Amidst the rapid development in AI, Silicon Valley recently marveled at DeepSeek, which is an AI model that’s made in China that matches the performance of the likes of Gemini and OpenAI despite using inferior chips.

Available for download in app stores like Google Play, DeepSeek functions like Gemini, and ChatGPT, as it is an AI assistant at its core. There’s a DeepThink R1 option that gives you a more organized answer to your queries–one that has bullet points for easy reading.

We tried a few queries, and the answers DeepSeek gave us are just as accurate as what you’d get on other AI models. What’s impressive, as Wired points out, is how this AI model from China managed to be accurate while having limited resources as a result of sanctions imposed by the US. It’s also more cost-efficient, with DeepSeek R1 costing around $5.6 million compared to the average of $100 million to $1 billion for building an AI model.

“DeepSeek has embraced open source methods, pooling collective expertise and fostering collaborative innovation. This approach not only mitigates resource constraints but also accelerates the development of cutting-edge technologies, setting DeepSeek apart from more insular competitors,” University of Technology Sydney Associate Professor Marina Zhang, who focuses on Chinese Innovations, explains.

So how was DeepSeek formed? It started back in 2015 as Fire-Flyer, which was a deep-learning branch of High-Flyer–known as one of China’s best-performing hedge funds. Through the years,  High-Flyer has been stockpiling GPUs to build the Fire-Flyer supercomputer for analyzing financial data. Fast forward to 2023, High-Flyer founder Liang Wenfeng assembled a small team for AI research, and that’s how DeepSeek was born.

Wenfeng’s vision for DeepSeek was bold, as it is one of the leading AI firms in China that does not depend on funding from tech giants like Baidu, Alibaba, or ByteDance. While they managed to secure 10,000 NVIDIA H100 chips before the export control ban in October 2022, Wenfeng faced the challenge of how to build an AI model that could rival those from the US. That’s where optimization played a key role.

Add to the fact that DeepSeek uses an open-source model and is willing to share its innovations with the public, this certainly caught the attention of the US, especially with how it implements its export controls against China.