What is DeepSeek, and Why It’s Time for VCs to Rethink Their Investment Strategies

By now, you’ve likely heard or read a headline about Deepseek, a Chinese artificial intelligence company that has all of Silicon Valley talking. If you were curious enough to read the article, you know that this scrappy startup has managed to match—and, in some cases, outperform—popular U.S. AI companies in head-to-head competitions, with its model, DeepSeek R1, going up against OpenAI’s o1 and Meta’s LLaMA 3.3.

That it has beaten U.S. AI companies is not as surprising as how it beat these companies, and how it has beaten the best companies has caused everyone from investors to lawmakers to turn their heads. It has caused AI chipmaker Nvidia's stocks to drop 18% in one day, and it has caused a frenzy of other AI startups, Chinese or otherwise, to reimagine what is possible for the future. While I won't dive into the details of how DeepSeek has managed to cause OpenAI to be in deep shit, the headline is that DeepSeek was able create and train its R1 model in a relatively short amount of time (~55 days), for a relatively small amount of money ($5.8M vs $billions USD), using a relatively inferior chip (Nvidia's H800 vs H100).

This breakthrough has significant implications for investors, particularly venture capitalists, whose investments in AI represented over 60% of total VC funding in Q4 2024. First, for VCs investing in AI startups—particularly those at the model layer—this means smaller check sizes and faster liquidity events may now be possible. That’s something VCs can celebrate.

Because DeepSeek and other Chinese companies have demonstrated the ability to develop powerful models with relatively minimal resources, U.S. startups could replicate this success, especially since R1 is open source. The reduced cost of training models at the model layer will have downstream effects on AI startups at the application layer as well. Companies building apps on less expensive, more powerful LLMs will benefit from lower computational costs, making their tools more accessible to a broader range of customers.

Second, because smaller investments can potentially generate outsized success, VCs can diversify by making more bets on AI startups, reducing the risk of concentrating their capital in just a few ventures. For reassurance, look no further than DeepSeek’s domestic competitor, Moonshot AI, a one-year old startup whose Kimi k series model is already challenging the likes of Anthropic's Claude. AI innovation is happening even faster than previously expected, which means VCs should look to make multiple bets to secure their position.

Finally, VCs may want to broaden their search for the next big thing in AI beyond the U.S. Despite ongoing U.S.-China trade tensions and restrictions on key technologies and materials, DeepSeek was able to achieve something remarkable with limited resources. This “Sputnik” moment could inspire engineers and entrepreneurs in other countries and regions to drive innovation. In an open-source world where knowledge and advancements are shared and built upon, the possibility that the next groundbreaking company could emerge from another part of the world is becoming increasingly plausible.