The Continuing Growth of AI Infrastructure Demand and the Rise of DeepSeek Signal a Greater Emphasis on Cost-Effectiveness in the Industry

In the dynamic landscape of the AI industry, recent developments have set the stage for significant transformations. According to the latest research by TrendForce, the consecutive releases of AI models such as DeepSeek-V3 and DeepSeek-R1 by DeepSeek are bound to prompt end customers to evaluate the rationality of their investment in AI infrastructure more carefully in the future. This evaluation will lead them to adopt more efficient software computing models, thereby reducing their reliance on hardware such as GPUs. Concurrently, Cloud Service Providers (CSPs) may increase their adoption of in-house ASIC infrastructure to cut down on construction costs. As a result, the actual demand for GPU AI chips or semiconductors in the industry may undergo changes after 2025.

The global AI Server market has witnessed rapid growth since 2023. It is expected that by 2025, the proportion of AI Servers in the overall Server shipments will exceed 15%, and by 2028, it is likely to approach 20%. In recent years, major CSP players have been actively expanding in line with the demands of AI training. Starting from 2025, their focus will shift towards edge AI inference. In addition to adopting new-generation GPU platforms like NVIDIA Blackwell, companies such as AWS are also ramping up their efforts in developing in-house ASICs. This is to enhance cost-effectiveness and meet the specific requirements of AI applications. Chinese CSPs and relevant AI players like DeepSeek are placing great emphasis on developing more efficient AI chips or algorithms to promote the diversified development of AI demand and applications.

The AI industry has traditionally relied on expanding models, increasing data, and enhancing hardware efficiency for its growth. However, cost and efficiency have now emerged as significant challenges. DeepSeek has employed the Model Distillation technology. This technology compresses large models to boost inference speed and reduce hardware requirements. At the same time, it makes full use of the degraded versions of NVIDIA Hopper chips, maximizing the utilization of computing resources. Its cost advantages stem from the selection of high-performance hardware, the adoption of new distillation techniques, and an open-source API strategy. This not only optimizes the balance between technology and commercial applications but also showcases the trend of the AI industry towards greater efficiency.

TrendForce points out that the Chinese AI market is expected to develop in two key directions. Firstly, AI-related enterprises will accelerate their investment in the development of self-owned AI chips or supply chains. For example, in addition to purchasing the currently available H20 as much as possible, large Chinese CSPs will accelerate the expansion of their own ASICs in the future, which will be applied in their own data centers. Secondly, China will leverage its existing Internet infrastructure advantages and use software to make up for hardware deficiencies. DeepSeek's unconventional approach of using distillation technology to enhance AI application opportunities is a prime example of this.

Overall, it is anticipated that changes in the international situation will prompt Chinese enterprises looking to invest in AI development to accelerate the development of their own AI chips or hardware such as High Bandwidth Memory (HBM). Although the performance of these may not match that of GPU solutions from companies like NVIDIA, their main purpose is to meet the infrastructure needs of self-use data centers in the Chinese market. In this context, the performance of a single chip is no longer the sole consideration. Moreover, players like DeepSeek are recently moving towards the development of multi-modal AI models. They strive to achieve similar performance in specific application fields at a lower training cost, thereby accelerating the process of commercialization.

The growth of the global AI Server market is a testament to the increasing importance of AI in the technological landscape. The expansion of AI Server shipments not only reflects the growing demand for AI computing power but also indicates the industry's continuous evolution. As CSPs around the world adjust their strategies, the competition in the market is becoming more intense. The development of in-house ASICs by major players is a strategic move to gain a competitive edge. By reducing reliance on external hardware suppliers, they can better control costs and customize solutions to meet their specific business needs.

DeepSeek's Model Distillation technology represents a significant innovation in the field. This technology not only addresses the challenges of high hardware costs but also opens up new possibilities for the widespread application of AI. By compressing large models, it enables faster inference on less powerful hardware, making AI more accessible to a wider range of users and applications. The open-source API strategy further encourages the development of a vibrant ecosystem, where developers can build on DeepSeek's technology and contribute to its growth.

In the Chinese AI market, the push for self-owned AI chips and supply chains is driven by multiple factors. The need for self-sufficiency in the face of potential international restrictions, as well as the desire to develop a competitive domestic AI industry, are among the key motivations. The development of more efficient algorithms, as demonstrated by DeepSeek, also showcases China's strength in software innovation. By leveraging its existing Internet infrastructure, China can bridge the gap in hardware performance and drive the development of AI applications in various sectors.

Looking ahead, the future of the AI industry is filled with both opportunities and challenges. The development of multi-modal AI models by companies like DeepSeek holds great promise for unlocking new applications and user experiences. However, ensuring the security and ethical use of AI remains a crucial concern. As the industry continues to evolve, collaboration between different stakeholders, including governments, enterprises, and research institutions, will be essential to address these issues and drive the sustainable development of AI. The balance between cost-effectiveness and performance will also be a key factor in determining the success of AI solutions in the market. Enterprises will need to constantly innovate and optimize their strategies to stay ahead in this highly competitive and rapidly changing landscape.