The American AI sector has a new kid on the block to contend with, as Chinese startup DeepSeek has topped rankings on app stores in the UK, the US, and China after launching its R1 reasoning model last week.
DeepSeek is an AI startup from China founded by Liang Wenfeng, a computer science graduate of Zhejiang University. In 2015 he and fellow graduates from his class including Lu Zhengzhe founded High-Flyer, a hedge fund based in Hangzhou (capital of the Zhejiang region) managing over 50 billion yuan ($7 billion) that has employed powerful computer clusters and AI to exploit market opportunities. The fund has achieved 13% annualized returns since 2017.
It is difficult to overstate how integral AI has been to High-Flyer’s investment strategy. The hedge fund’s technology “underpins every stage of its operation—from data collection and analysis to portfolio construction and trade execution,” according to Zichen Wang, former Director for International Communications at the Center for China and Globalization.
While the algorithmic AI-driven investment strategy was not foolproof, as the high and mid-frequency trading strategy led to unexpected losses at High-Flyer in 2021 and 2024, it would appear that the know-how accrued by relying so heavily on technology in that venture has paid dividends with DeepSeek, which was started in May 2023. In November 2024, the company unveiled a feature separating its AI from competitors that shows the logic leading to the answers it provides – essentially, their AI fact-checks itself. While this means that it spends more time crafting its response than competing models, it all but eliminates the “hallucinations” (i.e. inaccurate information) that bedevil other models like ChatGPT. VentureBeat reported at the time that this model succeeded where its competitors failed in answering “trick” questions.
Just as impressive is that Wenfeng’s AI is apparently running much leaner than its US-based competitors in a number of ways. The startup claims that the training for one of its latest models cost $5.6 million, a relative pittance compared to American competitors’ declared costs in the nine to ten-figure range. It’s also using cheaper technology by necessity, as the highest performing chips are kept out of China due to U.S. sanctions. While Wenfeng has acknowledged challenges posed by the embargo, the MIT Technology Review writes that the measures have engendered a culture at DeepSeek and other Chinese ventures to “innovate in ways that prioritize efficiency, resource-pooling, and collaboration.” Business Insider explains that the startup also employs an efficient architecture that its competitors are only just coming around to, dubbed a “mixture of experts,” which reportedly makes use of built-in expertise that goes into action or lies dormant depending on the query.
The efficiencies and results declared by DeepSeek are so impressive that some are doubting their authenticity. Neal Khosla, CEO of the AI health platform Curai, claims that the Chinese startup is “faking the cost was low to justify setting price low and hoping everyone switches to it [to] damage AI competitiveness in the US.” Analysts at Citi are similarly skeptical, saying that they “question the notion that its feats were done without the use of advanced GPUs to fine tune it and/or build the underlying LLMs the final model is based on.”
Only time will tell if DeepSeek’s tech is as good as they claim, but in the meantime, Silicon Valley is rattled. As of this writing, investors have dumped over $1 trillion of AI-related technology stocks in just one day of trading.
