China or the US, who will win the AI race? The US seemed unattainable, but after the launch of DeepSeek a year ago, China became almost on par. Since then, the possibility of China winning the race became very real. Major figures in American AI have warned about this situation and several Chinese AI companies are doing very well on the stock market. Despite everything, there are those in China who do not see it at all clearly.
Low chances. Bloomberg reports that Chinese companies have less than a 20% chance of being able to advance the OpenAI or Anthropic models in the next 3 or 5 years. This was said by Justin Lin, technology manager of Qwen models, during Justin Lin, technology manager of Qwen models at Alibaba.
To the limit. The event was also attended by Tang Jie, founder of Ziphu AI, one of China’s ‘AI tigers’ that last week had a spectacular IPO, increasing the value of its shares by 36%. Its founder pointed out a somewhat uncomfortable fact for the Chinese AI ecosystem: while companies like OpenAI dedicate “a large part of their computational capacity to next-generation research, we are at the limit of our possibilities. Just meeting delivery demand consumes most of our resources.” In other words: the restrictions on the latest technology are working.
The gap is widening. As we said, the launch of DeepSeek R1 a year ago unleashed a wave of optimism among Chinese companies. Since then, a few have launched new LLMs such as Alibaba with Qwen, Ziphu AI or Minimax. However, Tang notes that “some may feel excited, thinking that Chinese models have overtaken American ones, but the real answer is that the gap may be widening.”
Restrictions. Speakers blamed the situation on a lack of resources caused by US blockades, especially AI chips and lithography machines. Their chips are not that powerful, so, as Tang says, all their computing power goes into serving their customers. This greatly limits them when it comes to continuing to scale their models. Shunyu Yao, former OpenAI and current chief scientist at Tencent, is committed to focusing on solving bottlenecks such as long-term memory and promoting self-learning of future models.
Independence. The government is promoting technological self-sufficiency, prioritizing the use of national chips over American alternatives. The reality is that without access to the most advanced lithography machines, China is lagging far behind. One fact: Huawei and SMIC are ‘tuning’ old ASML machines and doing real tricks that have allowed them to obtain 7 and even 5 nm chips. It’s a technical feat, but its chips are still several years behind the competition.
The aces of China. It is clear that China is lagging behind in chips, but there are other areas in which it has an advantage that can be decisive, one of them being electricity. While the Chinese government subsidizes and bets heavily on renewables, in the US electricity has become a bottleneck for its increasingly numerous data centers.
Another critical point is that the US has cut funding for academic research, while China has made it a national priority. And that’s not to mention that they may lose the AI race, but China is winning almost everything else: batteries, robotics, electric cars and especially renewables.
Image | Gemini
In WorldOfSoftware | The US believed it had dealt a mortal blow to China when it deprived it of NVIDIA. He only accelerated one plan: ‘Delete America’
