Top artificial intelligence researchers in China are now expressing a view that contrasts with the optimistic narratives of the past year. They suggest that their country is unlikely to catch up to the United States in the near future, primarily due to challenges with computer chips.
“The truth may be that the gap is actually widening,” Tang Jie, who founded the Chinese AI company Zhipu, stated at a conference in Beijing last weekend. He added, “While we’re doing well in certain areas, we must still acknowledge the challenges and the disparities we face.”
The issue of chip scarcity became particularly apparent when Nvidia unveiled its new Rubin hardware in January. The company listed several American firms as purchasers but excluded all Chinese AI developers. This exclusion is a direct consequence of American regulations that prohibit Nvidia from selling directly to China.
In response to these restrictions, Chinese businesses have begun exploring options to rent computing power from data centers located in Southeast Asia and the Middle East to gain access to Rubin chips, according to individuals familiar with these discussions. This follows similar efforts last year to acquire chips from Nvidia's Blackwell line.
While these workarounds through other countries are largely legal, they result in Chinese AI developers having fewer chips and facing more complications compared to their well-funded American counterparts.
Industry Leaders Assess Catch-Up Odds
During the same conference, Justin Lin, who oversees the development of Alibaba’s AI model, Qwen, was asked about the likelihood of any Chinese company surpassing OpenAI and Anthropic within the next three to five years. He estimated the odds at 20% or less.
American export controls have discouraged many Chinese firms from pursuing the development of cutting-edge AI, which demands immense computing power. Consequently, these companies are shifting their focus to integrating AI into everyday products. In contrast, American companies continue to acquire the latest chips to drive further advancements.
“A massive amount of compute at OpenAI and other American companies is dedicated to next-generation research, whereas we are stretched thin,” Lin commented. “Just meeting delivery demands consumes most of our resources.”
Analysts at UBS estimate that China’s major internet companies invested approximately $57 billion in capital projects last year, with a significant portion allocated to AI. This figure represents roughly one-tenth of the spending by American companies.
Despite these challenges, China's AI sector is not being entirely dismissed. Developers such as DeepSeek have demonstrated the ability to achieve substantial results with limited resources. Furthermore, two other AI firms, Zhipu and MiniMax, collectively raised over $1 billion through stock offerings in Hong Kong this month, with MiniMax shares more than doubling their initial price.
“Despite a more challenging operating environment, investors continue to price in the possibility of technological catch-up or breakthrough,” observed Alyssa Lee, a seasoned tech investor now involved with an AI startup. “That optimism itself speaks to the level of innovation Chinese companies have demonstrated.”
DeepSeek Leverages Efficiency to Narrow the Gap
DeepSeek gained significant attention in the United States a year ago with the release of a powerful AI model. Since then, the company has shared methodologies for enhancing AI development efficiency, which have been adopted by some Western researchers. This month, DeepSeek published two papers detailing a new training setup designed to enable developers to build larger models with fewer chips, alongside a memory design aimed at improving model performance.
According to Epoch AI, models developed by DeepSeek and Alibaba have reduced the gap with leading American models to approximately four months, down from an average of seven months in recent years. A notable aspect is that many prominent Chinese AI models are open source, allowing for widespread download and modification. This strategy enhances the visibility of Chinese companies while leading American models remain proprietary.
However, DeepSeek has encountered obstacles. During the development of its latest flagship model last year, the company experimented with chips from Huawei and other Chinese manufacturers. The performance of these chips did not meet expectations, leading DeepSeek to utilize Nvidia chips for certain tasks, as reported by individuals familiar with the project. The company has since made progress and plans to release the model in the coming weeks.
“The primary bottleneck is chip-manufacturing capacity,” stated Yao Shunyu of Tencent at the Beijing event. Yao recently transitioned from OpenAI to lead Tencent’s AI initiatives.
H200 Chip Approval Has Limited Impact
Industry insiders suggest that Washington’s recent decision to permit Nvidia to sell its H200 chip to China is unlikely to significantly alter the competitive landscape. The H200 chip is two generations behind the Rubin line and is considered insufficient for training advanced AI models. Companies are still awaiting approval from Beijing to purchase these chips, with Chinese officials reportedly developing regulations to govern such acquisitions, as previously reported by Cryptopolitan.
Nvidia's business in China continues to face political challenges. Revenue from China decreased by 45% year-over-year, amounting to approximately $3 billion in the most recent quarter. Despite this, Nvidia reported overall third-quarter revenue of $57 billion, marking an increase of over 60%, and briefly became the first company to reach a $5 trillion valuation last fall.
A longer-term concern for Nvidia is the potential for Chinese companies to develop open-source software compatible with various chip types, not solely Nvidia's. A significant portion of Nvidia's competitive advantage stems from its CUDA software platform, which effectively ties developers to its hardware.
“That’s the real nightmare scenario,” commented Seaport analyst Jay Goldberg.
Should Chinese developers, compelled to use domestic chips, succeed in creating software tools that achieve global adoption, it could undermine Nvidia's established market position.
Nvidia CEO Jensen Huang, however, maintains a different perspective. “As I have long said, China is nanoseconds behind America in AI,” he wrote on X in November. “It’s vital that America wins by racing ahead and winning developers worldwide.”