The Man Behind the Chinese Hedge Fund That Spooked Silicon Valley With Its DeepSeek AI Platform

Published: 1/29/2025, 12:55:05 AM EST
The Man Behind the Chinese Hedge Fund That Spooked Silicon Valley With Its DeepSeek AI Platform
Screens display the DeepSeek logo in Toulouse, southwestern France, on Jan. 29, 2025. (Lionel Bonaventure/AFP via Getty Images)
News Analysis

This past weekend saw a major development in China’s drive to out-compete the United States in information technology.

High-Flyer, a mainland China-based hedge fund that few in the United States have even heard of, recently announced the commercial launch of its artificial intelligence platform, DeepSeek. Within less than 72 hours of its release, DeepSeek surpassed every other large language model (LLM) artificial intelligence (AI) platform in American app stores. Observers were shocked that DeepSeek’s R1 platform had come out of nowhere to garner such a commanding position.

High-Flyer founder Liang Wenfeng, 40, studied machine vision at Zhejiang University and was invested in advanced technologies at his 10-year-old multi-billion-dollar quant fund in China’s Hangzhou. Unlike other tech startups, Liang’s appears to have no connection with American start-ups or venture capital firms. DeepSeek’s creation is unlikely to be due to IP theft, and it’s not funded by American capital like TikTok’s ByteDance, which had Silicon Valley backing.

High-Flyer funded and created DeepSeek to build an AI platform for stock trading–called Firefly–in 2019. Under Liang’s direction and after years of acquiring thousands of high-performance Nvidia A100 advanced graphics processing units, or GPUs, the company got into the LLM business.

Upon its release, well-known Silicon Valley AI investor Marc Andreessen immediately lauded DeepSeek R1 as an “amazing and impressive breakthrough” in a post on X. Other observers said it was as good as or better than the leading OpenAI platform, ChatGPT.

DeepSeek’s R1 platform is free of charge and open-source, meaning that software developers can access the code and add to it, making modifications that suit their purposes. ChatGPT is not open source and has a monthly fee to use its most advanced features. DeepSeek potentially undercuts that subscription model, making China-developed AI cheaper to use at present.

After its successful launch and the explosion in user numbers, DeepSeek faced cyberattacks all day on Monday.

Experts have said that Washington’s export controls on advanced microchips like those made by Nvidia were designed so a company like DeepSeek could not simply pop up out of nowhere. These restrictions forced Chinese companies to find ways around the bans, and to do more work with less.

“Singapore buys a lot from Nvidia, and they are selling it to China,” said Christopher Balding, a former associate professor at Peking University’s HSBC Business School in China’s Shenzhen. “Singapore buys more advanced chips than Europe and Japan. Everyone knows that when there is shady stuff to go down in Asia, there is a good chance it is going through Singapore.”

Nvidia reported Singapore as its second-largest market after the United States in the third quarter of 2024, at $7.6 billion, against sales to China of $5.4 billion. Singapore is also building a large number of data centers.

DeepSeek’s instant success comes from its decision to take a handful of known search processes used in AI and place them under one umbrella. In simple terms, they achieved more with less computing power, making DeepSeek R1 more efficient and less costly than its American rivals.

It is unclear what Washington will do, or can do, to thwart China’s use of American-made microchips to build the supercomputing capacity AI machines need.

President Donald Trump said on Monday that, “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser-focused on competing to win.” He also congratulated the startup.

Liang said in an interview published in the WeChat publication “Waves” on May 23, 2023, that High-Flyer imported over 10,000 Nvidia H800 GPUs in 2021. These chips have long faced export restrictions by Washington, but some new bans were ordered by President Joe Biden in his last days in office.

He said High-Flyer began buying the GPUs in 2015, starting with 100 of them, before moving to 1,000 in 2019, and 10,000 by 2021–enough to build DeepSeek R1, the Chinese rival to OpenAI’s GPT-4.

“GPT-4 still has many mysteries. We will research on them at the same time as we replicate GPT-4,” Liang told Waves. He said High-Flyer is still funding the project, but that he was now looking for other funders. In the interview, Liang described DeepSeek as a passion project, often saying High-Flyer’s tech team was doing all of this work out of curiosity.

China made AI part of its Made in China 2025 goals. DeepSeek is privately held, but it has helped Beijing reach its goal of becoming the United States’ peer competitor in artificial intelligence.

“I would not say they are on par yet,” said Balding. “China has a lot of very smart people, and they do good things over there, make no mistake about that. But the reality is that the U.S. is still the leader.”

Liang could choose to spin DeepSeek off from High-Flyer. He said the next step is artificial general intelligence, or AGI, which would signify the move from supercomputing to human-like intelligence–potentially one of the most disruptive technologies in history.