On Monday, 22V hosted a Webinar on AI investing in China and the US with special guest Kevin Xu. Kevin is the founder of Interconnected Capital, a global technology hedge fund, and author of the Interconnected Newsletter on Substack. You can find a replay of the Webinar HERE.
Some key takeaways from our wide-ranging conversation:
DeepSeek is unique. Behind DeepSeek’s breakthrough AI models is a highly idiosyncratic story. DeepSeek is an AI fund that originated from a hedge fund, with no explicit commercial mandate and – crucially – with a dedication to leveraging the open source ecosystem. US export controls on high-performance GPUs forced the company to wring out efficiencies through clever engineering at the level between hardware and software – an area that US researchers had largely ignored.
‘Open vs. Closed’ matters more than ‘US vs. China’. Kevin drew on his own background in open source software to note that what is now playing out with generative AI is a pattern familiar from operating systems (Linux), Web browsers (Netscape) and databases. Open source language models will eat away at the economic moats of OpenAI and Anthropic unless those firms decide to open their own models. There are also high geopolitical stakes, as the US risks losing a source of “soft power” influence to China if US firms do not open their own models. (Note that after the Webinar, OpenAI announced it will release an open-weight language model this summer).
Potential AI winners in China. Kevin noted that among the more intriguing AI plays in China is the corporate ecosystem built around Xiaomi, with its combination of mobile, electric vehicles and self-driving technology, and cloud (Kingsoft Cloud). Among the largest e-commerce platforms, Alibaba is a pioneer in cloud and also well-positioned. A key challenge for Alibaba and its peers is capturing cloud market outside of China, where they face stiff competition from US firms. Geopolitics will matter significantly (see next point).
Nvidia and the geopolitics of AI. For now, DeepSeek and other high-performance open source models will boost AI diffusion and thus demand for computing power from Nvidia. Chinese firms such as Huawei are seeking to capture the GPU market in China but do not have the capability to match Nvidia or the capacity yet to meet demand in China. Further, Nvidia’s rollout of the Blackwell line of GPUs will likely boost the lead of US AI firms over Chinese rivals, at least temporarily. Over the longer-term, however, chip and cloud solutions provided by Huawei and other Chinese firms could be attractive for third countries worried about the risk of technology sanctions by the US government.
What to watch next: the software layer. Kevin noted that the question keeping him up at night is what the next software layer for AI will look like, and whether it will make good use of the enormous global buildout of AI infrastructure that is just now coming online. He believes that the software “picks and shovels,” not hardware, is the most interesting part of the technology stack.
