Growing Demand for GPUs in Chinese AI Industry Presents Opportunities and Challenges for Electronic Components Distributors

Recently, a Chinese AI entrepreneur, speaking at a closed-door seminar of the Chinese Association for Artificial Intelligence, likened LLM to the modern steam engine, emphasizing that the US is ahead of China in terms of computing power mainly because of its GPU advantage. GPUs have unique advantages over central processing units (CPUs) in machine learning because they use synchronous computing.

Since the launch of OpenAI's ChatGPT, a chatbot developed with the support of Microsoft, there has been a race among major tech companies to develop their own version of generative AI products. However, experts warn that China faces significant obstacles in producing products like ChatGPT, which may be hindered by the review system or increasing restrictions on advanced semiconductors.

A report released today by TrendForce, a consulting firm, states that generative AI is integrated through algorithms such as GAN, CLIP, Transformer, Diffusion, and other AI technologies such as pre-training models and multimodalities. Data, computing power, and algorithms are the three essential keys for developing generative AI.

The report points out that since generative AI requires a large amount of data for training, a large number of high-performance GPUs are required to shorten the training time. For example, the training parameters for the GPT model behind ChatGPT have skyrocketed from approximately 120 million in 2018 to nearly 180 billion in 2020.

TrendForce predicts that the estimated demand for GPUs will be around 20,000, and it is expected to increase significantly in the future as generative AI continues to grow. This will also benefit the relevant supply chain. The biggest beneficiary is NVIDIA, the leading GPU chip manufacturer, with its DGX A100, which can achieve up to 5 PetaFLOPS of computing power, making it the first choice for large-scale data analysis and AI acceleration. Additionally, AMD has also launched the MI100, MI200, and MI300 series chips.