High-Flyer, a quantitative hedge fund that amassed a substantial 100 billion yuan ($13.79 billion) portfolio leveraging AI-driven investment strategies, has announced a strategic pivot towards developing cutting-edge Artificial General Intelligence (AGI). The Hangzhou-based firm, officially known as Hangzhou Huanfang Technology Ltd Co., declared its commitment to advancing AI for the benefit of humanity in a recent WeChat post.
Table Content:
From AI-Powered Investments to AGI Research
High-Flyer’s ambitious goal is to create a new independent research group dedicated to exploring the core principles of AGI. This move aligns with the vision of industry leaders like OpenAI, which defines AGI as autonomous systems capable of surpassing human capabilities in most economically valuable tasks. While OpenAI CEO Sam Altman recently acknowledged that achieving true AGI remains elusive, High-Flyer’s commitment signifies a significant investment in this frontier field.
Central to High-Flyer’s AGI pursuit is DeepSeek, an independent research group led by High-Flyer’s founder and controlling shareholder, Liang Wenfeng. DeepSeek’s AI models have recently garnered global recognition, challenging established players in the technology sector. The company’s claim of achieving comparable results with significantly less computing power than leading U.S. firms sparked a notable sell-off in tech stocks worldwide.
A Decade-Long Journey to AI Leadership
DeepSeek’s rapid ascent may appear sudden, but High-Flyer’s trajectory reveals a decade-long commitment to AI innovation. Under Liang’s guidance, the fund dedicated years to studying and implementing overseas AI models, integrating them into its investment strategies. A crucial element of this approach involved substantial investments in high-end Nvidia chips to provide the necessary computing power.
Building AI Supercomputing Infrastructure
High-Flyer’s commitment to AI is exemplified by its construction of two AI supercomputing clusters, exclusively utilizing Nvidia’s powerful A100 chips. These clusters were built and operational prior to the U.S. imposing export restrictions on these chips to China in September 2022. The first cluster, comprising 1,100 A100 chips and costing 200 million yuan, became operational in 2020. A year later, a second, larger cluster with around 10,000 A100 chips was completed at a cost of 1 billion yuan. In 2022, High-Flyer researchers presented a strategy at an Nvidia conference outlining how they maximized the efficiency of the second cluster for AI model training.
DeepSeek’s Computing Power and the Future of AGI
While DeepSeek claims to utilize Nvidia’s less powerful H800 and H20 chips for training its DeepSeek-V3 and its predecessor DeepSeek-V2 models, speculation persists regarding its actual computing capabilities. Scale AI CEO Alexandr Wang recently alleged, without providing evidence, that DeepSeek possesses 50,000 undisclosed Nvidia H100 chips, potentially violating U.S. export controls. DeepSeek and Nvidia have not commented on this allegation.
Liang’s emphasis on computing power when discussing DeepSeek’s future resonates with High-Flyer’s prior investments in AI infrastructure. In a July interview, Liang addressed the possibility of DeepSeek’s public listing, stating that funding was not a constraint but rather the embargo on high-end chips posed the primary challenge. This statement underscores the critical role of advanced computing resources in the pursuit of AGI.
Conclusion: A Bold Bet on the Future of AI
High-Flyer’s strategic shift from quantitative investing to AGI research marks a significant development in the AI landscape. The company’s prior investments in AI infrastructure and DeepSeek’s impressive achievements position it as a potential leader in the global race towards AGI. While the path to AGI remains challenging and filled with uncertainties, High-Flyer’s commitment and resources signify a bold bet on the transformative potential of this technology.
