - Asian Tech Roundup

Welcome to Computing's weekly roundup of tech news in Asia. This time we look at AI news coming out of China suggesting progress in models, hardware and self-sufficiency, fallout from the cyberattack on Japanese brewer Asahi and trouble at the top at Taiwan’s TSMC.

The release of DeepSeek’s “open” R1 LLM back in January sent a shockwave through the markets. Could it be that all this money shovelled manically into the US-dominated AI firebox was being burned for nothing? Was the runaway train about to go off the rails? Subsequent weeks bought questions concerning the provenance of DeepSeek’s model, whether it had really been trained on homegrown hardware and the usual security issues with Chinese tech; the jitters subsided and markets resumed their upwards trend.

In this febrile atmosphere, all AI announcements should be taken with a pinch of salt. But with that important caveat, this week brings a number of stories that demonstrate China’s determination to dominate the AI space on its own terms, no matter what sanctions and hurdles may be placed in its way.

First is the news that China has overtaken the US in the global market for “open” AI models, with Chinese LLMs accounting for 17% of downloads, surpassing the 15.8% share from the likes of Meta, Google and OpenAI. This shift is significant because open models foster rapid global adoption and companies like DeepSeek and Alibaba Cloud are releasing models more frequently and with more variations than American rivals. Beijing has encouraged Chinese companies to adopt an open-source approach to AI, which contrasts with the closed, proprietary models favoured by most major US tech companies.

Next we have advances in AI hardware. China's top memory chipmaker CXMT has emerged as a rival to Korean giants Samsung and SK Hynix and the US’s Micron in the advanced DDR5 market which serves high-end AI computing servers and stacks. Meanwhile, AI chip start-up Zhonghao Xinying, whose founder once worked for Google, has emerged as a home-grown alternative to Nvidia, claiming its new TPU delivers up to 1.5 times the compute performance of Nvidia's A100 tensor core GPU while cutting energy consumption by 30%. At the same time, the Chinese government is reportedly banning ByteDance from using Nvidia GPUs in its datacentre. This news comes as Meta is eschewing Nvidia in favour of Google’s AI chips, a worrying trend for AI’s star performer.

China's push for AI self-sufficiency and ultimate dominance is heavily supported by government with massive investment in R&D and supply chains. And with Chinese companies also forming numerous strategic partnerships and making acquisitions to strengthen their position in the AI market, advances are happening on multiple fronts.

China

India

Japan

South Korea

Taiwan

Elsewhere in Asia