Chinese Tech Leaders Temper Expectations in Global AI Race

China’s AI sector has been portrayed as rapidly closing in on the United States in the race to lead the global AI market, but today’s experts offer a more down-to-earth perspective on whether China will push ahead.

The nation’s new models, flashy demos, and bold government targets have fuelled the idea that a turning point is just around the corner. But although China’s AI sector has made rapid strides over the past decade, leading researchers inside the country say there are limits to that progress when measured against the US.

Some of China’s most prominent AI scientists are now giving a more measured view of what lies ahead. Speaking at a major AI conference in Beijing, researchers from Alibaba, Tencent, and leading start-ups said the chances of a Chinese company overtaking US AI giants such as OpenAI or Google DeepMind within the next three to five years are slim, less than 20% by their estimates. Even that, one speaker suggested, may be “highly optimistic”.

The remarks, delivered at the AGI-Next summit hosted by Tsinghua University, highlight a sobering reassessment of China’s near-term AI ambitions. While domestic models have improved quickly and gained international attention, speakers warned that the structural gap with the US remains difficult to overcome.

A widening gap in compute power

According to reporting from the South China Morning Post, the main issue stems from computing power. Lin Junyang, technical lead of Alibaba’s Qwen large language model team, said the most significant disadvantage lies in computational scale, as US AI labs operate with one to two orders of magnitude more computational resources than their Chinese counterparts. That difference allows American companies to train larger, more experimental models and pursue long-horizon research that is difficult to replicate under tighter constraints.

“Most critically, OpenAI and others are pouring massive computational resources into next-generation research,” Lin said. “Meanwhile, in China, we are stretched to the limit just from meeting daily demand, which already takes up the vast majority of our compute,” he added, leaving little headroom for long-term, high-risk research.

The semiconductor access situation only compounds the problem, since China still lacks domestically produced extreme ultraviolet (EUV) lithography machines needed to manufacture the most advanced semiconductors. 

US export controls have further tightened access to the most powerful, cutting-edge chips, even as Washington recently approved the sale of Nvidia’s H200 AI processors to China late last year. Complicating matters, Beijing has reportedly urged some companies to pause those purchases and substitute foreign chips with domestic alternatives.

Tang Jie, co-founder and chief AI scientist at Zhipu AI, warned that recent enthusiasm around Chinese open source models can give a misleading picture of the broader race. While Chinese developers have released many of their most capable systems publicly, US companies have largely kept their strongest models behind closed doors.

“The gap between China and the US may in fact be widening because the US has many models that they have not released to the public,” he said, noting that performance benchmarks only capture what is publicly available. In his view, American firms likely have more advanced internal systems, supported by far greater compute and capital.

Open source models and hidden competition

One area where China has gained strength is open-source AI. Chinese developers have overwhelmingly chosen to release their models openly, accelerating adoption across industries and markets. This strategy has helped Chinese systems gain visibility and practical traction, particularly in regions and sectors looking for alternatives to proprietary US platforms.

Still, speakers say that openness alone will not drive the nation to leadership. US firms have access to vast private compute clusters, deep capital reserves, and advanced chips, allowing them to run parallel research tracks that may never be disclosed publicly. As a result, China’s apparent gains may also be up against a deeper, less visible US lead in foundational AI capabilities.

Long-term optimism and what to expect

Not everyone on the panel was as pessimistic. Yao Shunyu, Tencent’s newly appointed chief AI scientist and a former OpenAI researcher, said that China’s history of scaling technologies quickly should not be underestimated, and that the world’s leading AI company three to five years from now could be Chinese. He pointed to sectors such as electric vehicles and advanced manufacturing, where Chinese firms become global leaders in a relatively short time.

Still, Yao acknowledged that scale alone would not be enough, as China must make progress in advanced chipmaking, encourage broader enterprise adoption of AI, and allocate more resources to foundational research rather than deployment. “We are experts at optimising within existing frameworks, extracting as much as possible from as few GPUs as possible, but what’s still missing is a risk-taking spirit to define the next paradigm,” he said.

Tang expressed confidence in a new generation of Chinese researchers born in the 1990s and 2000s. If the innovation environment improves and the pressure of constant product delivery eases, he argued, these researchers could focus more deeply on breakthroughs rather than incremental gains.

For now, China’s AI elite looks to the future with a clear-eyed perspective rather than a defeatist one. While significant progress has been made, the combination of chip shortages, compute limits, and hidden US advantages makes a near-term upset unlikely.

Whether China can turn efficiency, talent, and open-source momentum into genuine leadership may ultimately determine its position in the global AI race over the next decade.

Also read: Baidu is positioning Kunlunxin chips as a domestic source of AI compute, as Nvidia’s GPUs remain restricted in China.

The post Chinese Tech Leaders Temper Expectations in Global AI Race appeared first on eWEEK.