Tiangong's big model continues to iterate, and AI music is being implemented at an accelerated pace
The company's “Tiangong Big Model” 3.0 will be released, and “Tiangong SkyMusic” has started closed beta. Currently, the company has built AI business matrices such as AI big models, AI search, AI music, AI stories, and AI games based on the self-developed “Tiangong” series of base models. At the same time, the company is promoting the research and development of AI computing power chips through capital increases, etc., and lays out the entire AI industry chain. We believe that the “Tiangong” model has a stable foundation and continues to iterate and launch AI-related applications around the model, which is expected to establish long-term competitiveness on the artificial intelligence circuit. We maintain the company's profit forecast of achieving net profit of 8.8/10.4/1.20 billion yuan in 23-25. The segmented valuation method calculates a target market value of 57.34 billion yuan, corresponding to a target price of 47.19 yuan (original value: 45.90 yuan), maintaining a “buy” rating.
“Tiangong Big Model” 3.0 will be released, the number of Tiangong AI application users is growing rapidly, and the big language model “Tiangong 1.0” of the Kunlun Tiangong Model will be officially released on April 17, 2023; on April 17, 2024, “Tiangong 3.0” will officially begin public testing. It uses a 400 billion parameter MoE hybrid expert model, and will simultaneously select open source. The C-end application “Tiangong AI Intelligent Assistant” supports functions such as AI search/creation. According to QM, as of March 31, the weekly MAU for Tiangong AI assistants (March 25 to March 31) was 5.326 million, an increase of 82.3% from 2,921 million at the beginning of the month (February 26 to March 3). According to point data, in the Iphone “efficiency” app list, the number of downloads of the “Tiangong” app quickly rose from 26th place on March 1 to 3rd place on March 30. Since then, it has declined somewhat, ranking 12th as of April 9.
“Tiangong SkyMusic” has started closed beta. Using the self-developed Sora model architecture, the “Tiangong SkyMusic” music model began closed beta on April 2 and will be officially released on April 17. “Tiangong SkyMusic” is based on the “Tiangong 3.0” large model base. It is currently the only publicly available AI music generation large model in China. It uses a Sora model architecture in the field of music audio, and supports the generation of dual-channel stereo songs with a sample rate of 44,100 Hz in 80 seconds. Currently, “Tiangong SkyMusic” has five major features: 1) generating 80s high-quality AI music, 2) leading vocal synthesis level; 3) controlling songs through lyric passages; 4) supporting various music styles such as rap; and 5) being able to learn singing techniques such as vibrato. It is expected that subsequent major music models will continue to be upgraded and optimized, so that the model has multi-modal emotional understanding and expression capabilities.
Opera continues to innovate AI and upgrade AI server clusters
Opera, a subsidiary of the company, continues to use AI as the core of product development and continues to innovate in the field of AI browsers. Currently, all of Opera's flagship browsers are equipped with native AI features, including Aria Browser AI. Opera deployed a new AI server cluster in Iceland, equipped with Nvidia's DGX SuperPod AI data center solution, equipped with the latest NVIDIA H100 TensorCore GPU and NVIDIA AI Enterprise software platform to support larger and more complex AI workloads.
Risk warning: AI development falls short of expectations, model development progress falls short of expectations, industry policy risks.