Panoramic Survey and Development Strategy Research of China's Artificial Intelligence Industry
-
This year, China's artificial intelligence industry has developed rapidly, with multiple large-scale AI model products including Baichuan being launched successively. AI technology, especially large model technology, will not only reshape existing industries but also create entirely new value.
China's AI large models are flourishing, and industrial implementation has moved beyond the 'novelty phase.' The practical effectiveness of industrial implementation has become a crucial evaluation dimension for the value of large models.
Generative AI poses significant security challenges. If search engines can be considered limited liability companies, AI large model products are akin to unlimited liability companies. The former produces recorded and traceable outputs, while the latter generates content independently, which can lead to risks such as 'confidently incorrect' outputs.
In response to these challenges, China has initiated relevant explorations, strengthening theoretical research in technology, continuously building the third generation of AI, enhancing technical infrastructure, improving the quality of training data and security evaluation capabilities, and achieving practical results in areas like AI security testing grounds.
Panoramic Survey and Development Strategy Research of China's Artificial Intelligence Industry
Recently, investment bank JPMorgan stated in its investment report that NVIDIA is expected to capture up to 60% of the artificial intelligence (AI) product market share this year, leveraging its GPU and networking hardware products.
It is reported that due to cyclical slowdown in its gaming division, the company's revenue for Q1 FY2024 decreased by 13% year-over-year to $7.19 billion. However, during the same period, its data center business revenue reached a record $4.28 billion, a 14% YoY increase, accounting for 60% of total revenue. Gaming revenue was $2.24 billion, down 38% YoY, representing 31% of total revenue.
Currently, NVIDIA leads the AI field, holding approximately 80% of the AI processor market share. Its high-end processors are already being used to train and run various chatbots. The company is highly favored by investors as a key supplier meeting AI computing demands.
The AI industry chain typically consists of upstream data and computing power layers, midstream algorithm layers, and downstream application layers. Recently, the market has focused more on the upstream industrial chain, particularly the computing power sector. Many new investment opportunities have emerged in AI hardware, as AI software applications rely on hardware computing power.
Domestic AI Computing Demand Will Maintain Growth Momentum
With continuous catalysis from ChatGPT, domestic AI computing demand is expected to sustain growth momentum, benefiting computing server manufacturers. Estimates suggest ChatGPT's total computing power requires 7-8 data centers with 500P computing power and ¥3 billion investment each to operate. In the digital economy era, global data volume and computing scale will show rapid growth trends.
With the simultaneous surge in demand for AI servers and AI chips, it is projected that the shipment volume of AI servers (including those equipped with GPUs, FPGAs, ASICs, and other main chips) will approach 1.2 million units in 2023, representing a year-on-year increase of 38.4%. This accounts for nearly 9% of total server shipments, and by 2026, this proportion is expected to rise further to 15%. The institution has also revised the compound annual growth rate (CAGR) for AI server shipments from 2022 to 2026 to 22%, while the shipment volume of AI chips is anticipated to grow by 46% in 2023.
The institution noted that NVIDIA GPUs have become the mainstream chips in AI servers, holding a market share of approximately 60-70%, followed by ASIC chips independently developed by cloud computing providers, which account for over 20% of the market.
Compared to general-purpose servers, AI servers utilize multiple accelerator cards, and their printed circuit boards (PCBs) adopt high-layer HDI structures, resulting in higher value. Additionally, the number of layers in the motherboard is significantly higher than that of general-purpose servers. The PCB value of an AI server is 5-6 times that of a standard server.
During his speech at NVIDIA Computex 2023, NVIDIA founder and CEO Jensen Huang announced that the generative AI engine NVIDIA DGX GH200 has now entered mass production. From the demonstration, it can be observed that the newly launched GH200 server architecture has undergone significant changes compared to the DGX H100. The PCB changes in the GH200 relative to the DGX H100 include the reduction of one UBB and one CPU motherboard, while adding three NVLink module boards. At the same time, the performance of the accelerator cards has been substantially improved, and the per-unit PCB value is expected to increase. This indicates that AI upgrades will continue to drive value growth in the PCB sector.
Computing power resources serve as a crucial foundation for the development of the digital economy. The emergence of new digital phenomena, business models, and patterns has driven application scenarios toward diversification, while the continuous expansion of computing power scale has led to a sustained increase in demand. According to data released by the Ministry of Industry and Information Technology, the total scale of operational data center racks nationwide exceeded 6.5 million standard racks in 2022. Over the past five years, the average annual growth rate of total computing power has exceeded 25%. As computing power is applied across various industries, different levels of precision in computing power need to "adapt" to diverse application scenarios. Particularly with the rapid advancement of artificial intelligence technology, the structure of computing power has evolved accordingly, leading to a growing demand for intelligent computing power.
From a policy perspective, China places high importance on the development of the AI industry, gradually solidifying the foundation for intelligent computing power. In February 2022, four ministries and commissions jointly issued a notice approving the launch of national computing power hub nodes in eight regions and planning 10 national data center clusters. By this point, the overall layout design of the national integrated data center system was completed. With the full implementation of the "East Data West Computing" project, the construction of intelligent computing centers has also entered a new phase of accelerated development. Data centers, as hubs for data and carriers of applications, form the foundation for AI development. In the long term, the demand for data centers is expected to recover. It is projected that the IDC market size will reach 612.3 billion yuan by 2024, with a compound annual growth rate of 15.9% from 2022 to 2024, marking the beginning of a new upward cycle for data centers.
The Future Direction of Artificial Intelligence
Artificial intelligence encompasses many concepts, some of which are difficult to measure and verify. For example, enabling machines to understand what society is or what responsibility means—while they may output representations, it is challenging to verify whether the machines truly comprehend these concepts. Therefore, it is possible to create a closed loop around verifiable and measurable concepts. Embodied intelligence happens to be such a closed loop, serving as an excellent starting point toward general intelligence.
The rapidly advancing large AI models hold the potential to break through limitations and endow robots with "intelligence".
Robotic large models encompass LLM (Large Language Models), VLM (Vision-Language Models), and VNM (Visual Navigation Models). The "brain" of robots in the AI domain is not limited to the language models used in ChatGPT. Google's LM-Nav research demonstrates that combining LLM, VLM, and VNM can translate from natural language (redundant verbal descriptions) to text (landmark strings) to images (locating objects in images based on text), ultimately generating path planning for robots. Based on this behavioral model, robots can engage in human-machine interaction while achieving a certain degree of "improvisation".
Recently, Professor Lu Cewu from Shanghai Jiao Tong University delivered a keynote speech titled "Embodied Intelligence" at the Machine Heart AI Technology Annual Conference, proposing the PIE solution. He suggested that embodied intelligence consists of three modules: embodied Perception, embodied Imagination, and embodied Execution, which could accelerate the practical implementation of embodied intelligence.
Currently, AI-powered robots may represent the current focal point for "embodied intelligence".
Compared to non-intelligent ordinary humanoid robots, embodied intelligence offers higher work efficiency. Its capabilities in comprehension, interaction, and planning make it highly practical as robots penetrate various industries. Additionally, its natural language control feature is essential for future large-scale assistance to ordinary workers.
Therefore, future attention should focus on the types of hardware robots and application scenarios that can be transformed using currently available large models, such as service robots primarily for conversation, industrial robots, and humanoid robots in complex scenarios.
Many major companies have already made arrangements in the field of embodied intelligence. Google has released PaLM-E, the largest generalist model in history; Microsoft is exploring how to extend ChatGPT to the robotics field; Alibaba's Qwen large model is experimenting with connecting to industrial robots, among others.