Hong Kong
—
When China’s greatest synthetic intelligence gamers gathered for a landmark assembly in Beijing in January, one query was in the highlight: What are the possibilities of a Chinese AI agency overtaking US frontrunners in the subsequent three to 5 years?
The reply from a prime AI scientist current at the gathering was surprisingly blunt: “Below 20 percent,” mentioned Justin Lin, technical lead for Chinese tech big Alibaba’s Qwen AI fashions. “And I think 20 percent is already very optimistic.”
The sobering evaluation stood in stark distinction to a yr of headlines celebrating China’s AI increase.
Since little-known startup DeepSeek shocked the world with a robust AI mannequin it mentioned was constructed at a fraction of the value of American equivalents, Chinese firms have topped world downloads for freely-available-to-use fashions and raised enormous sums in market debuts.
Yet regardless of the fanfare, some main Chinese AI builders have warned that China might have fallen additional behind in growing frontier fashions. Experts level to restricted entry to superior chips and restricted capital as lingering constraints.
Lin was not alone. Tang Jie, founding father of one in every of the outstanding Chinese AI startups Z.ai, also called Zhipu, mentioned the efficiency hole between Chinese and US fashions “may be widening.”
“In some areas we may be doing fairly well, but we also need to acknowledge the challenges and gaps we still face,” he mentioned at the identical Beijing assembly.
But that evaluation doesn’t imply China’s AI industry is stagnating.
Constraints in accessing high-performance chips and capital in addition to the nation’s distinctive tech ecosystem have fueled a divergent technique from the US – making AI fashions accessible for public use, or open supply.
The technique, which Beijing and builders see as a solution to speed up progress and compete with US rivals, has helped Chinese firms make notable features. Companies are aggressively rolling out AI purposes based mostly on these fashions for actual world makes use of. Industries are integrating the know-how in manufacturing, e-commerce and robotics.
In a televised new year address, Chinese chief Xi Jinping hailed the nation’s quick rising progressive capabilities, citing AI fashions which can be “racing ahead” and what he known as “breakthroughs” in homegrown chips, as Beijing pushes for technological self-reliance.
Qwen, for instance, overtook Meta’s Llama final September as the most-downloaded open mannequin on Hugging Face, a significant platform for AI fashions and instruments. Even American corporations like Airbnb have embraced it to energy AI customer support.
An unprecedented wave of Chinese AI listings has additionally unfolded in Hong Kong. In January, unicorn startups Z.ai and MiniMax, each builders of aggressive open fashions, went public, elevating $560 million and $620 million respectively, with their share costs surging.
Global tech firms are taking discover. In December, Meta announced it would purchase Manus, an AI agent agency based in China and later relocated to Singapore. While the deal has triggered Beijing’s regulatory review course of – elevating the chance it may very well be unwound, it is a testomony to the advances in Chinese AI know-how.
DeepSeek, which emerged as a poster little one for China’s AI increase after releasing an open mannequin a yr in the past that delivered close to industry-leading efficiency with far fewer sources, can be reportedly set to unveil a brand new mannequin with enhanced coding capabilities later this month.
A key driver of China’s AI momentum stems from its aggressive embrace of open fashions, a shift sparked by DeepSeek’s breakout success. Since then, Chinese AI labs have jumped on the open-model bandwagon, in distinction to their US rivals who’ve largely eschewed that method with a purpose to defend their mental property.
The flip to open fashions in China “has dramatically reduced costs for developers and enterprises,” mentioned Poe Zhao, analyst and founding father of Hello China Tech, a e-newsletter centered on the nation’s tech industry.
“Cloud providers like Alibaba use open models to drive cloud adoption, startups use openness to build developer ecosystems quickly.”
The increase of open fashions in China prompted their world utilization to skyrocket from a miniscule 1.2% in late 2024 to as a lot as practically 30% final yr, in line with a late 2025 study by OpenRouter, a market for AI fashions.
Alibaba alone, for instance, has launched greater than 400 open Qwen fashions, yielding over one billion downloads by earlier this month, in line with the firm.
“Open sourcing has to a certain extent become an industry consensus in China,” mentioned Lian Jye Su, chief analyst at Omdia, a know-how analysis agency.
The technique not solely reduces prices but additionally lessens Chinese firms’ publicity to geopolitical dangers.
“If one day the entire company were to be sanctioned (by the US), then at least the products based on its open-sourced models could still be used by others,” Su mentioned.
There can be a tough financial actuality behind the technique.
“There’s no other choice. The reason is that Chinese people, like consumers, as well as enterprises, do not pay for software,” mentioned Jenny Xiao, a accomplice at Leonis Capital, an AI-focused enterprise capital agency.
But whereas Chinese fashions reign supreme in the open mannequin area, closed fashions developed by US heavyweights – together with OpenAI’s GPT, Google’s Gemini and Anthropic’s Claude – proceed to dominate general efficiency benchmarks. Closed fashions nonetheless account for roughly 70% of complete downloads, in line with OpenRouter.
Bottlenecks stay
The efficiency hole between Chinese and US fashions, whereas not huge, persists largely on account of constraints on computing energy and capital.
“OpenAI, Anthropic and other American companies are putting massive compute into next-gen research, whereas we are relatively stretched thin,” Alibaba’s Lin mentioned.
Washington’s export controls have barred Chinese firms from acquiring cutting-edge chips, corresponding to Nvidia’s Blackwell and the recently launched Rubin series, in addition to the gear wanted to fabricate them. Turning to home chipmakers for much less superior semiconductors additionally doesn’t repair the drawback for these AI mannequin builders.
“Their problem is they can’t get the volume,” mentioned Paul Triolo, a China and tech skilled at consulting agency Albright Stonebridge, as US restrictions additionally hamper Chinese chipmakers’ capacity to scale output.
While Trump has authorized the export of Nvidia’s H200 chips, two generations behind Rubin, China has but to formally greenlight their import. Triolo mentioned this places Beijing in a troublesome bind, as it weighs the near-term want for superior chips in opposition to its push for self-reliance.
DeepSeek, Alibaba, and different Chinese tech giants ByteDance and Tencent have been granted conditional approvals by Beijing to buy a certain quantity of H200s, Reuters reported final month, citing nameless sources.
Unlike US startups that can elevate a number of rounds of enterprise capital funding, Chinese AI firms face a thinner investor base and rising stress to display industrial viability shortly. That is forcing firms like Z.ai and Minimax into public listings forward of their American rivals.
“A lot of these Chinese companies are trying to get out as soon as possible, because they’re burning a lot of cash and getting into public market is the easiest way to do that,” Xiao of Leonis Capital mentioned.
Adding to the problem is China’s home market, which is smaller and infrequently calls for heavy customization, making it more durable for AI mannequin builders to show a revenue, she added.
Still, specialists warning in opposition to underestimating China’s long-term prospects.
Chinese firms have excelled at speedy deployment in consumer-facing purposes and integrating AI into industrial use, mentioned Deepika Giri, head of AI analysis at market analysis agency IDC.
Even with limitations in China’s AI fashions, the know-how is “becoming more ubiquitous,” she mentioned. “It may not be the best (model) available globally, but still they’re integrating, so that the industrialization of AI is progressing extremely rapidly.”
Beijing has additionally made AI utility a spotlight, unveiling an action plan to deepen the use of AI in manufacturing final month, a part of a nationwide technique to improve its industrial sector.
From open fashions to real-world purposes, China has proven that constraints haven’t stopped it from gaining floor.
Yao Shunyu, a former OpenAI researcher who just lately joined Tencent as their chief AI scientist, mentioned China has repeatedly demonstrated a capability to “very quickly catch up or replicate” western technological developments and in some areas “go on to do it even better.”
He pointed to the nation’s highly effective manufacturing sector and electrical car manufacturing as examples in his remarks at the identical summit with Tang and Lin in Beijing.
For Yao, nevertheless, the larger problem for China looking for to drag forward of the US is cultural: a scarcity of risk-takers, regardless of an abundance of top-tier expertise.
“Can we actually lead the creation of new paradigms? I think this is, in some sense, the one key problem China still needs to solve,” he mentioned.