The quantitative hedge background High-Flyerwhich is behind the emergence of DeepSeekbuilt a portfolio of 100 billion yuans ($ 13.79 billion) using artificial intelligence models to make investment decisions, but by 2023 decided to change course to focus on the development of the most advanced AI.
In a publication on its official Wechat account, Hangzhou Huanfang Technology Ltd Co. – as the company is officially called – said it would focus on the search for General Artificial Intelligence (AGI).
“High-Flyer will concentrate its resources and strength, will completely dedicate itself to serving AI technology that benefits all humanity, will create a new independent research group and explore the essence of AGE,” the company said.
OpenAi, supported by Microsoft, which has developed ChatgPT, defines AG as autonomous systems that surpass humans in most economically valuable tasks.
It is the next generation of AI models and, in an X post last week, OpenAi CEO Sam Altman said his company had not yet reached this milestone.
DeepSeek
The independent research group planned by High-Flyer was Deepseek, whose models have shaken the global technology industry in recent weeks. High-Flyer’s founder and shareholder is also the discreet leader of.
Continues after advertising
The sophistication of DeepSek models was widely praised by its Silicon Valley competitors, a novelty for a Chinese AI model, but the allegations of the startup that used a fraction of computing power implemented by major US companies for their own models triggered a liquidation of technology actions around the world.
It is not clear how close DeepSeek is developing an AGE model.
Although DeepSeek’s success seems to have happened almost overnight, High-Flyer shows how this meteoric rise took more than a decade to happen.
Continues after advertising
Under Liang’s leadership, the fund spent years studying and experimenting with AI models abroad, applying this technology to its business and investing tens of millions of dollars in cutting -edge nvidia chips to provide the necessary computer power to support this AI-centered strategy, according to an analysis by the Reuters of High-Flyer websites and official Wechat accounts.
Supercomputation clusters
The strategy includes the construction of two AI supercomputation clusters, composed entirely by the powerful NVIDIA A100 chips, whose export to China was banned by Washington in September 2022.
High-Flyer clusters were built and placed in operation long before export controls were announced. The first cluster, consisting of 1,100 A100 chips, cost 200 million yuans and was placed in operation in 2020, while the second, consisting of about 10,000 A100 chips, was completed a year later at a cost of 1 billion yuan, from according to the company’s website and several posts on Wechat.
Continues after advertising
In 2022, High-Flyer researchers had a strategy at a NVIDIA conference that the company developed to maximize the efficiency of the second cluster when training AI models.
Nebula
It is unclear how High-Flyer has invested in Deepseek. High-Flyer has an office located in the same DeepSek building and has patents related to used chip clusters to train AI models.
Liang has a 55% stake in High-Flyer, closed capital, and holds 99% of voting rights, according to Chinese corporate records. The remaining actions are detained by other Fund executives.
Continues after advertising
So far, DeepSeek has only claimed to use Nvidia’s H800 and H20 chips to train its Deepseek-V3 model and its predecessor DeepSeek-V2, which triggered a war modeling price in China when it was released Last May.
However, some technology executives have publicly stated that Deepseek has much more computational power at its disposal.
Scale CEO Ai, Alexandr Wang, said during an interview with CNBC On Thursday, without evidence, that DeepSeek has 50,000 Nvidia H100 chips, which he claimed not to have been released because he would violate Washington’s export controls. The United States prohibited export of H100 chips to China while vetoing the less powerful A100.
DeepSeek did not respond to a request for comment Reuters About the claim. Nvidia also did not respond immediately to an email asking for comment.
But Liang’s concern for computer power, when discussing the future of DeepSek, echoes the massive investment of its quantitative fund in AI clusters. Asked in an interview with Chinese Waves last July if High-Flyer planned to separate DeepSeek from the company and make it public, Liang replied:
“We have no plans to raise money in the short term. The problem we faced has never been money, but the edge of cutting -edge chips. ”
