Big models explode with GPU supply shortages Rise in demand for algorithmic networks
With the explosion of ChatGPT and the intensive launch of large domestic AI models, the role of GPU chips, which provide the arithmetic foundation for model training, has become increasingly prominent. It is understood that many domestic GPU IP licenses are purchased from third parties, the advantage is that due to the use of mature technology can be quickly put into commercial use, but the disadvantage is that it is difficult to form a mature graphics card-level GPU, the core circuit patents can not be self-control and independent iteration. GPU has fallen into a supply shortage, a number of GPU prices from March and April so far increased by nearly 50%, orders are scheduled to the first half of next year.
The development of large models and related applications will also catalyze an explosion in demand for arithmetic power. However, in addition to large Internet companies and leading AI companies with more GPU arithmetic chip reserves, small and medium-sized enterprises encounter arithmetic bottlenecks in the development of AI models and applications. Some analysis points out that arithmetic leasing has become the optimal solution for the majority of SMEs to solve their arithmetic needs.
Among listed companies, Capital Online (300846) has taken the lead in establishing a large-scale, high-coverage GPU arithmetic resource pool, and has built a 10,000-chip GPU resource pool for graphics processing in 2022. Leon Technology (300603) has invested in the first phase of the Leon Cloud Data (Chengdu Jinyang) One Base project in Jinyang, Chengdu, which will build a supercomputing innovation application ecosystem of "computing power + data + algorithm" in the future.
The development of large models and related applications will also catalyze an explosion in demand for arithmetic power. However, in addition to large Internet companies and leading AI companies with more GPU arithmetic chip reserves, small and medium-sized enterprises encounter arithmetic bottlenecks in the development of AI models and applications. Some analysis points out that arithmetic leasing has become the optimal solution for the majority of SMEs to solve their arithmetic needs.
Among listed companies, Capital Online (300846) has taken the lead in establishing a large-scale, high-coverage GPU arithmetic resource pool, and has built a 10,000-chip GPU resource pool for graphics processing in 2022. Leon Technology (300603) has invested in the first phase of the Leon Cloud Data (Chengdu Jinyang) One Base project in Jinyang, Chengdu, which will build a supercomputing innovation application ecosystem of "computing power + data + algorithm" in the future.
