'Gamer Essentials' GPU Becomes Central Axis in US-China Hegemony Dispute

Semiconductor company Nvidia's graphics processing units (GPUs) have been included in the United States' export restriction list targeting China. / Photo by Song Hyundo, Asia Economy intern reporter

[Asia Economy Reporter Lim Juhyung, Intern Reporter Song Hyundo] The U.S. Biden administration has issued an export restriction order on semiconductor design company Nvidia's artificial intelligence (AI) graphics processing units (GPUs) to China. The Biden administration is reportedly concerned that America's advanced technology could be used in Chinese military projects. This effectively recognizes GPUs as strategic materials. Just 30 years ago, GPUs were regarded as expensive computer chips for running high-end games, but now they have emerged as a central axis in the U.S.-China hegemony conflict.

◆ Biden Administration Targets China's AI Research Capabilities

On the 31st of last month (local time), Nvidia disclosed to the U.S. Securities and Exchange Commission (SEC) that it had received an order from the U.S. government to restrict exports of high-performance GPUs to China and Hong Kong. The specific products banned from export are the AI and data center GPUs A100 (released in 2020) and H100 (released this year). Going forward, Nvidia must request a license from the U.S. Department of Commerce before selling A100 and H100 to Chinese customers, and if the license is denied, sales are prohibited.

Nvidia expects to suffer sales losses of approximately $400 million (about 545.8 billion KRW) due to this measure. Since the export ban news broke, Nvidia's stock price has dropped 14.77% over five days. The Chinese tech industry, which will immediately face a supply cutoff of high-performance GPUs, is also expected to experience setbacks in various AI projects. At a press conference on the 1st, the Chinese Ministry of Foreign Affairs strongly protested, stating, "The U.S. is trying to suppress developing countries by leveraging its scientific and technological superiority."

NVIDIA's A100 GPU-based AI computing system 'DGX Pod' / Photo by NVIDIA website capture

The Biden administration appears to have played the card of GPU export restrictions as a measure to curb China's "AI rise." The semiconductors specifically designated by the U.S. government for export license issuance are products with specifications such as the peak performance and input/output (I/O) performance of the A100. These indicators relate respectively to the maximum computational processing capacity per second of AI semiconductors and the data flow between semiconductors and memory devices. Additionally, the A100 is a representative high-performance AI GPU installed in data centers worldwide today. According to Bloomberg and others, Nvidia holds over 80% of the global discrete GPU market share, and particularly dominates 95% of the Chinese AI semiconductor market. In other words, the U.S. is precisely targeting only the semiconductors necessary for China to build an advanced AI research environment, aiming to disrupt the supply chain.

◆ Nvidia Emerges as the Core of AI Hegemony Competition

Nvidia, caught in the technological dispute between the two great powers, was not initially the core of the AI industry. Founded in 1993 by Taiwanese-American CEO Jensen Huang, Nvidia originally started as a "gaming semiconductor company." At that time, the PC and console gaming industries were transitioning from traditional 2D games to a more advanced 3D gaming era. However, to run 3D games smoothly, a GPU capable of handling complex graphic computations was needed, and Nvidia specialized in designing and selling these GPUs.

As games equipped with 3D graphics increased in the 1980s and 1990s, the demand for Nvidia's GPUs also surged rapidly. / Photo by Internet Community Capture

Effectively paralleling the development of the gaming industry, Nvidia has innovated gaming graphics technology over its 30-year corporate history, and the 'GeForce' GPU brand created by Nvidia has become an essential computer chip for gamers.

The transition of Nvidia from a "gaming semiconductor company" to an "AI company" was largely accidental. Google's AlphaGo, which gained worldwide attention in 2016 as an AI that plays Go, was implemented using an algorithm called "deep learning." Deep learning processes as many simple calculations simultaneously as possible to rapidly learn from vast amounts of data. This means the trend in computer algorithms shifted from "performing complex and large calculations one by one" to "performing many small and simple calculations simultaneously." Structurally, GPUs are very well suited for the latter.

Structural differences between CPU (left) and GPU. The CPU has fewer but high-performance ALUs (Arithmetic Logic Units) responsible for computation, whereas the GPU has a large number of much simpler ALUs arranged in parallel. Because of this, the GPU is better suited for processing algorithms where simultaneous computation is important, such as deep learning. / Photo by Online Community Capture

CEO Huang's keen business sense also played a role. Recognizing Nvidia's GPUs' potential to dominate the AI industry, he aggressively invested in research and development (R&D) and mergers and acquisitions to create an ideal system for AI computers, achieving technological maturity that competitors could not match. Moreover, Nvidia's engineers actively collaborated with AI researchers worldwide to optimize software for GPU operation. Naturally, scientists and engineers developing AI flocked to the "Nvidia ecosystem," and now it has become nearly impossible to operate outside the GPU development environment created by Nvidia.

◆ "U.S. Export Restrictions May Have 'Unintended Consequences'"

It seems inevitable that China's AI industry will suffer significant damage immediately. Chinese tech companies such as Baidu, Huawei, and Alibaba may soon be unable to equip their data centers with A100 or H100 GPUs. While they could use Nvidia's previous generation V100 as a substitute, its performance is significantly inferior to the later models. This means they cannot avoid falling behind in the AI competition, where data learning speed is critical.

However, Nvidia's absence from the Chinese AI market could present new opportunities for other competitors. Unlike the past when Nvidia held a peak position in chip performance, many "AI-dedicated semiconductors" have now been developed. British company Graphcore and U.S. startup Cerebras are representative examples, developing computer chip designs specialized for parallel computation and seeking market entry opportunities.

GPU for BR100 Data Center Released by Chinese Startup Biren / Photo by Biren

It could also be an opportunity for the Chinese tech industry to build a domestic supply chain to compete against Nvidia. For example, Chinese GPU design startup Biren Technology unveiled its data center GPU "BR100" at last month's global semiconductor event, the Hot Chips Conference, challenging Nvidia.

Some experts have also voiced concerns. Karl Freund, founder of U.S. AI semiconductor analysis firm Cambrian AI, said, "I understand the government's concerns about companies like Nvidia and AMD selling AI chips to China," but he also pointed out that "(the export restrictions) could have 'unintended consequences' that help Chinese companies like Biren and other startups, ultimately undermining U.S. competitiveness."

This content was produced with the assistance of AI translation services.