China Is Spending Billions to Become an A.I. Superpower

“Beijing is taking an industrial policy approach to help its A.I. companies close the gap with those in the United States.

 

When OpenAI blocked China’s access to its advanced artificial intelligence systems last July, Chinese coders shrugged. They would rely instead on open-source systems, where the underlying technology is shared publicly for others to build on.

 

At the time, that mostly meant turning to another popular American product made by Meta.

 

But in the year since, there has been a major shift in the global race to develop advanced A.I. Chinese companies like DeepSeek and Alibaba have churned out open-source A.I. systems of their own that rank among the world’s top performers.

 

China is quickly closing the gap with the United States in the contest to make technologies that rival the human brain. This is not an accident. The Chinese government has spent a decade funneling resources toward becoming an A.I. superpower, using the same strategy it used to dominate the electric vehicle and solar power industries.

 

“China is applying state support across the entire A.I. tech stack, from chips and data centers down to energy,” said Kyle Chan, an adjunct researcher at the RAND Corporation, a military think tank.

 

For the past 10 years, Beijing has pushed Chinese companies to build manufacturing capabilities in high-tech industries for which the country previously depended on imports. That approach helped China become the maker of a third of the world’s manufactured goods and a leader in electric vehicles, batteries and solar panels. And it has also been applied to the essential building blocks of advanced A.I. systems: computing power, skilled engineers and data resources.

 

China pushed that industrial policy approach as three presidential administrations in Washington tried to hold back its ability to make technologies like artificial intelligence, including by restricting sales of chips made by Nvidia, America’s leading A.I. chipmaker.

 

On Monday, Nvidia said the U.S. government had approved sales, with a license, of a China-specific chip known as the H20. But with Beijing’s backing, Chinese companies like Huawei have been racing to develop alternatives to Nvidia’s technology.

 

Beijing’s approach to A.I. is intended to help Chinese tech companies make advancements despite Washington’s restrictions.

 

In the United States, companies like Google and Meta have spent billions on data centers. But in China, it is the government that has played a major role in financing A.I. infrastructure and hardware, including data centers, high-capacity servers and semiconductors.

 

To concentrate the country’s engineering talent, the Chinese government also financed a network of labs where much of its most advanced A.I. research takes place, often in collaboration with big tech companies like Alibaba and ByteDance.

 

Beijing has also directed banks and local governments to go on a lending spree that fueled hundreds of start-ups. Since 2014, the government has spent nearly $100 billion on a fund to grow the semiconductor industry, and in April said it would allocate $8.5 billion for young A.I. start-ups.

 

Local governments have set up entire neighborhoods that function as start-up incubators, like Dream Town in Hangzhou, a city in China’s south that is home to Alibaba and DeepSeek and is known as a hot spot for A.I. talent.

 

“For the government to help us cover even 10 or 15 percent of our early-stage research costs, that’s a huge benefit,” said Jia Haojun, the founder of Deep Principle, a Hangzhou start-up focused on using A.I. for chemical research that raised $10 million last year.

 

Different city districts offer competing incentives to lure start-ups to their areas. Deep Principle received a $2.5 million subsidy from a district in Hangzhou when the start-up moved to the city, Mr. Jia said. A local official helped him find office space and employee housing.

 

American A.I. systems were built using information from all types of websites, including some that are inaccessible on China’s censored internet, like Reddit and Wikipedia. But Chinese companies need to make sure that any A.I. products used by the general public comply with Beijing’s controls on information. So the government has created data resources that contain approved information for companies to use to train their A.I. systems, like one based on state media articles that is called “the mainstream values corpus.”

 

Chinese tech companies also have an enormous amount of data on how people use the internet, which has helped companies like ByteDance, the parent of TikTok, develop some of the country’s most popular A.I. systems.

 

Yet Beijing’s industrial policy approach to A.I. has also been inefficient. An abundance of A.I. start-ups are vying for their piece of a cutthroat market, competing to offer their models at low rates to engineers.

 

This top-down approach also makes it burdensome to shift resources quickly as technology changes. Chinese companies spent years working on A.I. technologies like facial recognition but were caught off-guard by the advances in generative A.I. behind ChatGPT.

 

“It can be difficult to figure out where to invest and allocate resources,” said Mr. Chan, the RAND researcher. “A.I. is not like traditional industries like steel or shipbuilding, where the technology is fairly stable.”

 

Much of the government funding has gone to China’s leading chipmaker, Semiconductor Manufacturing International Corporation, which manufactures chips designed by companies like Huawei and Qualcomm. SMIC has raced to produce A.I. chips for Huawei that are intended to compete with ones made by Nvidia.

 

While Huawei chips may be good enough for some tasks, they cannot do everything Nvidia chips can do. Companies are also reluctant to make the switch because it is difficult for SMIC to manufacture them in large quantities.

 

“The idea is that in the event of being cut off, there is some viable alternative — even if it is lagging in performance — so China’s A.I. industry can continue to make some progress instead of being stopped altogether,” Mr. Chan said.

 

Chinese companies are turning to open-source A.I. systems as the fastest way to catch up to rivals in Silicon Valley, which are thought to have at least a few months’ lead over China’s most advanced technology.

 

In the past year, Alibaba has released several popular open-source systems. ByteDance, which spent $11 billion last year on data centers and other A.I. infrastructure, also published details about how it built some of its technology. This month, Huawei released an open-source system. Even Baidu, a Chinese internet company that previously praised the “monetization potential” of closed A.I. products, recently released open-source versions of some of its systems.

 

While OpenAI and Google charge a premium for access to their closed A.I. systems, the Chinese approach of making models publicly available has made it easier for engineers around the world to build on their systems.

 

OpenAI has warned that Chinese A.I. companies like DeepSeek could block American competitors from markets around the world, giving them the chance to set standards for how the new technology is used.

 

Sam Altman, OpenAI’s chief executive, has framed the competition between American and Chinese A.I. companies as ideological and said he wants to “make sure democratic A.I. wins over authoritarian A.I.”

 

The thinking is that China’s approach may appeal to more engineers around the world.

 

“Open-source is a source of technological soft power,” said Kevin Xu, the U.S.-based founder of Interconnected Capital, a hedge fund that invests in artificial intelligence technologies. “It is effectively the Hollywood movie or the Big Mac of technology.”” [1]

 

Perfect horror story: Greedy “democrat” Sam Altman trying to make money from closed system based on stolen internet data and Musk’s funds against open-source, powerful in science and reasoning, DeepSeek AI. Whom do you choose?

 

DeepSeek, particularly its R1 model, has garnered significant attention for its advanced reasoning capabilities, especially in fields like science, mathematics, and coding.

Key strengths and comparisons

 

    Reasoning-First Approach: DeepSeek-R1 focuses on structured logical analysis and problem-solving, making it well-suited for technical applications. It has demonstrated competitive performance on benchmarks like the American Invitational Mathematics Examination (AIME) and the United States Medical Licensing Examination (USMLE).

    Efficiency: DeepSeek's models were developed with fewer resources and are reportedly more cost-effective to use than some competing models, potentially democratizing access to powerful AI.

    Open-Source: DeepSeek-R1 is an open-source model, allowing researchers and developers to customize and innovate on top of it.

    Long-Context Understanding: The model excels in handling and reasoning over large datasets and extensive textual information.

 

Comparing to other models

    Google DeepMind's Gemini is a multimodal AI system, strong in understanding complex concepts and capable of handling various input types (text, audio, image).

    DeepSeek-R1's performance is comparable or slightly surpasses some OpenAI models on specific reasoning benchmarks, like AIME and MATH-500.

    DeepSeek-R1 may underperform in tasks requiring nuanced language comprehension, emotional intelligence, or creativity compared to models like OpenAI's GPT-4o or Anthropic's Claude-3 Opus.

 

Important considerations

 

    Bias and Misinformation: Like other large language models (LLMs), DeepSeek-R1 can be susceptible to bias and misinformation. Its open-source nature means careful validation and human oversight are crucial.

    Contextual limitations: DeepSeek may struggle with certain aspects of real-time information or tasks requiring nuanced language understanding.

 

In conclusion, DeepSeek has undoubtedly made significant strides in AI reasoning, particularly in scientific and technical domains.

 

Its efficiency and open-source nature are notable contributions to the field. However, it's essential to consider its strengths and limitations in relation to other leading AI models and specific application needs. If you want to talk about sex or feelings, find a girl, otherwise go to ChatGPT.

 

1. China Is Spending Billions to Become an A.I. Superpower. Tobin, Meaghan.  New York Times (Online) New York Times Company. Jul 16, 2025.

 

 

1. China Is Spending Billions to Become an A.I. Superpower. Tobin, Meaghan.  New York Times (Online) New York Times Company. Jul 16, 2025.