DeepSeek's $6M AI breakthrough challenges Silicon Valley's dominance

Two years after its founding, a Chinese startup called DeepSeek topped the iOS app store this week and sent markets into a spin—not because of flashy marketing or venture capital firepower, but because of a claim that upended the economics of artificial intelligence development. The company says it trained an AI model competitive with OpenAI's ChatGPT for less than $6 million. Microsoft, by contrast, is planning to spend around $80 billion on AI infrastructure this year. The gap between those numbers has prompted talk of a Sputnik moment in the global race for AI dominance.

DeepSeek is the creation of Liang Wenfeng, a hedge fund entrepreneur from Guangdong province in southern China who founded the company in May 2023. He is not a household name in Silicon Valley, and neither was his startup until this week. What sets him apart is his stated philosophy: he has said his aim is neither to lose money nor to make huge profits, but to be at the forefront of technology and advance the entire ecosystem. That framing—research first, commerce second—appears to have shaped how DeepSeek operates in ways that differ sharply from the venture-backed model that dominates American AI development.

Liang's ability to fund this research-focused operation comes from High-Flyer Capital, a hedge fund he started in 2015. After studying electronic information engineering at Zhejiang University, he built a trading operation that used AI to spot patterns in stock prices. By 2021, the fund's assets under management had surpassed 100 billion yuan, or roughly $110 million. That same year, rumors began circulating that Liang had accumulated an unusual collection of Nvidia graphics processing units—the specialized chips essential for training large AI models. By 2021, he was reported to own around 10,000 of them, ostensibly as a personal hobby. Only a handful of China's largest tech firms have comparable stockpiles.

The timing of that GPU accumulation proved consequential. In 2022, the Biden administration imposed sweeping export controls on semiconductors destined for China, explicitly designed to prevent the country from accessing the equipment needed for rapid AI development. Nvidia's most powerful H100 chip was banned outright. The company created a less capable H800 variant for the Chinese market, and DeepSeek trained its model on those constrained chips—though even the H800 was subsequently banned in 2023. Liang has acknowledged that semiconductor restrictions represent his company's biggest challenge, not funding.

DeepSeek's success suggests that Chinese researchers have found ways around those restrictions, a development that may prompt the US to tighten controls further or force a reckoning with whether such controls can actually work. The company is based in Hangzhou, where Alibaba is also headquartered, and Liang is reported to prefer hiring local talent over workers educated abroad. He has remained personally involved in the company's research direction. The startup is also known to pay aggressively for top engineers—job listings on Chinese recruitment sites advertise salaries up to 60,000 yuan per month, well above the roughly 35,000 yuan average at major tech firms like Tencent.

Some skeptics have questioned DeepSeek's claims about its efficiency. Alexandr Wang, an American AI entrepreneur, suggested in a CNBC interview that DeepSeek likely possesses a stockpile of advanced chips it has not publicly disclosed, given US sanctions. DeepSeek declined to comment when approached. Other analysts, however, have argued that the company's constraints may have forced genuine innovation. Karen Hao, an AI journalist, wrote that DeepSeek succeeded because of its limitations, not in spite of them.

The broader implication is that the US technological lead in AI, traditionally secured by dominance in chip manufacturing and top-tier university talent, may be narrowing faster than expected. China analyst Bill Bishop warned that regardless of how the Trump administration handles export controls, the result will likely deepen the divide over AI between the US and its allies on one side and the rest of the world, particularly the Global South, on the other. As for Liang himself, he has largely stayed out of public view. A childhood friend told the Chinese press that when congratulated on the breakthrough, Liang said he planned to return to his hometown for lunar new year—and that he would be hiding.