Chinese Tech Giant Succeeds in 'Cutting AI Costs with Only Chinese Chips'—Potential to Threaten NVIDIA's Dominance

Since 2024, the evolution of AI has been entering a new phase. The driving force behind this is the 'AI chip,' which supports massive computational power. Until now, NVIDIA GPUs from the United States have been the de facto industry standard for generative AI development. However, signs of change are beginning to appear in that position.

It has been reported that the Chinese fintech giant, Ant Group, has significantly reduced AI training costs using only Chinese-made chips. According to Bloomberg, Ant has developed a method to reduce AI training costs by up to 20% using domestic chips designed by Alibaba and Huawei. It is said that they have reached a level of performance comparable to NVIDIA chips.

This move has the potential to significantly redraw the geopolitical map of AI infrastructure supply.

1. Ant Group's New Strategy: Training AI Models with Chinese Chips

1-1. Who is Ant Group?

Ant Group is a fintech company supported by Alibaba founder Jack Ma, and it operates 'Alipay,' China's largest digital payment service. In addition to financial services, it focuses on AI, data analysis, and cloud computing, and is also involved in supporting national-scale infrastructure.

Ant has long focused on the development and utilization of AI models, and it boasts high competitiveness in the Chinese market, particularly in AI for finance (fintech AI). The 'AI chip replacement experiment' reported by Bloomberg is seen as part of this effort.

1-2. Specific Examples of Chinese Chips: The Roles of Alibaba and Huawei

According to reports, the chips used by Ant this time are the 'Hanguang' series of AI chips developed by Alibaba's subsidiary PingTouGe, and Huawei's 'Ascend' series.

• Hanguang 800: An AI inference chip announced by Alibaba in 2019. It has strengths in image recognition and NLP.

• Ascend 910: A high-performance AI training chip designed by Huawei. It boasts a computational performance of up to 256 TFLOPS.

Since the tightening of US export controls on China, research and development of these chips as alternatives to NVIDIA have accelerated rapidly within China.

2. Cost Reduction Mechanisms and Performance Comparison: NVIDIA vs. Chinese Chips

2-1. Cost Structure of AI Training

When training generative AI models, a large number of GPUs are required. Especially for large language models (LLMs), it is not uncommon for thousands of GPUs to be used in parallel. For example, it is estimated that over 10,000 NVIDIA A100s were used to train OpenAI's GPT-4.

For this reason, the key to success for AI development companies is 'which GPU they can use, and how cheaply and efficiently they can use it.'

2-2. The Significance of the 20% Reduction Seen in Bloomberg's Report

In Ant Group's initiative, it is stated that they were able to 'reduce AI training costs by up to 20%' as a result of developing optimal usage methods for Chinese chips. This 'reduction' is seen as the result of a combination of factors, including not just power consumption and hardware unit prices, but also software stack optimization, improved computational efficiency, and cluster management technology.

In addition, there are reports that results equivalent to NVIDIA chips were obtained in terms of performance (throughput, latency, and power consumption), and if this is true, a situation where 'Chinese chips have the advantage in cost efficiency' could occur.

3. Geopolitical Risks and NVIDIA's Predicament

3-1. Strengthening Export Controls and China's Self-Reliance Strategy

The U.S. government strictly restricts the export of high-performance NVIDIA GPUs (such as the A100, H100, and the latest Blackwell B200) to China. However, it has been reported that Chinese companies are continuing to procure them through channels like Hong Kong, meaning NVIDIA has not completely lost the Chinese market.

Nevertheless, the Chinese government and companies are clearly steering toward 'de-NVIDIA dependency.' If moves to demonstrate the practicality of domestic chips, like this one by Ant Group, continue to spread, a medium- to long-term impact on NVIDIA's sales in China will be unavoidable.

3-2. The 'DeepSeek Shock' That Affected Stock Prices

In fact, at the beginning of 2024, NVIDIA's stock price temporarily plummeted after another Chinese company, 'DeepSeek,' released information claiming they could 'run LLMs with only a small number of chips.' In this way, news about the 'rise of Chinese AI chips' has become a sensitive topic for financial markets as well.

4. Future Points of Interest and Industry Impact

4-1. The 'Reliability' of Chinese Chips Is the Next Challenge

While this report is positive news in terms of cost reduction and replacing NVIDIA, concerns remain regarding Chinese chips in areas such as 'long-term stability, software compatibility, and support systems.' Building a software foundation to replace CUDA (NVIDIA's proprietary development environment), which many developers are accustomed to, also remains an ongoing challenge.

4-2. A Spark for the U.S.-China AI Hegemony Competition

If Ant Group's success is proven, other Chinese giants (Baidu, Tencent, ByteDance, etc.) may follow suit. This would align with the 'semiconductor self-sufficiency strategy' promoted by the Chinese government, and the confrontational structure of U.S.-China technological hegemony in the AI field will likely become even clearer.

The U.S. side will also be required to continue striving to maintain its competitiveness through its own companies (NVIDIA, Intel, AMD, etc.).

Ant Group's announcement has the potential to be a major milestone for the AI industry. In particular, the claims of 'performance comparable to NVIDIA using Chinese-made chips' and '20% reduction in AI costs' suggest a transition from a mere prototype to the practical application stage.

However, at this point, there is a lack of verifiable data and third-party reviews, so we must wait for further information disclosure regarding actual reliability and reproducibility.

NVIDIA has reigned as the king of the AI chip world in terms of performance, ecosystem, and developer community. But now, as the evolution and self-reliance of Chinese companies accelerate, signs of a 'post-NVIDIA' era may be truly emerging.

Developments that cannot be overlooked by investors, AI researchers, and policymakers will likely continue to expand in the future.

Related Articles