The technical capabilities and prospects of artificial intelligence (AI) are driving many countries to increase their investment in related hardware to keep up with the pace of world competition. South Korea’s Acting President Choi Sang-moo said that the supply of 18,000 high-performance GPUs (graphics processors) will be expanded by the first half of next year. At the end of last month, India also announced that it would provide more than 18,000 GPUs to build AI computing facilities to help local startups and development researchers. The current competition requires countries to achieve independent control and diversified supply of hardware, while also improving the level of coordination between hardware and software.
10,000 GPUs to Be Secured by the End of the Year
Choi Sang-moo presided over a meeting of the “Special Committee on AI Computing Infrastructure” to learn more about the “National AI Computing Center” construction project currently being promoted. He said that in order to enhance South Korea’s artificial intelligence competitiveness, the processing capacity of 18,000 high-performance GPUs will be expanded by the first half of next year to support the business community and research institutions. Choi Sang-moo also emphasized that the government will fully support in the future to become a veritable “three major AI powers.” The South Korean government will secure (supply) 10,000 high-performance GPUs within this year through public-private partnerships and open the services of the “National AI Computing Center” as soon as possible. The sixth supercomputer with the processing power equivalent to 8,000 GPUs will be built in the first half of next year, with a focus on supporting research institutions.
The South Korean government has not yet decided which GPU products to purchase, but details such as the budget, GPU model and participating private companies will be finalized by September this year. Currently, South Korea ranks sixth in the global AI rankings, performing well in development and infrastructure, but lagging behind in talent training, research investment and venture capital. This year, the South Korean government increased its annual AI technology budget by 25% year-on-year, allocating 1.8 trillion won to the AI program. In addition, the government plans to launch a cross-departmental program to enhance the country’s AI capabilities.
Multiple Countries Compete for Key Hardware
The number of GPUs required for large AI models depends on the sophistication of the GPU, the amount of data used to train the large model, the size of the large model itself, and how much time developers want to spend on training the large model. Currently, US chipmaker Nvidia accounts for about 80% of the global GPU market, ahead of its competitors Intel and AMD.
Given the importance of GPUs to the development of artificial intelligence, in addition to South Korea, India, Saudi Arabia and other countries are also purchasing this key hardware in large quantities. Ashwini Vishnao, India’s Minister of Railways, Communications, and Electronics and Information Technology, said at the end of last month that India would develop a large-scale AI model and build a large AI computing facility with more than 18,000 GPUs. The country announced in March last year that it would invest $1.2 billion in the “India AI Mission” project, nearly half of which would be used to build computing infrastructure. In addition, as early as 2023, Saudi Arabia planned to purchase thousands of high-performance Nvidia chips. According to public reports, Saudi Arabia had purchased at least 3,000 Nvidia H100 chips at that time.
Countries that want to develop AI choose to “stock up” high-end chips such as GPUs, to a certain extent, because they are worried that it will be difficult to obtain these key hardware in the future. In January this year, the Biden administration introduced new regulations restricting the export of US AI chips and technologies before leaving office. The United States divides countries into three levels, keeping advanced computing power in the United States and its allies while looking for ways to block exports to some countries. The United States has restricted the number of GPUs India can import. Saudi officials told CNBC in an interview in April last year that they hope to obtain Nvidia’s high-performance chips “within the next year.”
Competition Among National Innovation Ecosystems
In his speech, Choi Sang-moo said that the competition for dominance in the AI industry has become increasingly fierce recently, and the competition pattern has also changed from a showdown between companies to a “competition between innovation ecosystems” by the state. However, “the example of DeepSeek tells us that even if there are deficiencies in capital and infrastructure, there are opportunities to surpass in performance.”
The competition in the AI innovation ecosystem covers multiple fields. Hardware includes chips, precision instruments, and high-performance equipment. Software includes infrastructure supply, talent cultivation, and technology transformation efficiency. Environmental aspects include institutional construction, financing channels, and the docking of supply and demand. The competition in the innovation ecosystem has put forward new requirements for the construction and development of the hardware industry chain in various countries. Key hardware must be self-controlled, non-key hardware must have diversified supply, and the level of coordination between hardware and software must be improved. Nowadays, for the security of AI supply chain, countries no longer pursue the most cost-effective solutions, but need to build their own solutions, so there are more market opportunities. If a country wants to gain an advantage in the competition of innovation ecology, on the one hand, it requires a large enough market to absorb innovation, and on the other hand, it needs to have strong innovation capabilities to realize supply.
