In the 21st century, artificial intelligence (AI) has gone from being a scientific curiosity to becoming a new technological battlefield. Have you ever wondered what is at stake in this competition?
The race for AI self-sufficiency between China and the US is not just a technological competition, but a fight for economic and geopolitical leadership. The pursuit of technological sovereignty has become a national priority for the development of Chinese models in deep learning.
We are going to explore how Beijing is weaving its strategy to achieve self-sufficiency in hardware, software, infrastructure and talent. Or at least what he publicly knows about it.
China’s national AI strategy
China has outlined a long-term plan to take the lead in global AI. In the 2017 Next Generation AI Development Plan, the goal was set to make AI a key driver of the economy by 2025 and a global innovation hub by 2030.
With regulations such as the Interim Measures for the Management of Generative AI Services (2023) and the AI Safety Governance Framework, the Asian giant seeks to combine innovation and ethical control.
The 14th Five-Year Plan (2021-2025) reinforces this commitment, prioritizing national chip production and AI R&D.
In April 2025, Xi Jinping urged the Politburo to achieve “autonomy and self-strengthening” in AI, leveraging a comprehensive national system to advance the industry and practical applications.
In addition, incentives are offered in public purchases, reinforcement of intellectual property and talent training programs.
Chinese hardware development for AI
Technological independence requires its own chips. Companies like Loongson Technology and Huawei take over from Intel, Nvidia and AMD. Loongson has introduced the 2K3000 and 3B6000M processors, with eight cores based on the LoongArch architecture and internal GPUs to accelerate AI tasks.
Although they do not yet reach the performance of their Western counterparts, they represent a decisive advance towards a local ecosystem.
Huawei, for its part, competes with the Ascend series. The Ascend910C, based on two 910B units, offers similar power to the Nvidia H100 and is preparing for a mass launch in May 2025.
Despite a low initial production yield (between 20% and 40%), caused by access restrictions to advanced lithography, the Chinese giant seems to be moving forward. The Ascend920 is already on its way, with 30-40% more efficiency, manufactured in 6nm by SMIC.
Additionally, US restrictions have directly targeted foreign manufacturers. In October 2022 and in successive extensions in 2023 and 2024, the US Department of Commerce imposed export controls on 7nm or more advanced chips, ordering TSMC to stop shipments of next-generation processors to unlicensed Chinese companies.
In addition, Washington prohibited the sale to China of AI training GPUs such as the Nvidia H100 and its H20 version, which led Nvidia to provision for losses of $5.5 billion in the first quarter of 2025 to cover inventories that it can no longer market in that market.
AI and Software Models
Did you think only chips mattered? Software is the other half of the equation. In January 2025, startup DeepSeek shook up the market by training a reasoning model for about a tenth of the usual cost.
This “Sputnik moment” for American AI demonstrated that innovation can thrive even under constraints on advanced hardware, encouraging other players to optimize resources.
By the end of 2024, the Beijing division of the Cyberspace Administration of China (CAC) had approved 105 large language models (LLMs), part of more than 300 generative AI services registered across the country, although only a fraction of them (around 10%) have deployed large-scale training in supercomputing).
In addition to DeepSeek, major Chinese technology companies have strengthened their AI portfolios:
- In April 2025, Baidu presented the Ernie 4.5 Turbo and Ernie
- Alibaba Cloud announced the Qwen 2.5-Max version in January 2025, which according to internal tests outperforms DeepSeek-V3 and GPT-4 in key benchmarks, and has released multiple visual and specialized variants under open licenses.
- Tencent launched the Hunyuan Turbo S model at the end of February 2025, capable of responding in less than a second with reasoning capabilities similar to DeepSeek-R1, and integrated into WeChat and Tencent Cloud AI.
Infrastructure
Without data centers, there is no AI. Between 2023 and 2024, more than 500 center projects were announced in China, of which 150 were already operating before the end of 2024.
However, many facilities suffer from low occupancy: up to 80% of the capacity remains unused, the result of excessive planning and a demand focused more on training than on inference.
Despite this, investments do not slow down. Alibaba allocated $50 billion to cloud and AI (2025-2028), while ByteDance plans to invest $20 billion in GPUs and hubs in 2025.
Additionally, the Shenzhen National Supercomputing Center envisions a 2exaFLOPS supercomputer by 2025, driving projects in healthcare, cybersecurity and smart cities.
Scientific research and talent retention
China strengthens its scientific base with national programs covering AI, quantum computing and data analysis. The National Science and Technology Program promotes strategic goals that reduce dependence on foreign countries.
At the same time, government initiatives seek to train elite scientists and cutting-edge teams, equipping laboratories and granting scholarships.
Why is China betting so much on AI?
The quest for self-sufficiency in AI is not just an industrial race, but a geopolitical pulse. US sanctions have been the catalyst that has driven China to accelerate its path towards technological independence.
In 2025, the ban on the sale of AI training GPUs (such as the Nvidia H100 and its H20 version) to Chinese customers exhausted the stocks available in the domestic market in just one year.
This forced companies like Huawei and others to look for quick alternatives, from increasing domestic chip production to redesigning their supply chains.
On the other hand, the lack of access to extreme ultraviolet (EUV) lithography machines has reduced the manufacturing yields of advanced semiconductors, but has also served as a stimulus for Chinese manufacturers to optimize their production processes and develop their own technologies.
Not everything is a bed of roses. The high underutilization of centers and the dependence on less advanced processes, such as SMIC’s N+2, raise questions: is China overinvesting in infrastructure that is then not used? Will your chips be able to compete globally if they don’t reach 3nm or less?
And in the end the big question: Who will win this race? If Beijing manages to overcome its obstacles, it will not only be able to stand up to the US, but also challenge the global technological order. And you, what do you think: are we facing the birth of a new digital superpower?
This post is also available in: