China vs. USA: AI is Decided by Money and Chips

What happened?

According to information published by Bloomberg and reported by Xataka, China is designing a plan to allocate approximately 2 trillion yuan (about $295 billion) over the next five years to build artificial intelligence data centers across the country. The project, still in early discussion stages, envisions creating a network of interconnected computing hubs that would pool resources currently scattered across regions, aiming to function as a cohesive system by 2028. The figure is comparable to the GDP of countries like Finland or Chile, and far exceeds the $50 billion the United States has allocated through the CHIPS and Science Act for semiconductor manufacturing, though that fund focuses on production, not data centers. This move is not isolated: since 2020, China has promoted the 'East Data West Computing' initiative, which seeks to relocate data centers from the eastern coast to the west, where renewable energy is cheaper. The new plan would integrate and expand those efforts.

In parallel, China is intensifying pressure on NVIDIA. Export restrictions imposed by the United States since October 2022 already limit the sale of the company's most advanced chips to China, and Beijing is actively promoting the development of domestic alternatives, such as Huawei's processors (Ascend 910B and 910C) and other local manufacturers like Cambricon or Biren Technology. The goal is to reduce technological dependence and strengthen its own supply chain. Bloomberg notes that the data center plan could prioritize the use of domestic chips, accelerating the adoption of Chinese hardware even if its performance is inferior to NVIDIA's H100 or B200. According to estimates from consultancy SemiAnalysis, Huawei's chips achieve between 60% and 80% of the performance of NVIDIA equivalents in inference tasks, but the gap is larger in training.

Why is it important?

Artificial intelligence has become critical economic infrastructure. Whoever controls the computing power and chips needed to train and run AI models will have a decisive strategic advantage. The United States and China are the two main contenders, but their approaches are radically different: while the U.S. relies on private innovation (Google, Microsoft, OpenAI, NVIDIA), China turns to the state as the central architect. This plan recalls China's massive investment in 5G, which allowed it to deploy over 2.3 million base stations by 2023, according to China's Ministry of Industry, surpassing the rest of the world combined. In AI, the bet is similar: using state financial muscle to leapfrog stages.

The Chinese plan is not just massive investment; it seeks to solve a structural problem: the fragmentation of its data centers. Currently, computing resources are spread across regions and state-owned enterprises, limiting efficiency. The national network would allow companies and agencies to access high-performance capacity more broadly and coordinatedly. Additionally, electricity costs for data centers in China are 30% lower in western provinces like Gansu or Ningxia, according to the National Energy Administration, making the plan economically viable. For global tech companies, this represents a paradigm shift: it is no longer enough to have the best algorithm; you need the infrastructure to scale it.

What consequences will it have?

If the plan materializes, China could accelerate the development of its own AI models, reduce its dependence on foreign chips, and pressure NVIDIA in a key market. For NVIDIA, losing access to the Chinese market would be a heavy blow, although global demand for its chips remains high. In 2024, NVIDIA generated approximately 15% of its revenue from China, according to its financial reports, but U.S. restrictions have already reduced that figure. In the long term, an independent Chinese tech ecosystem could consolidate, with its own standards and platforms. This is not new: in 2019, China banned the use of foreign software in public agencies, promoting local alternatives like Kirin (operating system) or Kingsoft (office suite). In AI, the dynamic would be similar.

For global tech companies, this means a bifurcation scenario: two tech blocs with different hardware, software, and regulations. Startups and developers will have to adapt to dual environments, increasing compatibility costs. Investors, meanwhile, will need to closely follow regulatory decisions and advances in Chinese semiconductors. A 2025 McKinsey report estimates that the global AI infrastructure market will reach $800 billion by 2030, and China could capture a quarter if it achieves autonomy. However, the risk of overcapacity exists: massive investment in data centers could lead to oversupply if AI demand does not grow as expected, similar to the dot-com bubble in 2000.

“The AI race is no longer just about who has the best models, but who can build the material base to feed them,” notes the Xataka analysis.

What should readers know?

• The Chinese plan is in early stages and details may change. It is not a firm decision but a draft under discussion. Bloomberg cites anonymous sources, and the Chinese government has not made official announcements.
• Pressure on NVIDIA is real, but the company remains dominant. Its chips are hard to replace in the short term, especially for training frontier models. However, China has already made progress with chips like the Ascend 910C, which Huawei claims matches the H100 in inference.
• China has already demonstrated the ability to mobilize massive resources in technology (5G, solar panels). AI could be the next front. In 2024, China installed 45 GW of solar capacity, more than any other country, according to the IEA. The same logic of scale now applies to computing.
• The U.S. response will be key: more investment in chips, alliances with allies, and possibly new restrictions. The U.S. government has already announced $39 billion in subsidies for semiconductor manufacturing and could expand sanctions on Chinese companies like Huawei. Additionally, cooperation with Japan, South Korea, and the Netherlands on export controls is crucial to limit China's access to advanced lithography technologies.