CITIC Securities: Decoding OpenAI Founder Sam Altman's AI Investment Map

According to the information from the Wise Finance APP, CITIC Securities released a research report stating that Sam Altman, the founder of OpenAI, focuses his AI investments on: horizontally expanding multimodal and cost-effective small models, vertically laying out upstream basics such as AI chips, nuclear energy, as well as humanoid robots, cameras, and other AI terminals, in addition to enterprise services, efficiency tools, financial technology, education and training, and healthcare as the top five application directions for Altman's layout. This reflects the investment opportunities in China's AI development, mainly including: horizontally focusing on companies with ample cash reserves and early implementation of cost-effective models; upstream focusing on domestic alternatives for AI chips, databases, as well as new infrastructure developments like nuclear power and AIDC; downstream focusing on the intelligent terminals of humanoid robots, cameras, and applications in fields such as enterprise services, finance, education, etc. Overview of Altman's Investment Portfolio: Altman has invested in over 400 projects, with AI being his key investment area. The rapid development of generative AI has attracted high global investments, with the United States in an absolute leading position. According to Stanford, by 2023, private investment in AI in the United States is 8.7 times that of China. OpenAI, an American company leading the global development of generative AI, with its founder Sam Altman also being an outstanding investor. According to Wall Street, as of early 2024, Altman and his venture capital fund have invested in over 400 companies, with the value of their holdings exceeding $2.8 billion, with AI being their key investment area. According to CB Insights, SVTR AI, various company websites, and the OpenAI Startup Fund website, Altman and his investment platform have a total of about 92 investment projects since 2019, with 82 of them related to AI. Among the 82 AI-related investment projects, 17 are related to AI infrastructure, 3 are related to AI terminals, and 62 are related to AI applications. Summary of Altman's Investment Direction: Horizontally expanding multimodal and small models, vertically laying out the AI industry upstream and downstream. Horizontal Expansion: ① Multimodal Products: Since the launch of GPT in 2018, OpenAI has continued to expand its layout with products such as DALL·E for text-to-image, Sora for text-to-video, and multimodal GPT-4o, which are expected to accelerate the practical implementation of AI in complex real-world scenarios. ② Cost-effective Small Models: In July 2024, OpenAI launched the GPT-4o mini, a small model with a parameter scale of about 7 billion, achieving higher cost-effectiveness by reducing model size and performance. ③ New Inference Model: On September 12, 2024, the o1 model was launched, improving performance by introducing internal thought chains and complex reasoning processes. Upstream: Layout of AI infrastructure. In the context of chip shortages, significant resource consumption like electricity, and the increasing importance of data, Altman has invested in important targets such as AI chip company Rain Neuromorphics, nuclear fusion companies Oklo and Helion Energy, real-time analytics database Rockset, to explore AI infrastructure solutions Downstream: Layout of AI terminals and applications.① In terms of terminals, Ultraman has invested in AI Pin company Humane, humanoid robot company Figure AI, high-end network camera company Opal, mainly focusing on human-machine interaction, labor substitution, and household consumption areas. ② In terms of applications, enterprise services, efficiency tools, financial technology, education and training, and healthcare are the top five application directions Ultraman is focusing on, with investments in companies such as Induced AI (workflow automation), qqbot (programming development tools), Slope (B2B payment platform), etc. The Chinese mapping of Ultraman's investments: Intense competition in domestic models, steadily advancing upstream infrastructure and downstream applications. Horizontal: Intense competition among domestic models, unicorn companies have emerged. China currently has 117 registered domestic large models (Cyberspace Administration, as of March 2024), among which Zhipu, Dark Side of the Moon, Zero One Everything, Bai Chuan Intelligence, Minimax are the "Five Tigers of Large Models", with Douyin's "Bean Bag" traffic performance leading significantly. Upstream infrastructure: Actively building computing infrastructure, accelerating domestic substitution. AI-driven domestic cloud vendors' capital expenditures have increased significantly, with a 124% year-on-year increase in total capital expenditures by BAT in 2024H1, reflecting the growing demand for infrastructure. Domestic substitution of AI chips, databases is accelerating; progress in basic infrastructure such as AIDC, nuclear power is speeding up. Downstream terminals and applications: ① In terms of terminals, the capital market enthusiasm for humanoid robots is rising, with a financing amount of 5.5 billion yuan in 23 years (IT Orange), products such as Yushu G-1, Zhiyuan Expedition A2 gradually going online; Omedia predicts that the penetration rate of intelligent cameras in China is expected to increase from 15% to 63% from 2019 to 2024. ② In terms of applications, Beisen Holdings launched "Beisen AI Family", Alibaba launched the "Tongyi" series, Tonghuashun launched "Wencai HithinkGPT Large Model", etc., domestic AI applications are actively advancing. Risk factors: Risks of technological innovation falling short of expectations; Risks of commercialization of AI applications progressing slower than expected; Risks of intensified competition in hot industry segments; Risks of policy support falling short of expectations; Risks of macroeconomic fluctuations leading to lower-than-expected downstream demand; Risks of intensified international conflicts restricting domestic AI development; Risks of potential ethical, moral, and user privacy risks in AI