2025-01-29

 
Research published by Blue Lotus Research on October 18, 2023





 
(11 Pages, 17 graphs and tables)
Listen to analyst’s audio summary
  • We believe Chinese LLM (Large Language Model) will be a sovereign market, which will benefit a range of supply chain companies from semiconductor, hardware, cloud computing and application developers. Adopt of LLM will follow the pattern of cloud, with players also falling into similar categories;
  • Game development seems to be the immediate beneficiary, we raise NTES TP by 29% to US$135 and make it our TOP ST PICK. Our TOP LT PICK is SMIC. We, however, caution on unrealistic expectations in the ST.
 
LLM is a sovereign market and thus worth to invest in
A key investable theme in China Internet has been its being a sovereign market, which creates barriers to entry. We encourage investors to view LLM/AIGC as such. We believe a good reference to the competitive outcome of LLM is key word search. Like search, data security will play a big role in shaping the landscape.
 
SMIC as the only local, advanced-node, logic IC player has value
We have noted the difference between an 100%, non-US semiconductor supply chain, which will likely take a decade to realize, and a localized, advanced-node, logic IC supply chain, which can volume produce chips like Kirin 9000s today, without sanction. We believe the former only has symbolic or deterrence value. Domestic substitution of Nvidia’s AIGC chip will enjoy solid demand support, due to Nvidia AIGC chips being so expensive. There are many investment choices along China’s semiconductor supply chain, we continue to recommend SMIC based on its uniqueness in advanced-node, silicon-based, logic IC.
 
Short-term LLM commercial value is highly platform-dependent
From a technology perspective, Baidu bears the most resemblance to OpenAI but it is unlikely to reap the ST commercial benefits, for three reasons: (1) China Internet has been highly platform-dominant with poor interconnectivity of users, data and workflows, (2) Skilled developers are scanty and congregated in large platforms, (3) vertical industry knowhows are localized and opaque. These differences already accounted for differences in SaaS and cloud adoptions, it will be the same for LLM adoptions, in our opinion. We see large over small, 2B over 2C, platform over corporate sequence of LLM adoption. Exhibit 16 shows the AIGC functions under development by major applications.

Click to download the full report