According to the latest research by TrendForce, the development of HBM technology is driven by the demand for AI Servers, and the three major original equipment manufacturers are actively promoting the progress of HBM4 products. Due to the increase in the number of I/O (input/output interface) of HBM4, the complex chip design has increased the wafer area, and some suppliers have changed their products to logic chip architecture to improve performance, which has pushed up costs. Given that the premium ratio of HBM3e was about 20% when it was first launched, it is expected that the premium of HBM4, which is more difficult to manufacture, will exceed 30%. NVIDIA, a leading AI chip manufacturer, unveiled its latest Rubin GPU at this year's GTC conference, and AMD has MI400 to compete with it. All of the above products will be equipped with HBM4. According to TrendForce's analysis, compared with previous generations of products, the number of I/Os of HBM4 has doubled from 1024 to 2048, and the data transmission rate has remained above 8.0Gbps, which is comparable to HBM3e. This means that at the same transmission speed, the amount of data transmitted by HBM4 with a higher number of channels will double.