Morgan Stanley: What does DeepSeek mean for China's data center and software industry?

Recently, the rise of the open-source large language model DeepSeek R1 has attracted industry attention. Morgan Stanley stated that this model achieves comparable performance at a training cost far lower than that of GPT, which may have a profound impact on China's data center and software industry.

On January 27, Morgan Stanley analysts Yang Liu, Tom Tang, and Lydia Lin released a report stating that the emergence of DeepSeek R1 has sparked discussions about the necessity of using large-scale high-end GPU clusters to train large language models (LLM). As an open-source project, its training cost is only 1/30 of that of GPT, yet it can achieve performance levels comparable to GPT.

Morgan Stanley believes that the impact of DeepSeek R1 on China's data centers may be as follows:

• In the short term, if large technology companies begin to adopt technology routes similar to DeepSeek, it may reduce the demand for AI model training, especially in data centers in remote areas.

• However, in the long run, low-cost models may drive growth in inference demand, which is positive for data center demand in first-tier cities.

On the other hand, regarding China's software industry, Morgan Stanley believes that the impact of DeepSeek R1 may bring a selective slight benefit. This is because the reduction in AI model costs may lower the threshold for applications to run AI functions, improving the industry environment from the supply side. However, this does not directly imply better monetization capabilities, as the biggest issue in the software industry still lies on the demand side.

Morgan Stanley believes that the ultimate impact will depend on specific software products and their downstream markets